536 debug: N052 sawyer step table + Pink log + arm-lock test

debug/536/n052_sawyer/pinklog_armlock_2026-11-09/README.md

@@ -0,0 +1,33 @@
+# N052 sawyer_umi seed 101 on 536 (L40S): Pink IK log and an arm-lock test
+
+Written by @nova (536 collector), 2026-11-09 20:58 CST. This follows `debug/536/n052_sawyer/` and `step_2026-11-09/`.
+
+## Summary
+- **The IK input is correct.** For all 20 `take_item:close` ticks, the pose given to Pink stays fixed at the `take_item:down` end pose (-0.1121, 1.1000, 0.7341; quat -0.5, 0.5, -0.5, -0.5). The pose actions are NaN, and the NaN fallback fills them with the last valid pose. The base dummy joints do not move.
+- **The IK output points the right way, but it cannot hold the arm.** On every physics step, `sol` is the measured `q` plus one step back toward the target. Joint velocity limit × 5 ms caps that step: about 0.0098 rad (right_j2, 1.957 rad/s) and 0.0227 rad (right_j6). Each step, the measured `q` is first pushed away, and the next solve starts again from the pushed `q`. The target therefore travels with the arm: over 15 ticks the hand moves +0.40 m in y (toward the base) and -0.05 m in z. See `pinklog_per_physics_step_2926_2928.txt`.
+- **The arm-lock test shows that IK is not the source of the force.** In `pink_task_space_actions_lock.py` (`MMB_ARM_LOCK`), the arm joint targets are frozen while the pose input is all-NaN (`armlock_s101_lock.log`: locked for the whole close). The hand is still dragged +0.33 m in y and -0.08 m in z. Result: `item_dropped` during lift (`armlock_s101_report.json`).
+- **What pushes is the red die.** See `armlock_close_lift_per_tick.txt`:
+  - For the first ~4 close ticks, the fingers close symmetrically (0.0208 → 0.0147).
+  - Then the left finger stops and is driven back to its open stop (0.0208). The right finger keeps closing to ~0.
+  - The die is ejected in +y at ~1.4 m/s (y 1.10 → 1.60 in 0.36 s), faster than the hand.
+  - The die presses on the left finger at its hard stop, and that drags the whole hand.
+  - The finger drives are at their 20 N effort limit, so the push looks like a contact-solver depenetration of the die squeezed between the fingers, not a commanded force.
+- **Lead to check on the build machine.** On 536, PhysX logs this for both dice:
+
+  `PhysicsUSD: Parse collision - triangle mesh collision (approximation None/MeshSimplification) cannot be a part of a dynamic body, falling back to convexHull approximation: /World/envs/env_0/dynamic/item_red/item_red_1`
+
+  Does the build machine's sawyer s101 log show the same line? If its dice collide with a different approximation (asset or Isaac/PhysX version), that would explain why the same code grasps there and not here.
+
+## Files
+| file | content |
+|---|---|
+| `pink_task_space_actions_log.py` | Pink action term with `MMB_PINK_LOG`: `IN` = raw/filled pose per `process_actions`, `OUT` = measured q and IK solution per solve |
+| `pink_task_space_actions_lock.py` | the same, plus `MMB_ARM_LOCK`: while the pose input is all-NaN, skip the IK solve and keep the cached joint targets |
+| `pinklog_s101_pink.log.gz`, `pinklog_s101_take_item_ticks.csv` | Pink log run (no lock); the tick CSV uses the build machine's recorder columns |
+| `pinklog_close_per_tick.txt` | per tick 2905–2960: IN pose, eef, base, max\|sol−q\|, per-joint sol change (output of `pinkana.py`; OUT index = 4·tick − 1907) |
+| `pinklog_per_physics_step_2926_2928.txt` | q and sol for every physics step of 3 close ticks (`pinkj.py`) |
+| `armlock_s101_*` | arm-lock run: tick CSV, lock log, report |
+| `armlock_close_lift_per_tick.txt` | eef, die, weigher pads, finger q/tau per tick 2925–2973 (`csvlook.py`) |
+
+## Run
+mobile-ready b3820fd82 node-local copy, Apptainer `xc.sif` (Isaac Sim 5.1), physics on CPU. `MEMORY_ROBOT=sawyer_umi TestMemoryN052 --level easy --seed 101 --headless --set 'env.camera={}' --set collect.record.save_in_video=false --set env.sim.contact_reports=true --spawn-jitter 0.3 --episodes 1 --max-steps 20000`, with `memory_runner_patched.py` and `n052_tick_csv.py` from `debug/536/n052_build_s101/`. The two runs differ only in the Pink action file.

debug/536/n052_sawyer/pinklog_armlock_2026-11-09/armlock_close_lift_per_tick.txt

@@ -0,0 +1,50 @@
+finger cols: ['q_right_gripper_l_finger_joint', 'tau_right_gripper_l_finger_joint', 'q_right_gripper_r_finger_joint', 'tau_right_gripper_r_finger_joint']
+2925 tem:down:0 eef_x=-0.1117 eef_y=1.1009 eef_z=0.7521 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=0.0107 q_r_finger=-0.0208 tau_r_finger=0.0105
+2926 tem:down:0 eef_x=-0.1117 eef_y=1.1009 eef_z=0.7506 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=0.0108 q_r_finger=-0.0208 tau_r_finger=0.0106
+2927 tem:down:0 eef_x=-0.1118 eef_y=1.1008 eef_z=0.7491 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=0.0110 q_r_finger=-0.0208 tau_r_finger=0.0107
+2928 tem:down:0 eef_x=-0.1118 eef_y=1.1007 eef_z=0.7476 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=0.0111 q_r_finger=-0.0208 tau_r_finger=0.0109
+2929 tem:down:0 eef_x=-0.1118 eef_y=1.1006 eef_z=0.7461 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=0.0112 q_r_finger=-0.0208 tau_r_finger=0.0110
+2930 tem:down:0 eef_x=-0.1118 eef_y=1.1006 eef_z=0.7446 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=0.0113 q_r_finger=-0.0208 tau_r_finger=0.0111
+2931 tem:down:0 eef_x=-0.1119 eef_y=1.1005 eef_z=0.7431 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=0.0114 q_r_finger=-0.0208 tau_r_finger=0.0112
+2932 tem:down:0 eef_x=-0.1119 eef_y=1.1004 eef_z=0.7416 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=0.0116 q_r_finger=-0.0208 tau_r_finger=0.0114
+2933 tem:down:0 eef_x=-0.1119 eef_y=1.1004 eef_z=0.7402 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=0.0006 q_r_finger=-0.0208 tau_r_finger=-0.0001
+2934  eef_x=-0.1119 eef_y=1.1003 eef_z=0.7388 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-0.0063 q_r_finger=-0.0208 tau_r_finger=-0.0001
+2935 item:close eef_x=-0.1120 eef_y=1.1003 eef_z=0.7376 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0188 tau_l_finger=-20.0000 q_r_finger=-0.0188 tau_r_finger=20.0000
+2936 item:close eef_x=-0.1121 eef_y=1.1003 eef_z=0.7367 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0168 tau_l_finger=-20.0000 q_r_finger=-0.0168 tau_r_finger=20.0000
+2937 item:close eef_x=-0.1121 eef_y=1.1003 eef_z=0.7360 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.4976 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0148 tau_l_finger=-20.0000 q_r_finger=-0.0148 tau_r_finger=20.0000
+2938 item:close eef_x=-0.1122 eef_y=1.1002 eef_z=0.7356 item_red_x=-0.1200 item_red_y=1.1016 item_red_z=0.4974 weigher_pad0=-0.0025 weigher_pad1=-0.0039 q_l_finger=0.0130 tau_l_finger=-20.0000 q_r_finger=-0.0129 tau_r_finger=20.0000
+2939 item:close eef_x=-0.1133 eef_y=1.1029 eef_z=0.7345 item_red_x=-0.1213 item_red_y=1.1082 item_red_z=0.4963 weigher_pad0=-0.0036 weigher_pad1=-0.0039 q_l_finger=0.0147 tau_l_finger=-20.0000 q_r_finger=-0.0111 tau_r_finger=20.0000
+2940 item:close eef_x=-0.1159 eef_y=1.1069 eef_z=0.7326 item_red_x=-0.1239 item_red_y=1.1168 item_red_z=0.4946 weigher_pad0=-0.0054 weigher_pad1=-0.0039 q_l_finger=0.0165 tau_l_finger=-20.0000 q_r_finger=-0.0093 tau_r_finger=20.0000
+2941 item:close eef_x=-0.1193 eef_y=1.1108 eef_z=0.7308 item_red_x=-0.1273 item_red_y=1.1248 item_red_z=0.4930 weigher_pad0=-0.0071 weigher_pad1=-0.0039 q_l_finger=0.0182 tau_l_finger=-20.0000 q_r_finger=-0.0076 tau_r_finger=20.0000
+2942 item:close eef_x=-0.1232 eef_y=1.1141 eef_z=0.7292 item_red_x=-0.1311 item_red_y=1.1319 item_red_z=0.4915 weigher_pad0=-0.0087 weigher_pad1=-0.0039 q_l_finger=0.0200 tau_l_finger=-20.0000 q_r_finger=-0.0058 tau_r_finger=20.0000
+2943 item:close eef_x=-0.1274 eef_y=1.1184 eef_z=0.7273 item_red_x=-0.1352 item_red_y=1.1402 item_red_z=0.4899 weigher_pad0=-0.0102 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0043 tau_r_finger=20.0000
+2944 item:close eef_x=-0.1342 eef_y=1.1335 eef_z=0.7194 item_red_x=-0.1404 item_red_y=1.1675 item_red_z=0.4833 weigher_pad0=-0.0119 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0035 tau_r_finger=20.0000
+2945 item:close eef_x=-0.1452 eef_y=1.1583 eef_z=0.7049 item_red_x=-0.1498 item_red_y=1.2106 item_red_z=0.4721 weigher_pad0=-0.0091 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0031 tau_r_finger=20.0000
+2946 item:close eef_x=-0.1618 eef_y=1.1901 eef_z=0.6872 item_red_x=-0.1682 item_red_y=1.2649 item_red_z=0.4607 weigher_pad0=-0.0071 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0029 tau_r_finger=20.0000
+2947 item:close eef_x=-0.1785 eef_y=1.2239 eef_z=0.6728 item_red_x=-0.1857 item_red_y=1.3193 item_red_z=0.4542 weigher_pad0=-0.0056 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0029 tau_r_finger=20.0000
+2948 item:close eef_x=-0.1936 eef_y=1.2588 eef_z=0.6633 item_red_x=-0.2003 item_red_y=1.3724 item_red_z=0.4536 weigher_pad0=-0.0045 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0029 tau_r_finger=20.0000
+2949 item:close eef_x=-0.2057 eef_y=1.2941 eef_z=0.6587 item_red_x=-0.2112 item_red_y=1.4235 item_red_z=0.4582 weigher_pad0=-0.0037 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0029 tau_r_finger=20.0000
+2950 item:close eef_x=-0.2146 eef_y=1.3293 eef_z=0.6582 item_red_x=-0.2180 item_red_y=1.4722 item_red_z=0.4672 weigher_pad0=-0.0031 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0029 tau_r_finger=20.0000
+2951 item:close eef_x=-0.2202 eef_y=1.3638 eef_z=0.6608 item_red_x=-0.2207 item_red_y=1.5186 item_red_z=0.4792 weigher_pad0=-0.0027 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0029 tau_r_finger=20.0000
+2952 item:close eef_x=-0.2223 eef_y=1.3976 eef_z=0.6653 item_red_x=-0.2194 item_red_y=1.5628 item_red_z=0.4931 weigher_pad0=-0.0023 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0029 tau_r_finger=20.0000
+2953 item:close eef_x=-0.2212 eef_y=1.4302 eef_z=0.6707 item_red_x=-0.2143 item_red_y=1.6047 item_red_z=0.5081 weigher_pad0=-0.0021 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0028 tau_r_finger=20.0000
+2954  eef_x=-0.2169 eef_y=1.4614 eef_z=0.6765 item_red_x=-0.2057 item_red_y=1.6444 item_red_z=0.5238 weigher_pad0=-0.0019 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0028 tau_r_finger=20.0000
+2955 _item:lift eef_x=-0.2083 eef_y=1.4970 eef_z=0.6837 item_red_x=-0.1840 item_red_y=1.6912 item_red_z=0.5472 weigher_pad0=-0.0018 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0025 tau_r_finger=20.0000
+2956 _item:lift eef_x=-0.1990 eef_y=1.5340 eef_z=0.6907 item_red_x=-0.1592 item_red_y=1.7369 item_red_z=0.5716 weigher_pad0=-0.0017 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0022 tau_r_finger=20.0000
+2957 _item:lift eef_x=-0.1876 eef_y=1.5700 eef_z=0.6984 item_red_x=-0.1309 item_red_y=1.7788 item_red_z=0.5977 weigher_pad0=-0.0017 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0019 tau_r_finger=20.0000
+2958 _item:lift eef_x=-0.1737 eef_y=1.6038 eef_z=0.7071 item_red_x=-0.0999 item_red_y=1.8156 item_red_z=0.6257 weigher_pad0=-0.0016 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0017 tau_r_finger=18.8801
+2959 _item:lift eef_x=-0.1585 eef_y=1.6340 eef_z=0.7171 item_red_x=-0.0682 item_red_y=1.8462 item_red_z=0.6555 weigher_pad0=-0.0016 weigher_pad1=-0.0039 q_l_finger=0.0208 tau_l_finger=-20.0000 q_r_finger=-0.0014 tau_r_finger=18.4831
+2960 _item:lift eef_x=-0.1453 eef_y=1.6567 eef_z=0.7254 item_red_x=-0.0421 item_red_y=1.8662 item_red_z=0.6820 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0206 tau_l_finger=-20.0000 q_r_finger=-0.0003 tau_r_finger=12.6863
+2961 _item:lift eef_x=-0.1418 eef_y=1.6644 eef_z=0.7271 item_red_x=-0.0253 item_red_y=1.8922 item_red_z=0.6751 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0189 tau_l_finger=-20.0000 q_r_finger=0.0000 tau_r_finger=19.9987
+2962 _item:lift eef_x=-0.1421 eef_y=1.6665 eef_z=0.7274 item_red_x=-0.0105 item_red_y=1.9041 item_red_z=0.6716 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0169 tau_l_finger=-20.0000 q_r_finger=-0.0000 tau_r_finger=20.0000
+2963 _item:lift eef_x=-0.1435 eef_y=1.6661 eef_z=0.7272 item_red_x=0.0011 item_red_y=1.9323 item_red_z=0.6497 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0149 tau_l_finger=-20.0000 q_r_finger=-0.0000 tau_r_finger=20.0000
+2964 _item:lift eef_x=-0.1453 eef_y=1.6644 eef_z=0.7267 item_red_x=0.0312 item_red_y=1.9495 item_red_z=0.6207 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0129 tau_l_finger=-20.0000 q_r_finger=0.0000 tau_r_finger=20.0000
+2965 _item:lift eef_x=-0.1472 eef_y=1.6621 eef_z=0.7261 item_red_x=0.0498 item_red_y=1.9595 item_red_z=0.6044 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0109 tau_l_finger=-20.0000 q_r_finger=0.0000 tau_r_finger=20.0000
+2966 _item:lift eef_x=-0.1492 eef_y=1.6594 eef_z=0.7254 item_red_x=0.0595 item_red_y=1.9858 item_red_z=0.5742 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0089 tau_l_finger=-20.0000 q_r_finger=0.0000 tau_r_finger=20.0000
+2967 _item:lift eef_x=-0.1511 eef_y=1.6566 eef_z=0.7247 item_red_x=0.0874 item_red_y=2.0064 item_red_z=0.5317 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0069 tau_l_finger=-20.0000 q_r_finger=0.0000 tau_r_finger=20.0000
+2968 _item:lift eef_x=-0.1530 eef_y=1.6536 eef_z=0.7241 item_red_x=0.1097 item_red_y=2.0156 item_red_z=0.5016 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0049 tau_l_finger=-20.0000 q_r_finger=0.0000 tau_r_finger=19.9946
+2969 _item:lift eef_x=-0.1549 eef_y=1.6505 eef_z=0.7234 item_red_x=0.1186 item_red_y=2.0392 item_red_z=0.4629 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0029 tau_l_finger=-16.7366 q_r_finger=0.0000 tau_r_finger=19.9958
+2970 _item:lift eef_x=-0.1568 eef_y=1.6473 eef_z=0.7228 item_red_x=0.1433 item_red_y=2.0627 item_red_z=0.4081 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0009 tau_l_finger=-12.7259 q_r_finger=0.0000 tau_r_finger=19.9971
+2971 _item:lift eef_x=-0.1587 eef_y=1.6441 eef_z=0.7221 item_red_x=0.1689 item_red_y=2.0723 item_red_z=0.3632 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0000 tau_l_finger=-20.0000 q_r_finger=0.0000 tau_r_finger=20.0000
+2972 _item:lift eef_x=-0.1605 eef_y=1.6409 eef_z=0.7215 item_red_x=0.1793 item_red_y=2.0874 item_red_z=0.3545 weigher_pad0=-0.0015 weigher_pad1=-0.0039 q_l_finger=0.0000 tau_l_finger=-20.0000 q_r_finger=0.0000 tau_r_finger=20.0000
+2973 _item:lift eef_x=-3.5563 eef_y=-2.2329 eef_z=0.5827 item_red_x=-0.1200 item_red_y=1.1000 item_red_z=0.5010 weigher_pad0=-0.0002 weigher_pad1=-0.0002 q_l_finger=0.0000 tau_l_finger=-20.0000 q_r_finger=0.0000 tau_r_finger=20.0000

debug/536/n052_sawyer/pinklog_armlock_2026-11-09/armlock_s101_lock.log

@@ -0,0 +1,4 @@
+LOCK 5484 [3.0018413066864014, -1.0007702112197876, 0.00045784973190166056, 2.0008583068847656, -6.614752783207223e-05, 1.4001474380493164, 3.179031773470342e-05]
+UNLOCK 5608
+LOCK 9828 [-0.22996696829795837, -0.6908884048461914, -0.5939531922340393, 1.1731380224227905, 0.46969982981681824, 1.219252347946167, -0.634748637676239]
+UNLOCK 9908

debug/536/n052_sawyer/pinklog_armlock_2026-11-09/armlock_s101_report.json

@@ -0,0 +1,1044 @@
+{
+  "task_id": "N052",
+  "config": "/mnt/scratch/xc536_atomic/n052saw_lock_s101/ptree/src/magicsim/Collect/Task/MemoryStation/Conf/memory_n052_easy.yaml",
+  "episodes": [
+    {
+      "task_id": "N052",
+      "task_version": "EXT60-MobileAvatar-v3",
+      "agent_id": "a11",
+      "family_id": "measured_attribute_memory",
+      "primary_category": "C6",
+      "level": "easy",
+      "embodiment": "sawyer_umi",
+      "scene_seed": 101,
+      "execution_mode": "expert_collect_chain",
+      "station_layout": {
+        "A": [
+          0.0,
+          -0.55,
+          1.571
+        ],
+        "A_obs": null,
+        "B": [
+          0.0,
+          2.05,
+          -1.571
+        ],
+        "S": [
+          -3.425,
+          -2.098,
+          0.251
+        ],
+        "flip_x": false,
+        "route_m": {
+          "S_to_A_m": 4.09,
+          "A_to_B_m": 4.45,
+          "A_to_B_straight_m": 2.6,
+          "S_to_A_straight_m": 3.76
+        }
+      },
+      "spawn_pose": [
+        -3.4252,
+        -2.0984,
+        0.2512
+      ],
+      "base_path": [
+        [
+          -3.425,
+          -2.098,
+          0.251,
+          0.02
+        ],
+        [
+          -3.322,
+          -2.077,
+          0.446,
+          1.7
+        ],
+        [
+          -3.221,
+          -2.068,
+          0.685,
+          2.0
+        ],
+        [
+          -3.116,
+          -2.054,
+          0.909,
+          2.28
+        ],
+        [
+          -3.017,
+          -2.034,
+          1.076,
+          2.52
+        ],
+        [
+          -2.921,
+          -1.999,
+          1.2,
+          2.76
+        ],
+        [
+          -2.83,
+          -1.954,
+          1.293,
+          3.0
+        ],
+        [
+          -2.742,
+          -1.902,
+          1.359,
+          3.24
+        ],
+        [
+          -2.652,
+          -1.846,
+          1.413,
+          3.5
+        ],
+        [
+          -2.566,
+          -1.794,
+          1.454,
+          3.76
+        ],
+        [
+          -2.476,
+          -1.74,
+          1.486,
+          4.04
+        ],
+        [
+          -2.388,
+          -1.69,
+          1.509,
+          4.32
+        ],
+        [
+          -2.296,
+          -1.639,
+          1.527,
+          4.62
+        ],
+        [
+          -2.204,
+          -1.591,
+          1.54,
+          4.92
+        ],
+        [
+          -2.112,
+          -1.544,
+          1.549,
+          5.22
+        ],
+        [
+          -2.022,
+          -1.499,
+          1.556,
+          5.52
+        ],
+        [
+          -1.931,
+          -1.456,
+          1.56,
+          5.82
+        ],
+        [
+          -1.841,
+          -1.413,
+          1.564,
+          6.12
+        ],
+        [
+          -1.744,
+          -1.369,
+          1.566,
+          6.44
+        ],
+        [
+          -1.648,
+          -1.325,
+          1.567,
+          6.76
+        ],
+        [
+          -1.552,
+          -1.281,
+          1.568,
+          7.08
+        ],
+        [
+          -1.456,
+          -1.237,
+          1.568,
+          7.4
+        ],
+        [
+          -1.361,
+          -1.194,
+          1.569,
+          7.72
+        ],
+        [
+          -1.267,
+          -1.153,
+          1.57,
+          8.04
+        ],
+        [
+          -1.172,
+          -1.112,
+          1.571,
+          8.36
+        ],
+        [
+          -1.078,
+          -1.071,
+          1.571,
+          8.68
+        ],
+        [
+          -0.984,
+          -1.029,
+          1.572,
+          9.0
+        ],
+        [
+          -0.889,
+          -0.988,
+          1.573,
+          9.32
+        ],
+        [
+          -0.795,
+          -0.947,
+          1.573,
+          9.64
+        ],
+        [
+          -0.7,
+          -0.905,
+          1.573,
+          9.96
+        ],
+        [
+          -0.606,
+          -0.864,
+          1.573,
+          10.28
+        ],
+        [
+          -0.511,
+          -0.823,
+          1.573,
+          10.6
+        ],
+        [
+          -0.416,
+          -0.78,
+          1.573,
+          10.92
+        ],
+        [
+          -0.327,
+          -0.734,
+          1.573,
+          11.22
+        ],
+        [
+          -0.238,
+          -0.687,
+          1.572,
+          11.58
+        ],
+        [
+          -0.15,
+          -0.639,
+          1.571,
+          12.1
+        ],
+        [
+          -0.054,
+          -0.601,
+          1.618,
+          21.38
+        ],
+        [
+          0.048,
+          -0.618,
+          1.829,
+          21.72
+        ],
+        [
+          0.153,
+          -0.638,
+          2.069,
+          22.02
+        ],
+        [
+          0.258,
+          -0.654,
+          2.277,
+          22.28
+        ],
+        [
+          0.365,
+          -0.657,
+          2.484,
+          22.54
+        ],
+        [
+          0.467,
+          -0.651,
+          2.708,
+          22.82
+        ],
+        [
+          0.572,
+          -0.642,
+          2.964,
+          23.14
+        ],
+        [
+          0.675,
+          -0.622,
+          -3.047,
+          23.48
+        ],
+        [
+          0.774,
+          -0.594,
+          -2.775,
+          23.82
+        ],
+        [
+          0.864,
+          -0.546,
+          -2.503,
+          24.16
+        ],
+        [
+          0.947,
+          -0.485,
+          -2.231,
+          24.5
+        ],
+        [
+          1.02,
+          -0.41,
+          -2.01,
+          24.84
+        ],
+        [
+          1.082,
+          -0.325,
+          -1.862,
+          25.18
+        ],
+        [
+          1.135,
+          -0.239,
+          -1.769,
+          25.5
+        ],
+        [
+          1.185,
+          -0.15,
+          -1.705,
+          25.82
+        ],
+        [
+          1.233,
+          -0.059,
+          -1.662,
+          26.14
+        ],
+        [
+          1.282,
+          0.034,
+          -1.633,
+          26.46
+        ],
+        [
+          1.332,
+          0.127,
+          -1.613,
+          26.78
+        ],
+        [
+          1.38,
+          0.216,
+          -1.6,
+          27.08
+        ],
+        [
+          1.429,
+          0.31,
+          -1.591,
+          27.4
+        ],
+        [
+          1.474,
+          0.402,
+          -1.583,
+          27.8
+        ],
+        [
+          1.523,
+          0.491,
+          -1.577,
+          28.42
+        ],
+        [
+          1.539,
+          0.592,
+          -1.572,
+          29.32
+        ],
+        [
+          1.506,
+          0.688,
+          -1.571,
+          29.64
+        ],
+        [
+          1.47,
+          0.784,
+          -1.57,
+          29.96
+        ],
+        [
+          1.434,
+          0.88,
+          -1.57,
+          30.28
+        ],
+        [
+          1.395,
+          0.976,
+          -1.571,
+          30.6
+        ],
+        [
+          1.356,
+          1.072,
+          -1.571,
+          30.92
+        ],
+        [
+          1.318,
+          1.168,
+          -1.57,
+          31.24
+        ],
+        [
+          1.279,
+          1.264,
+          -1.57,
+          31.56
+        ],
+        [
+          1.24,
+          1.36,
+          -1.57,
+          31.88
+        ],
+        [
+          1.201,
+          1.455,
+          -1.571,
+          32.2
+        ],
+        [
+          1.154,
+          1.55,
+          -1.57,
+          32.52
+        ],
+        [
+          1.09,
+          1.628,
+          -1.57,
+          32.78
+        ],
+        [
+          1.019,
+          1.7,
+          -1.569,
+          33.02
+        ],
+        [
+          0.947,
+          1.77,
+          -1.569,
+          33.26
+        ],
+        [
+          0.869,
+          1.839,
+          -1.569,
+          33.52
+        ],
+        [
+          0.785,
+          1.905,
+          -1.569,
+          33.8
+        ],
+        [
+          0.701,
+          1.959,
+          -1.569,
+          34.08
+        ],
+        [
+          0.611,
+          2.003,
+          -1.569,
+          34.38
+        ],
+        [
+          0.515,
+          2.038,
+          -1.569,
+          34.7
+        ],
+        [
+          0.413,
+          2.065,
+          -1.569,
+          35.04
+        ],
+        [
+          0.314,
+          2.087,
+          -1.57,
+          35.42
+        ],
+        [
+          0.214,
+          2.107,
+          -1.57,
+          36.0
+        ],
+        [
+          0.115,
+          2.086,
+          -1.571,
+          36.68
+        ]
+      ],
+      "stage_events": [
+        {
+          "event": "arrived_A",
+          "from": "S",
+          "t": 12.14,
+          "path_len_m": 3.613
+        },
+        {
+          "event": "left_A",
+          "t": 23.38
+        },
+        {
+          "event": "arrived_B",
+          "from": "A",
+          "t": 36.08,
+          "path_len_m": 8.177
+        },
+        {
+          "event": "move_target",
+          "label": "take_item:pre",
+          "mode": -1,
+          "pos": [
+            -0.1121,
+            1.1,
+            0.8541
+          ],
+          "quat": [
+            0.5,
+            -0.5,
+            0.5,
+            0.5
+          ],
+          "base": [
+            0.0906,
+            2.0799,
+            0.06,
+            -1.5707
+          ]
+        },
+        {
+          "event": "move_target",
+          "label": "take_item:down:0",
+          "mode": 4,
+          "pos": [
+            -0.1121,
+            1.1,
+            0.7341
+          ],
+          "quat": [
+            0.5,
+            -0.5,
+            0.5,
+            0.5
+          ],
+          "base": [
+            0.0906,
+            2.0799,
+            0.06,
+            -1.5707
+          ]
+        },
+        {
+          "event": "settle_err",
+          "label": "take_item:down",
+          "k": 0,
+          "err_mm": [
+            -0.2,
+            -0.3,
+            -4.7
+          ],
+          "norm_mm": 4.7
+        },
+        {
+          "event": "move_target",
+          "label": "take_item:lift",
+          "mode": 4,
+          "pos": [
+            -0.1121,
+            1.1,
+            0.8541
+          ],
+          "quat": [
+            0.5,
+            -0.5,
+            0.5,
+            0.5
+          ],
+          "base": [
+            0.0905,
+            2.0804,
+            0.06,
+            -1.571
+          ]
+        }
+      ],
+      "transport_state": {
+        "light": "red",
+        "on_delivery": {
+          "red": false,
+          "blue": false
+        },
+        "expected_deflection_m": {
+          "red": 0.0149,
+          "blue": 0.0446
+        }
+      },
+      "manipulation_success": false,
+      "mobile_success": true,
+      "full_success": false,
+      "failure_reason": "item_dropped",
+      "station_transition_order": [
+        "S",
+        "A",
+        "B"
+      ],
+      "distance_to_first_interaction": 3.613,
+      "interstation_path_lengths": {
+        "A": [
+          3.613
+        ],
+        "B": [
+          4.564
+        ]
+      },
+      "final_approach_pose": [
+        0.091,
+        2.081,
+        -1.571
+      ],
+      "cue_ledger": {
+        "read_s": 0.5,
+        "gap_s": 1.0,
+        "cues": {
+          "weight_red": {
+            "reads": 1,
+            "first_read_t": 14.86,
+            "first_read_end_t": 16.12,
+            "gap_open_t": 17.12,
+            "reobserved_t": [],
+            "visible_now": false
+          },
+          "weight_blue": {
+            "reads": 1,
+            "first_read_t": 18.86,
+            "first_read_end_t": 20.12,
+            "gap_open_t": 21.12,
+            "reobserved_t": [],
+            "visible_now": false
+          }
+        },
+        "decisions": [
+          {
+            "label": "pick_lighter",
+            "t": 36.96,
+            "class": "memory",
+            "per_cue": {
+              "weight_red": "memory",
+              "weight_blue": "memory"
+            },
+            "readings_m": {
+              "red": 0.038800700753927234,
+              "blue": 0.057989563196897506
+            },
+            "lighter": "red"
+          }
+        ],
+        "protocol_valid": true
+      },
+      "public": {
+        "demo_running": false,
+        "demo_done": true
+      },
+      "truth": {
+        "light": "red",
+        "heavy": "blue",
+        "red_pad": 0,
+        "order": [
+          0,
+          1
+        ]
+      },
+      "episode_time_s": 59.46,
+      "extra": {
+        "item_pos": {
+          "red": [
+            0.1846,
+            2.0953,
+            0.35
+          ],
+          "blue": [
+            0.12,
+            1.1,
+            0.4959
+          ]
+        },
+        "demo_start_t": 12.14,
+        "weigh_events": [
+          {
+            "pad": 0,
+            "event": "freed",
+            "t": 13.14
+          },
+          {
+            "pad": 0,
+            "event": "locked",
+            "t": 16.14,
+            "deflection_m": 0.0388
+          },
+          {
+            "pad": 1,
+            "event": "freed",
+            "t": 17.14
+          },
+          {
+            "pad": 1,
+            "event": "locked",
+            "t": 20.14,
+            "deflection_m": 0.058
+          }
+        ],
+        "cue_diag": {
+          "weight_red": {
+            "active_ticks": 89,
+            "visible_ticks": 89
+          },
+          "weight_blue": {
+            "active_ticks": 89,
+            "visible_ticks": 89
+          }
+        },
+        "demo_done_t": 21.14,
+        "transport_state": {
+          "light": "red",
+          "on_delivery": {
+            "red": false,
+            "blue": false
+          },
+          "expected_deflection_m": {
+            "red": 0.0149,
+            "blue": 0.0446
+          }
+        },
+        "first_pick": {
+          "color": "red",
+          "correct": true,
+          "t": 59.1
+        }
+      },
+      "env_terminated": false,
+      "env_truncated": true,
+      "control_ticks": 2973,
+      "diagnostics": {
+        "atomic_skill_info": null,
+        "global_planner_info": {},
+        "nav_reports": {
+          "0": {
+            "success": true,
+            "native_path": [
+              [
+                -0.08055705577135086,
+                -0.6008435487747192,
+                0.030429374426603317
+              ],
+              [
+                0.06742338836193085,
+                -0.5833139419555664,
+                0.03042936697602272
+              ],
+              [
+                0.1435028314590454,
+                -0.5743016600608826,
+                0.03042936511337757
+              ],
+              [
+                0.36825087666511536,
+                -0.5476782917976379,
+                0.03042936511337757
+              ],
+              [
+                0.41083288192749023,
+                -0.5426340699195862,
+                0.03042936511337757
+              ],
+              [
+                0.6295801997184753,
+                -0.5167214870452881,
+                0.03042936511337757
+              ],
+              [
+                0.6568048000335693,
+                -0.5134965181350708,
+                0.03042936511337757
+              ],
+              [
+                0.6950979232788086,
+                -0.5089603662490845,
+                0.03042936511337757
+              ],
+              [
+                0.9177659749984741,
+                -0.48258334398269653,
+                0.03042936697602272
+              ],
+              [
+                0.9177660346031189,
+                -0.48258328437805176,
+                0.03042936511337757
+              ],
+              [
+                0.9732123613357544,
+                -0.3931257724761963,
+                0.03042936511337757
+              ],
+              [
+                1.0029863119125366,
+                -0.34508827328681946,
+                0.030429363250732422
+              ],
+              [
+                1.1462059020996094,
+                -0.1140168309211731,
+                0.030429350212216377
+              ],
+              [
+                1.1631228923797607,
+                -0.08672289550304413,
+                0.030429350212216377
+              ],
+              [
+                1.2482998371124268,
+                0.05070227384567261,
+                0.030429350212216377
+              ],
+              [
+                1.2932368516921997,
+                0.12320400029420853,
+                0.030429350212216377
+              ],
+              [
+                1.4125683307647705,
+                0.3157340884208679,
+                0.030429350212216377
+              ],
+              [
+                1.4687683582305908,
+                0.40640759468078613,
+                0.030429348349571228
+              ],
+              [
+                1.5230579376220703,
+                0.49399882555007935,
+                0.030429348349571228
+              ],
+              [
+                1.5939898490905762,
+                0.6073216199874878,
+                0.03042934462428093
+              ],
+              [
+                1.508687973022461,
+                0.7216188311576843,
+                0.03042934462428093
+              ],
+              [
+                1.508687973022461,
+                0.7216188311576843,
+                0.03042934462428093
+              ],
+              [
+                1.4620769023895264,
+                0.8330859541893005,
+                0.03042934648692608
+              ],
+              [
+                1.388969898223877,
+                1.0079160928726196,
+                0.030429333448410034
+              ],
+              [
+                1.3795855045318604,
+                1.0303581953048706,
+                0.030429333448410034
+              ],
+              [
+                1.3748223781585693,
+                1.0417490005493164,
+                0.030429333448410034
+              ],
+              [
+                1.3629900217056274,
+                1.0700452327728271,
+                0.030429331585764885
+              ],
+              [
+                1.2629833221435547,
+                1.3092041015625,
+                0.03042932040989399
+              ],
+              [
+                1.197399616241455,
+                1.4660428762435913,
+                0.03042932040989399
+              ],
+              [
+                1.170562982559204,
+                1.5302207469940186,
+                0.03042931854724884
+              ],
+              [
+                1.1551445722579956,
+                1.5670928955078125,
+                0.03042931668460369
+              ],
+              [
+                1.127187728881836,
+                1.6339495182037354,
+                0.030429314821958542
+              ],
+              [
+                1.0492475032806396,
+                1.8203377723693848,
+                0.030429307371377945
+              ],
+              [
+                1.04924738407135,
+                1.8203377723693848,
+                0.030429305508732796
+              ],
+              [
+                0.9635880589485168,
+                1.8671214580535889,
+                0.030429303646087646
+              ],
+              [
+                0.7433589696884155,
+                1.9874018430709839,
+                0.030429303646087646
+              ],
+              [
+                0.6857983469963074,
+                2.018839120864868,
+                0.030429301783442497
+              ],
+              [
+                0.6285221576690674,
+                2.0501210689544678,
+                0.030429299920797348
+              ],
+              [
+                0.4817010462284088,
+                2.1303091049194336,
+                0.030429288744926453
+              ],
+              [
+                0.34276700019836426,
+                2.1320085525512695,
+                0.030429285019636154
+              ],
+              [
+                0.34276700019836426,
+                2.1320085525512695,
+                0.030429285019636154
+              ],
+              [
+                0.20285966992378235,
+                2.132150411605835,
+                0.030429285019636154
+              ],
+              [
+                0.0631614476442337,
+                2.1322922706604004,
+                0.030429288744926453
+              ],
+              [
+                -7.566995918750763e-09,
+                2.049999952316284,
+                0.0304292943328619
+              ]
+            ],
+            "validated_points": 44,
+            "repaired": false,
+            "clearance_radius_m": 0.62,
+            "window_m": 12.0,
+            "window": "path",
+            "scan_s": 1.38
+          }
+        }
+      }
+    }
+  ],
+  "success": false,
+  "planner_seeds": "default",
+  "dr": {
+    "seed": 101,
+    "spawn_jitter": 0.3,
+    "start": [
+      -3.4252,
+      -2.0984,
+      0.2512
+    ],
+    "nominal_start": [
+      -3.3,
+      -2.3,
+      0.0
+    ]
+  },
+  "output_dir": "/mnt/scratch/xc536_atomic/n052saw_lock_s101/rec/memorystage/000",
+  "steps": 2973,
+  "elapsed_s": 801.013,
+  "episodes_run": 1,
+  "episodes_full_success": 0
+}

debug/536/n052_sawyer/pinklog_armlock_2026-11-09/csvlook.py

@@ -0,0 +1,11 @@
+import sys, csv
+f, t0, t1 = sys.argv[1], int(sys.argv[2]), int(sys.argv[3])
+rows = list(csv.DictReader(open(f)))
+h = rows[0].keys()
+fing = [c for c in h if 'finger' in c and (c.startswith('q_') or c.startswith('tau_'))]
+print('finger cols:', fing)
+cols = ['eef_x', 'eef_y', 'eef_z', 'item_red_x', 'item_red_y', 'item_red_z', 'weigher_pad0_joint', 'weigher_pad1_joint'] + fing
+for r in rows:
+    t = int(r['tick'])
+    if t0 <= t <= t1:
+        print(t, r['phase'][-10:], ' '.join('%s=%.4f' % (c.replace('right_gripper_', '').replace('_joint', ''), float(r[c])) for c in cols if r.get(c, '') != ''))

debug/536/n052_sawyer/pinklog_armlock_2026-11-09/pink_task_space_actions_lock.py

@@ -0,0 +1,1175 @@
+# Copyright (c) 2022-2025, The Isaac Lab Project Developers (https://github.com/isaac-sim/IsaacLab/blob/main/CONTRIBUTORS.md).
+# All rights reserved.
+#
+# SPDX-License-Identifier: BSD-3-Clause
+
+from __future__ import annotations
+
+import os
+
+import torch
+from collections.abc import Sequence
+from typing import TYPE_CHECKING
+from dataclasses import MISSING
+
+from pink.tasks import FrameTask
+
+from gymnasium import spaces
+
+import isaaclab.utils.math as math_utils
+from isaaclab.assets.articulation import Articulation
+from magicsim.Env.Robot.mdp.pink_ik import LocalFrameTask, PinkIKController
+from magicsim.Env.Robot.mdp.action_manager import ActionTerm
+from magicsim.Env.Robot.mdp.differential_ik import DifferentialIKController
+
+if TYPE_CHECKING:
+    from isaaclab.envs import ManagerBasedEnv
+    from isaaclab.envs.utils.io_descriptors import GenericActionIODescriptor
+
+    from . import pink_actions_cfg
+
+
+class PinkInverseKinematicsAction(ActionTerm):
+    r"""Pink Inverse Kinematics action term.
+
+    This action term processes the action tensor and sets these setpoints in the pink IK framework.
+    The action tensor is ordered in the order of the tasks defined in PinkIKControllerCfg.
+    """
+
+    cfg: pink_actions_cfg.PinkInverseKinematicsActionCfg
+    """Configuration for the Pink Inverse Kinematics action term."""
+
+    _asset: Articulation
+    """The articulation asset to which the action term is applied."""
+
+    def __init__(
+        self, cfg: pink_actions_cfg.PinkInverseKinematicsActionCfg, env: ManagerBasedEnv
+    ):
+        """Initialize the Pink Inverse Kinematics action term.
+
+        Args:
+            cfg: The configuration for this action term.
+            env: The environment in which the action term will be applied.
+        """
+        self._step_count = 0
+        super().__init__(cfg, env)
+
+        self._env = env
+        self._sim_dt = env.sim.get_physics_dt()
+
+        # Initialize joint information
+        self._initialize_joint_info()
+
+        # Initialize IK controllers
+        self._initialize_ik_controllers()
+
+        # Initialize action tensors
+        self._raw_actions = torch.zeros(
+            self.num_envs, self.action_dim, device=self.device
+        )
+        # Last passed (valid) action per env; used when input becomes NaN (first time: use current EEF)
+        self._last_passed_action = torch.zeros_like(self._raw_actions)
+        # NaN=HOLD anchors (cfg.nan_hold, 2026-10-02 idle-arm flail fix):
+        # per frame-task group, the pose captured when the group turned NaN —
+        # commanded as a CONSTANT target until a real target returns. NaN =
+        # no anchor armed.
+        self._nan_anchor = torch.full_like(self._raw_actions, float("nan"))
+        # _processed_actions stores joint positions output from IK, shape (num_envs, num_controlled_joints)
+        self._processed_actions = torch.zeros(
+            self.num_envs, len(self._controlled_joint_ids), device=self.device
+        )
+
+        # PhysX Articulation Floating joint indices offset from IsaacLab Articulation joint indices
+        self._physx_floating_joint_indices_offset = 6
+
+        # Initialize env_ids (all envs by default)
+        self._env_ids = torch.arange(self.num_envs, device=self.device)
+        # Wrist override (opt-in per env, see set_wrist_override): while set,
+        # the named frame's task ignores rotation about its own z (5-DoF) and
+        # the named joint is driven straight to the given target instead of
+        # whatever the IK solution says. Lets a skill spin the wrist in joint
+        # space (a screw unwind) with the TCP pinned on the wrist axis by
+        # construction, instead of asking the IK to track a rotating pose.
+        # {env_id: (task_index, joint_column, target_rad)}
+        self._wrist_override: dict[int, tuple] = {}
+
+        # Pre-allocate tensors for runtime use
+        self._initialize_helper_tensors()
+
+        # Initialize action space
+        self._action_space = spaces.Box(
+            low=self.cfg.action_space[0].cpu().numpy(),
+            high=self.cfg.action_space[1].cpu().numpy(),
+            dtype=float,
+        )
+
+    def _initialize_joint_info(self) -> None:
+        """Initialize joint IDs and names based on configuration."""
+        # Resolve pink controlled joints
+        self._isaaclab_controlled_joint_ids, self._isaaclab_controlled_joint_names = (
+            self._asset.find_joints(self.cfg.pink_controlled_joint_names)
+        )
+        self.cfg.controller.joint_names = self._isaaclab_controlled_joint_names
+        self._isaaclab_all_joint_ids = list(range(len(self._asset.data.joint_names)))
+        self.cfg.controller.all_joint_names = self._asset.data.joint_names
+
+        # Resolve hand joints (optional)
+        if self.cfg.hand_joint_names not in (None, MISSING):
+            self._hand_joint_ids, self._hand_joint_names = self._asset.find_joints(
+                self.cfg.hand_joint_names
+            )
+        else:
+            self._hand_joint_ids, self._hand_joint_names = [], []
+
+        # Combine all joint information
+        self._controlled_joint_ids = (
+            self._isaaclab_controlled_joint_ids + self._hand_joint_ids
+        )
+        self._controlled_joint_names = (
+            self._isaaclab_controlled_joint_names + self._hand_joint_names
+        )
+        assert len(self._controlled_joint_ids) == self.cfg.num_joints, (
+            f"Number of controlled joints does not match the number of joints to control: {len(self._controlled_joint_ids)} != {self.cfg.num_joints}"
+        )
+
+    def _initialize_ik_controllers(self) -> None:
+        """Initialize Pink IK controllers for all environments."""
+        assert self._env.num_envs > 0, (
+            "Number of environments specified are less than 1."
+        )
+
+        self._ik_controllers = []
+        robot_cfg = self._asset.cfg
+        for _ in range(self._env.num_envs):
+            controller_cfg = self.cfg.controller.copy()
+            # If urdf_path is not specified in controller config, try to get it from various sources
+            if controller_cfg.urdf_path is None:
+                # 1. Try to get from asset configuration
+                if hasattr(robot_cfg, "urdf_path"):
+                    controller_cfg.urdf_path = robot_cfg.urdf_path
+                # 2. Try to get from spawn configuration if it's a UsdFileCfg (common in IsaacLab)
+                elif hasattr(robot_cfg.spawn, "urdf_path"):
+                    controller_cfg.urdf_path = robot_cfg.spawn.urdf_path
+                # 3. Try to infer from G1 specific configs if available
+                # Note: This is a fallback for G1 robot specific structure
+                elif (
+                    hasattr(robot_cfg, "spawn")
+                    and hasattr(robot_cfg.spawn, "usd_path")
+                    and "g1" in str(robot_cfg.spawn.usd_path).lower()
+                ):
+                    # Attempt to find URDF in expected location relative to asset root or known path
+                    pass
+
+            self._ik_controllers.append(
+                PinkIKController(
+                    cfg=controller_cfg,
+                    robot_cfg=robot_cfg,
+                    device=self.device,
+                    controlled_joint_indices=self._isaaclab_controlled_joint_ids,
+                )
+            )
+
+    def _initialize_helper_tensors(self) -> None:
+        """Pre-allocate tensors and cache values for performance optimization."""
+        # Cache frequently used tensor versions of joint IDs to avoid repeated creation
+        self._controlled_joint_ids_tensor = torch.tensor(
+            self._controlled_joint_ids, device=self.device
+        )
+
+        # Cache base link index to avoid string lookup every time
+        if self.cfg.controller.base_link_name == "root":
+            self._base_link_idx = None
+        else:
+            articulation_data = self._env.scene[
+                self.cfg.controller.articulation_name
+            ].data
+            self._base_link_idx = articulation_data.body_names.index(
+                self.cfg.controller.base_link_name
+            )
+
+        # Pre-allocate working tensors
+        # Count only FrameTask instances in variable_input_tasks (not all tasks)
+        num_frame_tasks = sum(
+            1
+            for task in self._ik_controllers[0].cfg.variable_input_tasks
+            if isinstance(task, FrameTask)
+        )
+        self._num_frame_tasks = num_frame_tasks
+        self._controlled_frame_poses = torch.zeros(
+            num_frame_tasks, self.num_envs, 4, 4, device=self.device
+        )
+
+        # Pre-allocate tensor for base frame computations
+        self._base_link_frame_buffer = torch.zeros(
+            self.num_envs, 4, 4, device=self.device
+        )
+
+        # Cache EEF body indices for NaN action handling
+        # Maps frame task index → body index in articulation data
+        self._eef_body_indices: list[int] = []
+        if self.cfg.target_eef_link_names not in (None, MISSING):
+            for link_name in self.cfg.target_eef_link_names.values():
+                body_idx = self._asset.data.body_names.index(link_name)
+                self._eef_body_indices.append(body_idx)
+
+    # ==================== Properties ====================
+
+    @property
+    def hand_joint_dim(self) -> int:
+        """Dimension for hand joint positions."""
+        return self.cfg.controller.num_hand_joints
+
+    @property
+    def position_dim(self) -> int:
+        """Dimension for position (x, y, z)."""
+        return 3
+
+    @property
+    def orientation_dim(self) -> int:
+        """Dimension for orientation (w, x, y, z)."""
+        return 4
+
+    @property
+    def pose_dim(self) -> int:
+        """Total pose dimension (position + orientation)."""
+        return self.position_dim + self.orientation_dim
+
+    @property
+    def action_dim(self) -> int:
+        """Dimension of the action space (based on number of tasks and pose dimension)."""
+        # Count only FrameTask instances in variable_input_tasks
+        frame_tasks_count = sum(
+            1
+            for task in self._ik_controllers[0].cfg.variable_input_tasks
+            if isinstance(task, FrameTask)
+        )
+        return frame_tasks_count * self.pose_dim + self.hand_joint_dim
+
+    @property
+    def env_ids(self) -> torch.Tensor:
+        """Get the current valid environment IDs (excluding NaN action envs)."""
+        return self._env_ids
+
+    @property
+    def raw_actions(self) -> torch.Tensor:
+        """Get the raw actions tensor."""
+        return self._raw_actions
+
+    @property
+    def processed_actions(self) -> torch.Tensor:
+        """Get the processed actions tensor."""
+        return self._processed_actions
+
+    @property
+    def action_space(self) -> spaces.Box:
+        """Get the action space."""
+        return self._action_space
+
+    @property
+    def IO_descriptor(self) -> GenericActionIODescriptor:
+        """The IO descriptor of the action term.
+
+        This descriptor is used to describe the action term of the pink inverse kinematics action.
+        It adds the following information to the base descriptor:
+        - scale: The scale of the action term.
+        - offset: The offset of the action term.
+        - clip: The clip of the action term.
+        - pink_controller_joint_names: The names of the pink controller joints.
+        - hand_joint_names: The names of the hand joints.
+        - controller_cfg: The configuration of the pink controller.
+
+        Returns:
+            The IO descriptor of the action term.
+        """
+        super().IO_descriptor
+        self._IO_descriptor.shape = (self.action_dim,)
+        self._IO_descriptor.dtype = str(self.raw_actions.dtype)
+        self._IO_descriptor.action_type = "PinkInverseKinematicsAction"
+        self._IO_descriptor.pink_controller_joint_names = (
+            self._isaaclab_controlled_joint_names
+        )
+        self._IO_descriptor.hand_joint_names = self._hand_joint_names
+        self._IO_descriptor.extras["controller_cfg"] = self.cfg.controller.__dict__
+        return self._IO_descriptor
+
+    # """
+    # Operations.
+    # """
+
+    def _handle_nan_actions(
+        self, actions: torch.Tensor, selected_env_ids: torch.Tensor
+    ) -> tuple[torch.Tensor | None, torch.Tensor]:
+        """Handle NaN actions: use last passed action if available, else current EEF pose.
+
+        Actions are structured as groups of pose_dim (7) per frame task, optionally
+        followed by hand joint values. The logic is:
+
+        1. Include all envs by default (do not skip all-NaN envs).
+        2. For each group of 7 that has any NaN: use last passed action if valid;
+           else (no valid previous action) behavior depends on
+           ``cfg.fallback_to_current``:
+             - True (default): fall back to current EEF pose.
+             - False: skip this env (drop it from the returned batch).
+        3. For hand joints: same logic — last passed if valid, else current joint
+           positions (or skip when ``fallback_to_current`` is False).
+
+        Args:
+            actions: Input action tensor of shape ``(len(selected_env_ids), action_dim)``.
+                Rows are **already** the slice for those envs (same order as
+                ``selected_env_ids``), as produced by ``ActionManager.process_action``.
+                Do not index again with global env ids as row indices.
+            selected_env_ids: Selected environment IDs (global indices).
+
+        Returns:
+            Tuple of (processed_actions, valid_env_ids).
+            processed_actions is None if no valid envs exist.
+        """
+        env_ids = selected_env_ids
+        if len(env_ids) == 0:
+            return None, env_ids
+
+        if actions.shape[0] != len(env_ids):
+            raise ValueError(
+                f"Pink IK actions batch ({actions.shape[0]}) must match len(env_ids) ({len(env_ids)}); "
+                "actions should be the env slice in env_ids order, not the full buffer indexed by id."
+            )
+
+        valid_actions = actions.clone()
+        fallback_to_current = self.cfg.fallback_to_current
+        skip_mask = torch.zeros(
+            len(env_ids), dtype=torch.bool, device=valid_actions.device
+        )
+
+        # For each group of pose_dim (7), fill NaN with last passed action or current EEF pose
+        for task_idx in range(self._num_frame_tasks):
+            start = task_idx * self.pose_dim
+            end = start + self.pose_dim
+
+            group_nan_mask = torch.isnan(valid_actions[:, start:end]).any(dim=1)
+            if getattr(self.cfg, "nan_hold", False):
+                # NaN=HOLD (2026-10-02 idle-arm flail): a NaN group commands
+                # the pose ANCHORED when it turned NaN — a constant target
+                # pink holds with real stiffness. The legacy fallback
+                # ("last passed, else current pose") is a zero-stiffness
+                # follow target: the idle arm drifts (null-space posture
+                # pull + gravity-comp residual) and the drift is chased.
+                real_mask = ~group_nan_mask
+                if real_mask.any():
+                    # live target -> disarm the anchor
+                    self._nan_anchor[env_ids[real_mask], start:end] = float("nan")
+                if group_nan_mask.any():
+                    anchors = self._nan_anchor[env_ids, start:end]
+                    need_cap = group_nan_mask & torch.isnan(anchors).any(dim=1)
+                    if need_cap.any():
+                        cap_ids = env_ids[need_cap]
+                        self._nan_anchor[cap_ids, start:end] = (
+                            self._get_current_eef_pose(cap_ids, task_idx)
+                        )
+                    valid_actions[group_nan_mask, start:end] = self._nan_anchor[
+                        env_ids[group_nan_mask], start:end
+                    ]
+                continue
+            if group_nan_mask.any():
+                nan_env_ids = env_ids[group_nan_mask]
+                last_poses = self._last_passed_action[nan_env_ids, start:end]
+                has_valid_last = (~torch.isnan(last_poses)).all(dim=1) & (
+                    last_poses[:, 3:7].abs().sum(dim=1) > 0.01
+                )
+                if fallback_to_current:
+                    current_poses = self._get_current_eef_pose(nan_env_ids, task_idx)
+                    fill_values = torch.where(
+                        has_valid_last.unsqueeze(1).expand_as(current_poses),
+                        last_poses,
+                        current_poses,
+                    )
+                    valid_actions[group_nan_mask, start:end] = fill_values
+                else:
+                    # Use last_poses only where valid; mark the rest for skipping.
+                    valid_actions[group_nan_mask, start:end] = last_poses
+                    group_local_idx = group_nan_mask.nonzero(as_tuple=True)[0]
+                    skip_mask[group_local_idx[~has_valid_last]] = True
+
+        # Handle NaN in hand joints: last passed if valid, else current joint positions
+        if self.hand_joint_dim > 0:
+            hand_start = self._num_frame_tasks * self.pose_dim
+            hand_nan_mask = torch.isnan(valid_actions[:, hand_start:]).any(dim=1)
+            if hand_nan_mask.any():
+                nan_env_ids = env_ids[hand_nan_mask]
+                last_hand = self._last_passed_action[nan_env_ids, hand_start:]
+                has_valid_last_hand = ~torch.isnan(last_hand).any(dim=1)
+                if fallback_to_current:
+                    current_hand_pos = self._asset.data.joint_pos[nan_env_ids][
+                        :, self._hand_joint_ids
+                    ]
+                    fill_hand = torch.where(
+                        has_valid_last_hand.unsqueeze(1).expand_as(current_hand_pos),
+                        last_hand,
+                        current_hand_pos,
+                    )
+                    valid_actions[hand_nan_mask, hand_start:] = fill_hand
+                else:
+                    valid_actions[hand_nan_mask, hand_start:] = last_hand
+                    hand_local_idx = hand_nan_mask.nonzero(as_tuple=True)[0]
+                    skip_mask[hand_local_idx[~has_valid_last_hand]] = True
+
+        if skip_mask.any():
+            keep_mask = ~skip_mask
+            valid_actions = valid_actions[keep_mask]
+            env_ids = env_ids[keep_mask]
+            if len(env_ids) == 0:
+                return None, env_ids
+
+        return valid_actions, env_ids
+
+    def _get_current_eef_pose(
+        self, env_ids: torch.Tensor, task_idx: int
+    ) -> torch.Tensor:
+        """Get current EEF pose for specified envs and frame task index.
+
+        Returns the pose as [pos_x, pos_y, pos_z, quat_w, quat_x, quat_y, quat_z]
+        in the same coordinate frame as the input actions (base-relative when
+        ``relative_to_base`` is True, otherwise env-origin-relative world frame).
+
+        Args:
+            env_ids: Environment IDs for which to get the current EEF pose.
+            task_idx: Frame task index (0-based, counting only FrameTask instances).
+
+        Returns:
+            Current EEF pose tensor of shape ``(len(env_ids), 7)``.
+        """
+        body_idx = self._eef_body_indices[task_idx]
+
+        # Get world-frame pose from articulation data: [pos(3), quat(4)]
+        eef_state_w = self._asset.data.body_link_state_w[env_ids, body_idx, :7]
+        eef_pos_w = eef_state_w[:, :3]
+        eef_quat_w = eef_state_w[:, 3:7]
+
+        # Convert position to env-origin-relative frame
+        eef_pos = eef_pos_w - self._env.scene.env_origins[env_ids]
+
+        if self.cfg.relative_to_base:
+            # Compute base link pose in env-origin frame
+            if self.cfg.controller.base_link_name == "root":
+                base_pos = (
+                    self._asset.data.root_pos_w[env_ids]
+                    - self._env.scene.env_origins[env_ids]
+                )
+                base_quat = self._asset.data.root_quat_w[env_ids]
+            else:
+                articulation_data = self._env.scene[
+                    self.cfg.controller.articulation_name
+                ].data
+                base_state = articulation_data.body_link_state_w[
+                    env_ids, self._base_link_idx, :7
+                ]
+                base_pos = base_state[:, :3] - self._env.scene.env_origins[env_ids]
+                base_quat = base_state[:, 3:7]
+
+            # Express EEF pose relative to the base link frame
+            eef_pos_base, eef_quat_base = math_utils.subtract_frame_transforms(
+                base_pos, base_quat, eef_pos, eef_quat_w
+            )
+            return torch.cat([eef_pos_base, eef_quat_base], dim=1)
+        else:
+            return torch.cat([eef_pos, eef_quat_w], dim=1)
+
+    def process_actions(self, actions: torch.Tensor, env_ids: torch.Tensor) -> None:
+        """Process the input actions and set targets for each task.
+
+        Args:
+            actions: The input actions tensor.
+            env_ids: The environment IDs to process.
+        """
+        if self._step_count % self.cfg.decimation != 0:
+            return
+
+        # Handle env_ids
+        if env_ids is None:
+            selected_env_ids = torch.arange(self.num_envs, device=self.device)
+        elif not isinstance(env_ids, torch.Tensor):
+            selected_env_ids = torch.tensor(
+                env_ids, device=self.device, dtype=torch.int32
+            )
+        else:
+            selected_env_ids = env_ids
+
+        # Handle NaN actions: skip all-NaN envs, fill partial NaN groups with current EEF pose
+        valid_actions, self._env_ids = self._handle_nan_actions(
+            actions, selected_env_ids
+        )
+        # MMB_ARM_LOCK experiment (536, N052 sawyer): while the pose input is
+        # all-NaN (a hold), keep the arm joint targets frozen at their value
+        # when the hold began instead of re-solving IK from measured q.
+        if os.environ.get("MMB_ARM_LOCK"):
+            _hold = bool(actions is not None and len(actions) and torch.isnan(actions[0, : self.pose_dim]).all())
+            if _hold and not getattr(self, "_arm_locked", False):
+                with open(os.environ["MMB_ARM_LOCK"], "a") as _f:
+                    _f.write("LOCK %d %s\n" % (self._step_count, self._processed_actions[0].tolist()))
+            elif not _hold and getattr(self, "_arm_locked", False):
+                with open(os.environ["MMB_ARM_LOCK"], "a") as _f:
+                    _f.write("UNLOCK %d\n" % self._step_count)
+            self._arm_locked = _hold
+        if os.environ.get("MMB_PINK_LOG"):
+            with open(os.environ["MMB_PINK_LOG"], "a") as _f:
+                _raw = actions[0].tolist() if actions is not None and len(actions) else None
+                _f.write("IN %d raw=%s valid=%s envs=%s\n" % (self._step_count, _raw, None if valid_actions is None else valid_actions[0].tolist(), self._env_ids.tolist()))
+
+        # If no valid envs, return early
+        if valid_actions is None or len(self._env_ids) == 0:
+            return
+
+        # Store raw actions and record last passed action for NaN fallback
+        self._raw_actions[self._env_ids] = valid_actions
+        self._last_passed_action[self._env_ids] = valid_actions.clone()
+
+        # Extract hand joint positions directly (no cloning needed)
+        # Handle case when hand_joint_dim = 0: -0 is same as 0, so slice would be whole tensor
+        if self.hand_joint_dim > 0:
+            self._target_hand_joint_positions = valid_actions[:, -self.hand_joint_dim :]
+        else:
+            self._target_hand_joint_positions = torch.zeros(
+                (valid_actions.shape[0], 0), device=self.device
+            )
+
+        # Get base link frame transformation for valid envs only
+        all_base_link_frames = self._get_base_link_frame_transform()
+        self.base_link_frame_in_world_rf = all_base_link_frames[self._env_ids]
+
+        # Process controlled frame poses for valid envs
+        controlled_frame_poses = self._extract_controlled_frame_poses(valid_actions)
+
+        # DEBUG: Print raw action and controlled frame poses
+        # print(f"\n=== Pink IK process_actions DEBUG ===")
+        # print(f"Raw action (first env): pos={valid_actions[0, :3].cpu().numpy()}, quat={valid_actions[0, 3:7].cpu().numpy()}")
+        # print(f"relative_to_base: {self.cfg.relative_to_base}")
+        # print(f"Controlled frame poses (world frame):\n{controlled_frame_poses[0, 0].cpu().numpy()}")
+
+        # NOTE(magics): We assume actions are already in the Base Link Frame (Pelvis) for LocalFrameTask.
+        # This bypasses the World-to-Base transformation to allow direct local control.
+        # If using standard FrameTask (World Frame), this would imply inputs are also in Local Frame
+        # but interpreted as World Frame, which might be incorrect for mixed usage.
+        # But for G1 LocalFrameTask usage, this is the correct behavior for Body-relative control.
+        if self.cfg.relative_to_base:
+            positions, rotation_matrices = math_utils.unmake_pose(
+                controlled_frame_poses
+            )
+            transformed_poses = (positions, rotation_matrices)
+        else:
+            transformed_poses = self._transform_poses_to_base_link_frame(
+                controlled_frame_poses
+            )
+            # DEBUG: Print after transformation
+            # print(f"After transform_to_base_link_frame:")
+            # print(f"  positions: {transformed_poses[0].cpu().numpy()}")
+            # print(f"  rotation_matrices: {transformed_poses[1][0, 0].cpu().numpy()}")
+
+        # Set targets for valid envs only
+        self._set_task_targets(transformed_poses)
+
+    def _get_base_link_frame_transform(self) -> torch.Tensor:
+        """Get the base link frame transformation matrix.
+
+        Returns:
+            Base link frame transformation matrix.
+        """
+        if self.cfg.controller.base_link_name == "root":
+            base_pos_w = self._asset.data.root_pos_w
+            base_quat_w = self._asset.data.root_quat_w
+            torch.sub(
+                base_pos_w,
+                self._env.scene.env_origins,
+                out=self._base_link_frame_buffer[:, :3, 3],
+            )
+            return math_utils.make_pose(
+                self._base_link_frame_buffer[:, :3, 3],
+                math_utils.matrix_from_quat(base_quat_w),
+            )
+
+        # Get base link frame pose in world origin using cached index
+        articulation_data = self._env.scene[self.cfg.controller.articulation_name].data
+        base_link_frame_in_world_origin = articulation_data.body_link_state_w[
+            :, self._base_link_idx, :7
+        ]
+
+        # Transform to environment origin frame (reuse buffer to avoid allocation)
+        torch.sub(
+            base_link_frame_in_world_origin[:, :3],
+            self._env.scene.env_origins,
+            out=self._base_link_frame_buffer[:, :3, 3],
+        )
+
+        # Copy orientation (avoid clone)
+        base_link_frame_quat = base_link_frame_in_world_origin[:, 3:7]
+
+        # Create transformation matrix
+        return math_utils.make_pose(
+            self._base_link_frame_buffer[:, :3, 3],
+            math_utils.matrix_from_quat(base_link_frame_quat),
+        )
+
+    def _extract_controlled_frame_poses(self, actions: torch.Tensor) -> torch.Tensor:
+        """Extract controlled frame poses from action tensor for valid envs.
+
+        Args:
+            actions: The action tensor for valid envs, shape (num_valid_envs, action_dim).
+
+        Returns:
+            Controlled frame poses tensor, shape (num_frame_tasks, num_valid_envs, 4, 4).
+        """
+        num_valid_envs = actions.shape[0]
+        # Create tensor for valid envs (cannot use pre-allocated when size varies)
+        controlled_frame_poses = torch.zeros(
+            self._num_frame_tasks, num_valid_envs, 4, 4, device=self.device
+        )
+
+        for task_index in range(self._num_frame_tasks):
+            # Extract position and orientation for this task
+            pos_start = task_index * self.pose_dim
+            pos_end = pos_start + self.position_dim
+            quat_start = pos_end
+            quat_end = (task_index + 1) * self.pose_dim
+
+            position = actions[:, pos_start:pos_end]
+            quaternion = actions[:, quat_start:quat_end]
+
+            # Create pose matrix
+            controlled_frame_poses[task_index] = math_utils.make_pose(
+                position, math_utils.matrix_from_quat(quaternion)
+            )
+
+        return controlled_frame_poses
+
+    def _transform_poses_to_base_link_frame(
+        self, poses: torch.Tensor
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        """Transform poses from world frame to base link frame.
+
+        Args:
+            poses: Poses in world frame.
+
+        Returns:
+            Tuple of (positions, rotation_matrices) in base link frame.
+        """
+        # Transform poses to base link frame
+        base_link_inv = math_utils.pose_inv(self.base_link_frame_in_world_rf)
+        transformed_poses = math_utils.pose_in_A_to_pose_in_B(poses, base_link_inv)
+
+        # Extract position and rotation
+        positions, rotation_matrices = math_utils.unmake_pose(transformed_poses)
+
+        return positions, rotation_matrices
+
+    def _set_task_targets(
+        self, transformed_poses: tuple[torch.Tensor, torch.Tensor]
+    ) -> None:
+        """Set targets for all tasks across valid environments only.
+
+        Args:
+            transformed_poses: Tuple of (positions, rotation_matrices) in base link frame,
+                               with shape (num_frame_tasks, num_valid_envs, ...).
+        """
+        positions, rotation_matrices = transformed_poses
+
+        # DEBUG: Print base link frame and transformed target positions
+        # print(f"\n=== Pink IK _set_task_targets DEBUG ===")
+        # print(f"base_link_frame_in_world_rf (env 0):\n{self.base_link_frame_in_world_rf[0].cpu().numpy()}")
+        # print(f"Target positions (in base link frame): {positions.cpu().numpy()}")
+
+        # Iterate only over valid envs
+        for valid_idx, env_index in enumerate(self._env_ids.tolist()):
+            ik_controller = self._ik_controllers[env_index]
+            for frame_task_index, task in enumerate(
+                ik_controller.cfg.variable_input_tasks
+            ):
+                if isinstance(task, LocalFrameTask):
+                    target = task.transform_target_to_base
+                elif isinstance(task, FrameTask):
+                    target = task.transform_target_to_world
+                else:
+                    continue
+
+                # Use valid_idx to index into transformed_poses
+                target.translation = (
+                    positions[frame_task_index, valid_idx, :].cpu().numpy()
+                )
+                target.rotation = (
+                    rotation_matrices[frame_task_index, valid_idx, :].cpu().numpy()
+                )
+
+                task.set_target(target)
+
+    def set_wrist_override(
+        self, env_id: int, frame: str, joint_name: str, target: float | None
+    ) -> bool:
+        """Drive ``joint_name`` of ``env_id`` straight to ``target`` (rad)
+        while the task for ``frame`` ignores rotation about its own z.
+        ``target=None`` clears the override (6-DoF task restored). Returns
+        False when the frame or joint is unknown to this term."""
+        env_id = int(env_id)
+        if target is None:
+            self._wrist_override.pop(env_id, None)
+            return True
+        tasks = self._ik_controllers[env_id].cfg.variable_input_tasks
+        task_index = next(
+            (i for i, t in enumerate(tasks) if getattr(t, "frame", None) == frame),
+            None,
+        )
+        if (
+            task_index is None
+            or joint_name not in self._isaaclab_controlled_joint_names
+        ):
+            return False
+        col = self._isaaclab_controlled_joint_names.index(joint_name)
+        self._wrist_override[env_id] = (task_index, col, float(target))
+        return True
+
+    def _wrist_target_clamped(self, env_id: int, col: int, target: float) -> float:
+        joint_id = self._isaaclab_controlled_joint_ids[col]
+        lim = self._asset.data.joint_pos_limits[env_id, joint_id]
+        return float(min(max(target, float(lim[0])), float(lim[1])))
+
+    # ==================== Action Application ====================
+
+    def apply_actions(self) -> None:
+        """Apply the computed joint positions based on the inverse kinematics solution.
+
+        IK solve is only performed every ``cfg.decimation`` steps.  Cached
+        joint-position targets are applied every step so the robot never
+        stalls.
+        """
+        self._step_count += 1
+
+        if len(self._env_ids) == 0:
+            return
+
+        # Recompute IK only on decimation steps (matches process_actions gate)
+        if (self._step_count - 1) % self.cfg.decimation == 0 and getattr(self, "_arm_locked", False):
+            if os.environ.get("MMB_PINK_LOG"):
+                with open(os.environ["MMB_PINK_LOG"], "a") as _f:
+                    _f.write("OUT %d q=%s sol=LOCKED\n" % (self._step_count, [round(float(v), 5) for v in self._asset.data.joint_pos[0, self._controlled_joint_ids].tolist()]))
+        elif (self._step_count - 1) % self.cfg.decimation == 0:
+            ik_joint_positions = self._compute_ik_solutions()
+            if os.environ.get("MMB_PINK_LOG"):
+                with open(os.environ["MMB_PINK_LOG"], "a") as _f:
+                    _f.write("OUT %d q=%s sol=%s\n" % (self._step_count, [round(float(v), 5) for v in self._asset.data.joint_pos[0, self._controlled_joint_ids].tolist()], [round(float(v), 5) for v in ik_joint_positions[0].tolist()]))
+
+            if self.hand_joint_dim > 0:
+                all_joint_positions = torch.cat(
+                    (ik_joint_positions, self._target_hand_joint_positions), dim=1
+                )
+            else:
+                all_joint_positions = ik_joint_positions
+
+            self._processed_actions[self._env_ids] = all_joint_positions
+
+        for env_id, (_, col, target) in self._wrist_override.items():
+            self._processed_actions[env_id, col] = self._wrist_target_clamped(
+                env_id, col, target
+            )
+
+        # Always apply cached targets + gravity compensation
+        if self.cfg.enable_gravity_compensation:
+            self._apply_gravity_compensation()
+
+        self._asset.set_joint_position_target(
+            self._processed_actions[self._env_ids],
+            self._controlled_joint_ids,
+            self._env_ids,
+        )
+
+    def _apply_gravity_compensation(self) -> None:
+        """Apply gravity compensation to arm joints for valid envs only."""
+        if not self._asset.cfg.spawn.rigid_props.disable_gravity:
+            # Get gravity compensation forces using cached tensor for valid envs
+            all_forces = self._asset.root_physx_view.get_gravity_compensation_forces()
+            if self._asset.is_fixed_base:
+                if getattr(self.cfg, "fixed_base_gravity_compensation", False):
+                    # PhysX returns (count, max_dofs) for fixed-base articulations
+                    # (no root block), so the joint ids index directly — the same
+                    # indexing IsaacLab's own pink / OSC actions use.
+                    gravity = all_forces[self._env_ids][
+                        :, self._controlled_joint_ids_tensor
+                    ]
+                else:  # legacy: no compensation on fixed-base robots
+                    gravity = torch.zeros_like(
+                        all_forces[self._env_ids][:, self._controlled_joint_ids_tensor]
+                    )
+            else:
+                # If floating base, then need to skip the first 6 joints (base)
+                gravity = all_forces[self._env_ids][
+                    :,
+                    self._controlled_joint_ids_tensor
+                    + self._physx_floating_joint_indices_offset,
+                ]
+
+            # Apply gravity compensation to arm joints for valid envs
+            self._asset.set_joint_effort_target(
+                gravity, self._controlled_joint_ids, self._env_ids
+            )
+
+    def _solve_ik(self, ik_controller, current_joint_pos) -> torch.Tensor:
+        """One IK solve of one env: the iterative solver when the controller asks for more than one QP, else the historical call."""
+        if getattr(ik_controller.cfg, "ik_iterations", 1) > 1:
+            return ik_controller.compute(
+                current_joint_pos,
+                self._sim_dt,
+                control_dt=self._sim_dt * self.cfg.decimation,
+            )
+        # Keep the historical single-QP call and its 5 ms dt exactly.
+        return ik_controller.compute(current_joint_pos, self._sim_dt)
+
+    def _compute_ik_solutions(self) -> torch.Tensor:
+        """Compute IK solutions for valid environments only.
+
+        Returns:
+            IK joint positions tensor for valid environments.
+        """
+        ik_solutions = []
+
+        for env_index in self._env_ids.tolist():
+            ik_controller = self._ik_controllers[env_index]
+            # Get current joint positions for this environment
+            current_joint_pos = self._asset.data.joint_pos.cpu().numpy()[env_index]
+
+            ov = self._wrist_override.get(int(env_index))
+            if ov is None:
+                joint_pos_des = self._solve_ik(ik_controller, current_joint_pos)
+            else:
+                # The task objects are shared by every env's controller, so
+                # the cost is swapped only around this env's solve.
+                task_index, col, target = ov
+                task = ik_controller.cfg.variable_input_tasks[task_index]
+                saved = task.orientation_cost.copy()
+                free = saved.copy()
+                free[2] = 0.0
+                task.set_orientation_cost(free)
+                try:
+                    joint_pos_des = self._solve_ik(ik_controller, current_joint_pos)
+                finally:
+                    task.set_orientation_cost(saved)
+                joint_pos_des = joint_pos_des.clone()
+                joint_pos_des[col] = self._wrist_target_clamped(
+                    int(env_index), col, target
+                )
+            ik_solutions.append(joint_pos_des)
+
+        return torch.stack(ik_solutions)
+
+    # ==================== Reset ====================
+
+    def reset(self, env_ids: Sequence[int] | None = None) -> None:
+        """Reset the action term for specified environments.
+
+        Args:
+            env_ids: A list of environment IDs to reset. If None, all environments are reset.
+        """
+        self._raw_actions[env_ids] = torch.zeros(self.action_dim, device=self.device)
+        self._last_passed_action[env_ids] = torch.zeros(
+            self.action_dim, device=self.device
+        )
+        # Episode boundary: drop the NaN=HOLD anchors — the next NaN tick
+        # re-captures from the freshly reset pose (the yaml park).
+        self._nan_anchor[env_ids] = float("nan")
+        # Also clear ``_processed_actions`` (the cached IK solution applied
+        # in ``apply_actions`` every step). Without this, after a reset the
+        # next NaN-action tick has ``valid_actions=None`` and
+        # ``process_actions`` returns early — but ``apply_actions`` still
+        # ships the *previous* episode's IK joint targets to PhysX,
+        # snapping the arms back to whatever pose ended the last attempt.
+        # Replace with the robot's current joint positions ("hold here")
+        # so the freshly-reset default joint pose is what gets commanded.
+        if env_ids is None:
+            env_ids_tensor = torch.arange(self.num_envs, device=self.device)
+        elif isinstance(env_ids, torch.Tensor):
+            env_ids_tensor = env_ids.to(self.device, dtype=torch.long)
+        else:
+            env_ids_tensor = torch.tensor(
+                list(env_ids), device=self.device, dtype=torch.long
+            )
+        if env_ids_tensor.numel() > 0:
+            cur_joint_pos = self._asset.data.joint_pos[env_ids_tensor][
+                :, self._controlled_joint_ids
+            ]
+            self._processed_actions[env_ids_tensor] = cur_joint_pos
+            for e in env_ids_tensor.tolist():
+                self._wrist_override.pop(int(e), None)
+
+
+class PinkDualDifferentialInverseKinematicsAction(PinkInverseKinematicsAction):
+    """Hybrid Pink IK + per-arm differential IK action.
+
+    On decimation fires this runs the full Pink IK solve (via the parent
+    implementation). Between fires it falls back to per-arm Jacobian tracking
+    toward the *same* target, matching the computation in
+    :class:`DualDifferentialInverseKinematicsAction`. Hand-joint targets are
+    refreshed only on Pink fires; PhysX retains them between steps.
+
+    The action layout is identical to the parent: ``[right_pose(7), left_pose(7)]``
+    for dual-arm robots, because the Pink ``variable_input_tasks`` must be
+    ordered right-first, left-second.
+    """
+
+    cfg: "pink_actions_cfg.PinkDualDifferentialInverseKinematicsActionCfg"
+
+    def __init__(
+        self,
+        cfg: "pink_actions_cfg.PinkDualDifferentialInverseKinematicsActionCfg",
+        env,
+    ):
+        super().__init__(cfg, env)
+        self._setup_dual_diff()
+
+    def _setup_dual_diff(self) -> None:
+        cfg = self.cfg
+
+        self._right_joint_ids, _ = self._asset.find_joints(cfg.right_joint_names)
+        self._left_joint_ids, _ = self._asset.find_joints(cfg.left_joint_names)
+
+        r_ids, r_names = self._asset.find_bodies(cfg.right_body_name)
+        l_ids, l_names = self._asset.find_bodies(cfg.left_body_name)
+        if len(r_ids) != 1 or len(l_ids) != 1:
+            raise ValueError(
+                f"Expected one body match each: right={r_names}, left={l_names}"
+            )
+        self._right_body_idx = r_ids[0]
+        self._left_body_idx = l_ids[0]
+
+        self._right_ref_idx = None
+        self._left_ref_idx = None
+        if cfg.right_command_reference_body_name is not None:
+            ids, _ = self._asset.find_bodies(cfg.right_command_reference_body_name)
+            if len(ids) != 1:
+                raise ValueError(
+                    f"right_command_reference_body_name resolved to {len(ids)} bodies."
+                )
+            self._right_ref_idx = ids[0]
+        if cfg.left_command_reference_body_name is not None:
+            ids, _ = self._asset.find_bodies(cfg.left_command_reference_body_name)
+            if len(ids) != 1:
+                raise ValueError(
+                    f"left_command_reference_body_name resolved to {len(ids)} bodies."
+                )
+            self._left_ref_idx = ids[0]
+
+        if self._asset.is_fixed_base:
+            self._right_jacobi_body_idx = self._right_body_idx - 1
+            self._left_jacobi_body_idx = self._left_body_idx - 1
+            self._right_jacobi_joint_ids = self._right_joint_ids
+            self._left_jacobi_joint_ids = self._left_joint_ids
+        else:
+            self._right_jacobi_body_idx = self._right_body_idx
+            self._left_jacobi_body_idx = self._left_body_idx
+            self._right_jacobi_joint_ids = [i + 6 for i in self._right_joint_ids]
+            self._left_jacobi_joint_ids = [i + 6 for i in self._left_joint_ids]
+
+        self._right_offset_pos, self._right_offset_rot = self._build_diff_offset(
+            cfg.right_body_offset
+        )
+        self._left_offset_pos, self._left_offset_rot = self._build_diff_offset(
+            cfg.left_body_offset
+        )
+
+        self._right_diff_ik = DifferentialIKController(
+            cfg=cfg.diff_ik_controller, num_envs=self.num_envs, device=self.device
+        )
+        self._left_diff_ik = DifferentialIKController(
+            cfg=cfg.diff_ik_controller, num_envs=self.num_envs, device=self.device
+        )
+
+        if self._num_frame_tasks != 2:
+            raise ValueError(
+                "PinkDualDifferentialInverseKinematicsAction expects exactly 2 "
+                f"FrameTask entries in variable_input_tasks, got {self._num_frame_tasks}."
+            )
+
+    def _build_diff_offset(self, offset_cfg):
+        if offset_cfg is None:
+            return None, None
+        pos = torch.tensor(offset_cfg.pos, device=self.device).repeat(self.num_envs, 1)
+        rot = torch.tensor(offset_cfg.rot, device=self.device).repeat(self.num_envs, 1)
+        return pos, rot
+
+    # --- per-arm geometry helpers (mirror DualDifferentialInverseKinematicsAction)
+
+    def _arm_pose_in_ref(
+        self,
+        env_ids,
+        body_idx,
+        ref_idx,
+        offset_pos,
+        offset_rot,
+    ):
+        ee_pos_w = self._asset.data.body_pos_w[env_ids, body_idx]
+        ee_quat_w = self._asset.data.body_quat_w[env_ids, body_idx]
+        if ref_idx is not None:
+            ref_pos_w = self._asset.data.body_pos_w[env_ids, ref_idx]
+            ref_quat_w = self._asset.data.body_quat_w[env_ids, ref_idx]
+        else:
+            ref_pos_w = self._asset.data.root_pos_w[env_ids]
+            ref_quat_w = self._asset.data.root_quat_w[env_ids]
+        ee_pos_b, ee_quat_b = math_utils.subtract_frame_transforms(
+            ref_pos_w, ref_quat_w, ee_pos_w, ee_quat_w
+        )
+        if offset_pos is not None:
+            ee_pos_b, ee_quat_b = math_utils.combine_frame_transforms(
+                ee_pos_b, ee_quat_b, offset_pos[env_ids], offset_rot[env_ids]
+            )
+        return ee_pos_b, ee_quat_b
+
+    def _arm_jacobian(
+        self,
+        env_ids,
+        jacobi_body_idx,
+        jacobi_joint_ids,
+        ref_idx,
+        offset_pos,
+        offset_rot,
+    ):
+        jac_w = self._asset.root_physx_view.get_jacobians()[
+            :, jacobi_body_idx, :, jacobi_joint_ids
+        ][env_ids]
+        if ref_idx is not None:
+            ref_quat = self._asset.data.body_quat_w[env_ids, ref_idx]
+        else:
+            ref_quat = self._asset.data.root_quat_w[env_ids]
+        R = math_utils.matrix_from_quat(math_utils.quat_inv(ref_quat))
+        jac = jac_w.clone()
+        jac[:, 0:3, :] = torch.bmm(R, jac[:, 0:3, :])
+        jac[:, 3:, :] = torch.bmm(R, jac[:, 3:, :])
+        if offset_pos is not None:
+            skew = math_utils.skew_symmetric_matrix(offset_pos[env_ids])
+            jac[:, 0:3, :] += torch.bmm(-skew, jac[:, 3:, :])
+            jac[:, 3:, :] = torch.bmm(
+                math_utils.matrix_from_quat(offset_rot[env_ids]), jac[:, 3:, :]
+            )
+        return jac
+
+    def _transform_command_to_ref(self, command, env_ids, ref_idx):
+        """Convert an env-local world-frame pose command into the arm's ref frame."""
+        if ref_idx is not None:
+            base_pos_w = self._asset.data.body_pos_w[env_ids, ref_idx]
+            base_quat_w = self._asset.data.body_quat_w[env_ids, ref_idx]
+        else:
+            base_pos_w = self._asset.data.root_pos_w[env_ids]
+            base_quat_w = self._asset.data.root_quat_w[env_ids]
+        env_origins = getattr(self._env.scene, "env_origins", None)
+        positions_w = command[:, 0:3]
+        if env_origins is not None:
+            positions_w = positions_w + env_origins[env_ids]
+        if self.cfg.diff_ik_controller.command_type == "position":
+            pos_b, _ = math_utils.subtract_frame_transforms(
+                base_pos_w, base_quat_w, positions_w, None
+            )
+            return pos_b
+        quats_w = command[:, 3:7]
+        pos_b, quat_b = math_utils.subtract_frame_transforms(
+            base_pos_w, base_quat_w, positions_w, quats_w
+        )
+        return torch.cat((pos_b, quat_b), dim=1)
+
+    # --- overridden apply ---------------------------------------------------
+
+    def apply_actions(self) -> None:
+        self._step_count += 1
+
+        if len(self._env_ids) == 0:
+            return
+
+        is_pink_fire = (self._step_count - 1) % max(1, self.cfg.decimation) == 0
+
+        if is_pink_fire:
+            ik_joint_positions = self._compute_ik_solutions()
+            if self.hand_joint_dim > 0:
+                all_joint_positions = torch.cat(
+                    (ik_joint_positions, self._target_hand_joint_positions), dim=1
+                )
+            else:
+                all_joint_positions = ik_joint_positions
+            self._processed_actions[self._env_ids] = all_joint_positions
+
+            if self.cfg.enable_gravity_compensation:
+                self._apply_gravity_compensation()
+
+            self._asset.set_joint_position_target(
+                self._processed_actions[self._env_ids],
+                self._controlled_joint_ids,
+                self._env_ids,
+            )
+        else:
+            self._apply_dual_diff_ik()
+
+    def _apply_dual_diff_ik(self) -> None:
+        """Per-arm Jacobian tracking toward the stored Pink pose target."""
+        raw = self._raw_actions[self._env_ids]
+        right_cmd = raw[:, : self.pose_dim]
+        left_cmd = raw[:, self.pose_dim : 2 * self.pose_dim]
+
+        # Pink interprets actions in env-local world frame when relative_to_base is False.
+        if not self.cfg.relative_to_base:
+            right_cmd = self._transform_command_to_ref(
+                right_cmd, self._env_ids, self._right_ref_idx
+            )
+            left_cmd = self._transform_command_to_ref(
+                left_cmd, self._env_ids, self._left_ref_idx
+            )
+
+        r_pos, r_quat = self._arm_pose_in_ref(
+            self._env_ids,
+            self._right_body_idx,
+            self._right_ref_idx,
+            self._right_offset_pos,
+            self._right_offset_rot,
+        )
+        l_pos, l_quat = self._arm_pose_in_ref(
+            self._env_ids,
+            self._left_body_idx,
+            self._left_ref_idx,
+            self._left_offset_pos,
+            self._left_offset_rot,
+        )
+
+        self._right_diff_ik.set_command(right_cmd, r_pos, r_quat)
+        self._left_diff_ik.set_command(left_cmd, l_pos, l_quat)
+
+        r_q = self._asset.data.joint_pos[
+            self._env_ids.unsqueeze(1), self._right_joint_ids
+        ]
+        l_q = self._asset.data.joint_pos[
+            self._env_ids.unsqueeze(1), self._left_joint_ids
+        ]
+        r_jac = self._arm_jacobian(
+            self._env_ids,
+            self._right_jacobi_body_idx,
+            self._right_jacobi_joint_ids,
+            self._right_ref_idx,
+            self._right_offset_pos,
+            self._right_offset_rot,
+        )
+        l_jac = self._arm_jacobian(
+            self._env_ids,
+            self._left_jacobi_body_idx,
+            self._left_jacobi_joint_ids,
+            self._left_ref_idx,
+            self._left_offset_pos,
+            self._left_offset_rot,
+        )
+
+        r_q_des = self._right_diff_ik.compute(r_pos, r_quat, r_jac, r_q)
+        l_q_des = self._left_diff_ik.compute(l_pos, l_quat, l_jac, l_q)
+
+        r_lo = self._asset.data.joint_pos_limits[0, self._right_joint_ids, 0]
+        r_hi = self._asset.data.joint_pos_limits[0, self._right_joint_ids, 1]
+        l_lo = self._asset.data.joint_pos_limits[0, self._left_joint_ids, 0]
+        l_hi = self._asset.data.joint_pos_limits[0, self._left_joint_ids, 1]
+        r_q_des = torch.clamp(r_q_des, min=r_lo, max=r_hi)
+        l_q_des = torch.clamp(l_q_des, min=l_lo, max=l_hi)
+
+        if self.cfg.enable_gravity_compensation:
+            self._apply_gravity_compensation()
+
+        self._asset.set_joint_position_target(
+            r_q_des, self._right_joint_ids, self._env_ids
+        )
+        self._asset.set_joint_position_target(
+            l_q_des, self._left_joint_ids, self._env_ids
+        )

debug/536/n052_sawyer/pinklog_armlock_2026-11-09/pink_task_space_actions_log.py

@@ -0,0 +1,1159 @@
+# Copyright (c) 2022-2025, The Isaac Lab Project Developers (https://github.com/isaac-sim/IsaacLab/blob/main/CONTRIBUTORS.md).
+# All rights reserved.
+#
+# SPDX-License-Identifier: BSD-3-Clause
+
+from __future__ import annotations
+
+import os
+
+import torch
+from collections.abc import Sequence
+from typing import TYPE_CHECKING
+from dataclasses import MISSING
+
+from pink.tasks import FrameTask
+
+from gymnasium import spaces
+
+import isaaclab.utils.math as math_utils
+from isaaclab.assets.articulation import Articulation
+from magicsim.Env.Robot.mdp.pink_ik import LocalFrameTask, PinkIKController
+from magicsim.Env.Robot.mdp.action_manager import ActionTerm
+from magicsim.Env.Robot.mdp.differential_ik import DifferentialIKController
+
+if TYPE_CHECKING:
+    from isaaclab.envs import ManagerBasedEnv
+    from isaaclab.envs.utils.io_descriptors import GenericActionIODescriptor
+
+    from . import pink_actions_cfg
+
+
+class PinkInverseKinematicsAction(ActionTerm):
+    r"""Pink Inverse Kinematics action term.
+
+    This action term processes the action tensor and sets these setpoints in the pink IK framework.
+    The action tensor is ordered in the order of the tasks defined in PinkIKControllerCfg.
+    """
+
+    cfg: pink_actions_cfg.PinkInverseKinematicsActionCfg
+    """Configuration for the Pink Inverse Kinematics action term."""
+
+    _asset: Articulation
+    """The articulation asset to which the action term is applied."""
+
+    def __init__(
+        self, cfg: pink_actions_cfg.PinkInverseKinematicsActionCfg, env: ManagerBasedEnv
+    ):
+        """Initialize the Pink Inverse Kinematics action term.
+
+        Args:
+            cfg: The configuration for this action term.
+            env: The environment in which the action term will be applied.
+        """
+        self._step_count = 0
+        super().__init__(cfg, env)
+
+        self._env = env
+        self._sim_dt = env.sim.get_physics_dt()
+
+        # Initialize joint information
+        self._initialize_joint_info()
+
+        # Initialize IK controllers
+        self._initialize_ik_controllers()
+
+        # Initialize action tensors
+        self._raw_actions = torch.zeros(
+            self.num_envs, self.action_dim, device=self.device
+        )
+        # Last passed (valid) action per env; used when input becomes NaN (first time: use current EEF)
+        self._last_passed_action = torch.zeros_like(self._raw_actions)
+        # NaN=HOLD anchors (cfg.nan_hold, 2026-10-02 idle-arm flail fix):
+        # per frame-task group, the pose captured when the group turned NaN —
+        # commanded as a CONSTANT target until a real target returns. NaN =
+        # no anchor armed.
+        self._nan_anchor = torch.full_like(self._raw_actions, float("nan"))
+        # _processed_actions stores joint positions output from IK, shape (num_envs, num_controlled_joints)
+        self._processed_actions = torch.zeros(
+            self.num_envs, len(self._controlled_joint_ids), device=self.device
+        )
+
+        # PhysX Articulation Floating joint indices offset from IsaacLab Articulation joint indices
+        self._physx_floating_joint_indices_offset = 6
+
+        # Initialize env_ids (all envs by default)
+        self._env_ids = torch.arange(self.num_envs, device=self.device)
+        # Wrist override (opt-in per env, see set_wrist_override): while set,
+        # the named frame's task ignores rotation about its own z (5-DoF) and
+        # the named joint is driven straight to the given target instead of
+        # whatever the IK solution says. Lets a skill spin the wrist in joint
+        # space (a screw unwind) with the TCP pinned on the wrist axis by
+        # construction, instead of asking the IK to track a rotating pose.
+        # {env_id: (task_index, joint_column, target_rad)}
+        self._wrist_override: dict[int, tuple] = {}
+
+        # Pre-allocate tensors for runtime use
+        self._initialize_helper_tensors()
+
+        # Initialize action space
+        self._action_space = spaces.Box(
+            low=self.cfg.action_space[0].cpu().numpy(),
+            high=self.cfg.action_space[1].cpu().numpy(),
+            dtype=float,
+        )
+
+    def _initialize_joint_info(self) -> None:
+        """Initialize joint IDs and names based on configuration."""
+        # Resolve pink controlled joints
+        self._isaaclab_controlled_joint_ids, self._isaaclab_controlled_joint_names = (
+            self._asset.find_joints(self.cfg.pink_controlled_joint_names)
+        )
+        self.cfg.controller.joint_names = self._isaaclab_controlled_joint_names
+        self._isaaclab_all_joint_ids = list(range(len(self._asset.data.joint_names)))
+        self.cfg.controller.all_joint_names = self._asset.data.joint_names
+
+        # Resolve hand joints (optional)
+        if self.cfg.hand_joint_names not in (None, MISSING):
+            self._hand_joint_ids, self._hand_joint_names = self._asset.find_joints(
+                self.cfg.hand_joint_names
+            )
+        else:
+            self._hand_joint_ids, self._hand_joint_names = [], []
+
+        # Combine all joint information
+        self._controlled_joint_ids = (
+            self._isaaclab_controlled_joint_ids + self._hand_joint_ids
+        )
+        self._controlled_joint_names = (
+            self._isaaclab_controlled_joint_names + self._hand_joint_names
+        )
+        assert len(self._controlled_joint_ids) == self.cfg.num_joints, (
+            f"Number of controlled joints does not match the number of joints to control: {len(self._controlled_joint_ids)} != {self.cfg.num_joints}"
+        )
+
+    def _initialize_ik_controllers(self) -> None:
+        """Initialize Pink IK controllers for all environments."""
+        assert self._env.num_envs > 0, (
+            "Number of environments specified are less than 1."
+        )
+
+        self._ik_controllers = []
+        robot_cfg = self._asset.cfg
+        for _ in range(self._env.num_envs):
+            controller_cfg = self.cfg.controller.copy()
+            # If urdf_path is not specified in controller config, try to get it from various sources
+            if controller_cfg.urdf_path is None:
+                # 1. Try to get from asset configuration
+                if hasattr(robot_cfg, "urdf_path"):
+                    controller_cfg.urdf_path = robot_cfg.urdf_path
+                # 2. Try to get from spawn configuration if it's a UsdFileCfg (common in IsaacLab)
+                elif hasattr(robot_cfg.spawn, "urdf_path"):
+                    controller_cfg.urdf_path = robot_cfg.spawn.urdf_path
+                # 3. Try to infer from G1 specific configs if available
+                # Note: This is a fallback for G1 robot specific structure
+                elif (
+                    hasattr(robot_cfg, "spawn")
+                    and hasattr(robot_cfg.spawn, "usd_path")
+                    and "g1" in str(robot_cfg.spawn.usd_path).lower()
+                ):
+                    # Attempt to find URDF in expected location relative to asset root or known path
+                    pass
+
+            self._ik_controllers.append(
+                PinkIKController(
+                    cfg=controller_cfg,
+                    robot_cfg=robot_cfg,
+                    device=self.device,
+                    controlled_joint_indices=self._isaaclab_controlled_joint_ids,
+                )
+            )
+
+    def _initialize_helper_tensors(self) -> None:
+        """Pre-allocate tensors and cache values for performance optimization."""
+        # Cache frequently used tensor versions of joint IDs to avoid repeated creation
+        self._controlled_joint_ids_tensor = torch.tensor(
+            self._controlled_joint_ids, device=self.device
+        )
+
+        # Cache base link index to avoid string lookup every time
+        if self.cfg.controller.base_link_name == "root":
+            self._base_link_idx = None
+        else:
+            articulation_data = self._env.scene[
+                self.cfg.controller.articulation_name
+            ].data
+            self._base_link_idx = articulation_data.body_names.index(
+                self.cfg.controller.base_link_name
+            )
+
+        # Pre-allocate working tensors
+        # Count only FrameTask instances in variable_input_tasks (not all tasks)
+        num_frame_tasks = sum(
+            1
+            for task in self._ik_controllers[0].cfg.variable_input_tasks
+            if isinstance(task, FrameTask)
+        )
+        self._num_frame_tasks = num_frame_tasks
+        self._controlled_frame_poses = torch.zeros(
+            num_frame_tasks, self.num_envs, 4, 4, device=self.device
+        )
+
+        # Pre-allocate tensor for base frame computations
+        self._base_link_frame_buffer = torch.zeros(
+            self.num_envs, 4, 4, device=self.device
+        )
+
+        # Cache EEF body indices for NaN action handling
+        # Maps frame task index → body index in articulation data
+        self._eef_body_indices: list[int] = []
+        if self.cfg.target_eef_link_names not in (None, MISSING):
+            for link_name in self.cfg.target_eef_link_names.values():
+                body_idx = self._asset.data.body_names.index(link_name)
+                self._eef_body_indices.append(body_idx)
+
+    # ==================== Properties ====================
+
+    @property
+    def hand_joint_dim(self) -> int:
+        """Dimension for hand joint positions."""
+        return self.cfg.controller.num_hand_joints
+
+    @property
+    def position_dim(self) -> int:
+        """Dimension for position (x, y, z)."""
+        return 3
+
+    @property
+    def orientation_dim(self) -> int:
+        """Dimension for orientation (w, x, y, z)."""
+        return 4
+
+    @property
+    def pose_dim(self) -> int:
+        """Total pose dimension (position + orientation)."""
+        return self.position_dim + self.orientation_dim
+
+    @property
+    def action_dim(self) -> int:
+        """Dimension of the action space (based on number of tasks and pose dimension)."""
+        # Count only FrameTask instances in variable_input_tasks
+        frame_tasks_count = sum(
+            1
+            for task in self._ik_controllers[0].cfg.variable_input_tasks
+            if isinstance(task, FrameTask)
+        )
+        return frame_tasks_count * self.pose_dim + self.hand_joint_dim
+
+    @property
+    def env_ids(self) -> torch.Tensor:
+        """Get the current valid environment IDs (excluding NaN action envs)."""
+        return self._env_ids
+
+    @property
+    def raw_actions(self) -> torch.Tensor:
+        """Get the raw actions tensor."""
+        return self._raw_actions
+
+    @property
+    def processed_actions(self) -> torch.Tensor:
+        """Get the processed actions tensor."""
+        return self._processed_actions
+
+    @property
+    def action_space(self) -> spaces.Box:
+        """Get the action space."""
+        return self._action_space
+
+    @property
+    def IO_descriptor(self) -> GenericActionIODescriptor:
+        """The IO descriptor of the action term.
+
+        This descriptor is used to describe the action term of the pink inverse kinematics action.
+        It adds the following information to the base descriptor:
+        - scale: The scale of the action term.
+        - offset: The offset of the action term.
+        - clip: The clip of the action term.
+        - pink_controller_joint_names: The names of the pink controller joints.
+        - hand_joint_names: The names of the hand joints.
+        - controller_cfg: The configuration of the pink controller.
+
+        Returns:
+            The IO descriptor of the action term.
+        """
+        super().IO_descriptor
+        self._IO_descriptor.shape = (self.action_dim,)
+        self._IO_descriptor.dtype = str(self.raw_actions.dtype)
+        self._IO_descriptor.action_type = "PinkInverseKinematicsAction"
+        self._IO_descriptor.pink_controller_joint_names = (
+            self._isaaclab_controlled_joint_names
+        )
+        self._IO_descriptor.hand_joint_names = self._hand_joint_names
+        self._IO_descriptor.extras["controller_cfg"] = self.cfg.controller.__dict__
+        return self._IO_descriptor
+
+    # """
+    # Operations.
+    # """
+
+    def _handle_nan_actions(
+        self, actions: torch.Tensor, selected_env_ids: torch.Tensor
+    ) -> tuple[torch.Tensor | None, torch.Tensor]:
+        """Handle NaN actions: use last passed action if available, else current EEF pose.
+
+        Actions are structured as groups of pose_dim (7) per frame task, optionally
+        followed by hand joint values. The logic is:
+
+        1. Include all envs by default (do not skip all-NaN envs).
+        2. For each group of 7 that has any NaN: use last passed action if valid;
+           else (no valid previous action) behavior depends on
+           ``cfg.fallback_to_current``:
+             - True (default): fall back to current EEF pose.
+             - False: skip this env (drop it from the returned batch).
+        3. For hand joints: same logic — last passed if valid, else current joint
+           positions (or skip when ``fallback_to_current`` is False).
+
+        Args:
+            actions: Input action tensor of shape ``(len(selected_env_ids), action_dim)``.
+                Rows are **already** the slice for those envs (same order as
+                ``selected_env_ids``), as produced by ``ActionManager.process_action``.
+                Do not index again with global env ids as row indices.
+            selected_env_ids: Selected environment IDs (global indices).
+
+        Returns:
+            Tuple of (processed_actions, valid_env_ids).
+            processed_actions is None if no valid envs exist.
+        """
+        env_ids = selected_env_ids
+        if len(env_ids) == 0:
+            return None, env_ids
+
+        if actions.shape[0] != len(env_ids):
+            raise ValueError(
+                f"Pink IK actions batch ({actions.shape[0]}) must match len(env_ids) ({len(env_ids)}); "
+                "actions should be the env slice in env_ids order, not the full buffer indexed by id."
+            )
+
+        valid_actions = actions.clone()
+        fallback_to_current = self.cfg.fallback_to_current
+        skip_mask = torch.zeros(
+            len(env_ids), dtype=torch.bool, device=valid_actions.device
+        )
+
+        # For each group of pose_dim (7), fill NaN with last passed action or current EEF pose
+        for task_idx in range(self._num_frame_tasks):
+            start = task_idx * self.pose_dim
+            end = start + self.pose_dim
+
+            group_nan_mask = torch.isnan(valid_actions[:, start:end]).any(dim=1)
+            if getattr(self.cfg, "nan_hold", False):
+                # NaN=HOLD (2026-10-02 idle-arm flail): a NaN group commands
+                # the pose ANCHORED when it turned NaN — a constant target
+                # pink holds with real stiffness. The legacy fallback
+                # ("last passed, else current pose") is a zero-stiffness
+                # follow target: the idle arm drifts (null-space posture
+                # pull + gravity-comp residual) and the drift is chased.
+                real_mask = ~group_nan_mask
+                if real_mask.any():
+                    # live target -> disarm the anchor
+                    self._nan_anchor[env_ids[real_mask], start:end] = float("nan")
+                if group_nan_mask.any():
+                    anchors = self._nan_anchor[env_ids, start:end]
+                    need_cap = group_nan_mask & torch.isnan(anchors).any(dim=1)
+                    if need_cap.any():
+                        cap_ids = env_ids[need_cap]
+                        self._nan_anchor[cap_ids, start:end] = (
+                            self._get_current_eef_pose(cap_ids, task_idx)
+                        )
+                    valid_actions[group_nan_mask, start:end] = self._nan_anchor[
+                        env_ids[group_nan_mask], start:end
+                    ]
+                continue
+            if group_nan_mask.any():
+                nan_env_ids = env_ids[group_nan_mask]
+                last_poses = self._last_passed_action[nan_env_ids, start:end]
+                has_valid_last = (~torch.isnan(last_poses)).all(dim=1) & (
+                    last_poses[:, 3:7].abs().sum(dim=1) > 0.01
+                )
+                if fallback_to_current:
+                    current_poses = self._get_current_eef_pose(nan_env_ids, task_idx)
+                    fill_values = torch.where(
+                        has_valid_last.unsqueeze(1).expand_as(current_poses),
+                        last_poses,
+                        current_poses,
+                    )
+                    valid_actions[group_nan_mask, start:end] = fill_values
+                else:
+                    # Use last_poses only where valid; mark the rest for skipping.
+                    valid_actions[group_nan_mask, start:end] = last_poses
+                    group_local_idx = group_nan_mask.nonzero(as_tuple=True)[0]
+                    skip_mask[group_local_idx[~has_valid_last]] = True
+
+        # Handle NaN in hand joints: last passed if valid, else current joint positions
+        if self.hand_joint_dim > 0:
+            hand_start = self._num_frame_tasks * self.pose_dim
+            hand_nan_mask = torch.isnan(valid_actions[:, hand_start:]).any(dim=1)
+            if hand_nan_mask.any():
+                nan_env_ids = env_ids[hand_nan_mask]
+                last_hand = self._last_passed_action[nan_env_ids, hand_start:]
+                has_valid_last_hand = ~torch.isnan(last_hand).any(dim=1)
+                if fallback_to_current:
+                    current_hand_pos = self._asset.data.joint_pos[nan_env_ids][
+                        :, self._hand_joint_ids
+                    ]
+                    fill_hand = torch.where(
+                        has_valid_last_hand.unsqueeze(1).expand_as(current_hand_pos),
+                        last_hand,
+                        current_hand_pos,
+                    )
+                    valid_actions[hand_nan_mask, hand_start:] = fill_hand
+                else:
+                    valid_actions[hand_nan_mask, hand_start:] = last_hand
+                    hand_local_idx = hand_nan_mask.nonzero(as_tuple=True)[0]
+                    skip_mask[hand_local_idx[~has_valid_last_hand]] = True
+
+        if skip_mask.any():
+            keep_mask = ~skip_mask
+            valid_actions = valid_actions[keep_mask]
+            env_ids = env_ids[keep_mask]
+            if len(env_ids) == 0:
+                return None, env_ids
+
+        return valid_actions, env_ids
+
+    def _get_current_eef_pose(
+        self, env_ids: torch.Tensor, task_idx: int
+    ) -> torch.Tensor:
+        """Get current EEF pose for specified envs and frame task index.
+
+        Returns the pose as [pos_x, pos_y, pos_z, quat_w, quat_x, quat_y, quat_z]
+        in the same coordinate frame as the input actions (base-relative when
+        ``relative_to_base`` is True, otherwise env-origin-relative world frame).
+
+        Args:
+            env_ids: Environment IDs for which to get the current EEF pose.
+            task_idx: Frame task index (0-based, counting only FrameTask instances).
+
+        Returns:
+            Current EEF pose tensor of shape ``(len(env_ids), 7)``.
+        """
+        body_idx = self._eef_body_indices[task_idx]
+
+        # Get world-frame pose from articulation data: [pos(3), quat(4)]
+        eef_state_w = self._asset.data.body_link_state_w[env_ids, body_idx, :7]
+        eef_pos_w = eef_state_w[:, :3]
+        eef_quat_w = eef_state_w[:, 3:7]
+
+        # Convert position to env-origin-relative frame
+        eef_pos = eef_pos_w - self._env.scene.env_origins[env_ids]
+
+        if self.cfg.relative_to_base:
+            # Compute base link pose in env-origin frame
+            if self.cfg.controller.base_link_name == "root":
+                base_pos = (
+                    self._asset.data.root_pos_w[env_ids]
+                    - self._env.scene.env_origins[env_ids]
+                )
+                base_quat = self._asset.data.root_quat_w[env_ids]
+            else:
+                articulation_data = self._env.scene[
+                    self.cfg.controller.articulation_name
+                ].data
+                base_state = articulation_data.body_link_state_w[
+                    env_ids, self._base_link_idx, :7
+                ]
+                base_pos = base_state[:, :3] - self._env.scene.env_origins[env_ids]
+                base_quat = base_state[:, 3:7]
+
+            # Express EEF pose relative to the base link frame
+            eef_pos_base, eef_quat_base = math_utils.subtract_frame_transforms(
+                base_pos, base_quat, eef_pos, eef_quat_w
+            )
+            return torch.cat([eef_pos_base, eef_quat_base], dim=1)
+        else:
+            return torch.cat([eef_pos, eef_quat_w], dim=1)
+
+    def process_actions(self, actions: torch.Tensor, env_ids: torch.Tensor) -> None:
+        """Process the input actions and set targets for each task.
+
+        Args:
+            actions: The input actions tensor.
+            env_ids: The environment IDs to process.
+        """
+        if self._step_count % self.cfg.decimation != 0:
+            return
+
+        # Handle env_ids
+        if env_ids is None:
+            selected_env_ids = torch.arange(self.num_envs, device=self.device)
+        elif not isinstance(env_ids, torch.Tensor):
+            selected_env_ids = torch.tensor(
+                env_ids, device=self.device, dtype=torch.int32
+            )
+        else:
+            selected_env_ids = env_ids
+
+        # Handle NaN actions: skip all-NaN envs, fill partial NaN groups with current EEF pose
+        valid_actions, self._env_ids = self._handle_nan_actions(
+            actions, selected_env_ids
+        )
+        if os.environ.get("MMB_PINK_LOG"):
+            with open(os.environ["MMB_PINK_LOG"], "a") as _f:
+                _raw = actions[0].tolist() if actions is not None and len(actions) else None
+                _f.write("IN %d raw=%s valid=%s envs=%s\n" % (self._step_count, _raw, None if valid_actions is None else valid_actions[0].tolist(), self._env_ids.tolist()))
+
+        # If no valid envs, return early
+        if valid_actions is None or len(self._env_ids) == 0:
+            return
+
+        # Store raw actions and record last passed action for NaN fallback
+        self._raw_actions[self._env_ids] = valid_actions
+        self._last_passed_action[self._env_ids] = valid_actions.clone()
+
+        # Extract hand joint positions directly (no cloning needed)
+        # Handle case when hand_joint_dim = 0: -0 is same as 0, so slice would be whole tensor
+        if self.hand_joint_dim > 0:
+            self._target_hand_joint_positions = valid_actions[:, -self.hand_joint_dim :]
+        else:
+            self._target_hand_joint_positions = torch.zeros(
+                (valid_actions.shape[0], 0), device=self.device
+            )
+
+        # Get base link frame transformation for valid envs only
+        all_base_link_frames = self._get_base_link_frame_transform()
+        self.base_link_frame_in_world_rf = all_base_link_frames[self._env_ids]
+
+        # Process controlled frame poses for valid envs
+        controlled_frame_poses = self._extract_controlled_frame_poses(valid_actions)
+
+        # DEBUG: Print raw action and controlled frame poses
+        # print(f"\n=== Pink IK process_actions DEBUG ===")
+        # print(f"Raw action (first env): pos={valid_actions[0, :3].cpu().numpy()}, quat={valid_actions[0, 3:7].cpu().numpy()}")
+        # print(f"relative_to_base: {self.cfg.relative_to_base}")
+        # print(f"Controlled frame poses (world frame):\n{controlled_frame_poses[0, 0].cpu().numpy()}")
+
+        # NOTE(magics): We assume actions are already in the Base Link Frame (Pelvis) for LocalFrameTask.
+        # This bypasses the World-to-Base transformation to allow direct local control.
+        # If using standard FrameTask (World Frame), this would imply inputs are also in Local Frame
+        # but interpreted as World Frame, which might be incorrect for mixed usage.
+        # But for G1 LocalFrameTask usage, this is the correct behavior for Body-relative control.
+        if self.cfg.relative_to_base:
+            positions, rotation_matrices = math_utils.unmake_pose(
+                controlled_frame_poses
+            )
+            transformed_poses = (positions, rotation_matrices)
+        else:
+            transformed_poses = self._transform_poses_to_base_link_frame(
+                controlled_frame_poses
+            )
+            # DEBUG: Print after transformation
+            # print(f"After transform_to_base_link_frame:")
+            # print(f"  positions: {transformed_poses[0].cpu().numpy()}")
+            # print(f"  rotation_matrices: {transformed_poses[1][0, 0].cpu().numpy()}")
+
+        # Set targets for valid envs only
+        self._set_task_targets(transformed_poses)
+
+    def _get_base_link_frame_transform(self) -> torch.Tensor:
+        """Get the base link frame transformation matrix.
+
+        Returns:
+            Base link frame transformation matrix.
+        """
+        if self.cfg.controller.base_link_name == "root":
+            base_pos_w = self._asset.data.root_pos_w
+            base_quat_w = self._asset.data.root_quat_w
+            torch.sub(
+                base_pos_w,
+                self._env.scene.env_origins,
+                out=self._base_link_frame_buffer[:, :3, 3],
+            )
+            return math_utils.make_pose(
+                self._base_link_frame_buffer[:, :3, 3],
+                math_utils.matrix_from_quat(base_quat_w),
+            )
+
+        # Get base link frame pose in world origin using cached index
+        articulation_data = self._env.scene[self.cfg.controller.articulation_name].data
+        base_link_frame_in_world_origin = articulation_data.body_link_state_w[
+            :, self._base_link_idx, :7
+        ]
+
+        # Transform to environment origin frame (reuse buffer to avoid allocation)
+        torch.sub(
+            base_link_frame_in_world_origin[:, :3],
+            self._env.scene.env_origins,
+            out=self._base_link_frame_buffer[:, :3, 3],
+        )
+
+        # Copy orientation (avoid clone)
+        base_link_frame_quat = base_link_frame_in_world_origin[:, 3:7]
+
+        # Create transformation matrix
+        return math_utils.make_pose(
+            self._base_link_frame_buffer[:, :3, 3],
+            math_utils.matrix_from_quat(base_link_frame_quat),
+        )
+
+    def _extract_controlled_frame_poses(self, actions: torch.Tensor) -> torch.Tensor:
+        """Extract controlled frame poses from action tensor for valid envs.
+
+        Args:
+            actions: The action tensor for valid envs, shape (num_valid_envs, action_dim).
+
+        Returns:
+            Controlled frame poses tensor, shape (num_frame_tasks, num_valid_envs, 4, 4).
+        """
+        num_valid_envs = actions.shape[0]
+        # Create tensor for valid envs (cannot use pre-allocated when size varies)
+        controlled_frame_poses = torch.zeros(
+            self._num_frame_tasks, num_valid_envs, 4, 4, device=self.device
+        )
+
+        for task_index in range(self._num_frame_tasks):
+            # Extract position and orientation for this task
+            pos_start = task_index * self.pose_dim
+            pos_end = pos_start + self.position_dim
+            quat_start = pos_end
+            quat_end = (task_index + 1) * self.pose_dim
+
+            position = actions[:, pos_start:pos_end]
+            quaternion = actions[:, quat_start:quat_end]
+
+            # Create pose matrix
+            controlled_frame_poses[task_index] = math_utils.make_pose(
+                position, math_utils.matrix_from_quat(quaternion)
+            )
+
+        return controlled_frame_poses
+
+    def _transform_poses_to_base_link_frame(
+        self, poses: torch.Tensor
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        """Transform poses from world frame to base link frame.
+
+        Args:
+            poses: Poses in world frame.
+
+        Returns:
+            Tuple of (positions, rotation_matrices) in base link frame.
+        """
+        # Transform poses to base link frame
+        base_link_inv = math_utils.pose_inv(self.base_link_frame_in_world_rf)
+        transformed_poses = math_utils.pose_in_A_to_pose_in_B(poses, base_link_inv)
+
+        # Extract position and rotation
+        positions, rotation_matrices = math_utils.unmake_pose(transformed_poses)
+
+        return positions, rotation_matrices
+
+    def _set_task_targets(
+        self, transformed_poses: tuple[torch.Tensor, torch.Tensor]
+    ) -> None:
+        """Set targets for all tasks across valid environments only.
+
+        Args:
+            transformed_poses: Tuple of (positions, rotation_matrices) in base link frame,
+                               with shape (num_frame_tasks, num_valid_envs, ...).
+        """
+        positions, rotation_matrices = transformed_poses
+
+        # DEBUG: Print base link frame and transformed target positions
+        # print(f"\n=== Pink IK _set_task_targets DEBUG ===")
+        # print(f"base_link_frame_in_world_rf (env 0):\n{self.base_link_frame_in_world_rf[0].cpu().numpy()}")
+        # print(f"Target positions (in base link frame): {positions.cpu().numpy()}")
+
+        # Iterate only over valid envs
+        for valid_idx, env_index in enumerate(self._env_ids.tolist()):
+            ik_controller = self._ik_controllers[env_index]
+            for frame_task_index, task in enumerate(
+                ik_controller.cfg.variable_input_tasks
+            ):
+                if isinstance(task, LocalFrameTask):
+                    target = task.transform_target_to_base
+                elif isinstance(task, FrameTask):
+                    target = task.transform_target_to_world
+                else:
+                    continue
+
+                # Use valid_idx to index into transformed_poses
+                target.translation = (
+                    positions[frame_task_index, valid_idx, :].cpu().numpy()
+                )
+                target.rotation = (
+                    rotation_matrices[frame_task_index, valid_idx, :].cpu().numpy()
+                )
+
+                task.set_target(target)
+
+    def set_wrist_override(
+        self, env_id: int, frame: str, joint_name: str, target: float | None
+    ) -> bool:
+        """Drive ``joint_name`` of ``env_id`` straight to ``target`` (rad)
+        while the task for ``frame`` ignores rotation about its own z.
+        ``target=None`` clears the override (6-DoF task restored). Returns
+        False when the frame or joint is unknown to this term."""
+        env_id = int(env_id)
+        if target is None:
+            self._wrist_override.pop(env_id, None)
+            return True
+        tasks = self._ik_controllers[env_id].cfg.variable_input_tasks
+        task_index = next(
+            (i for i, t in enumerate(tasks) if getattr(t, "frame", None) == frame),
+            None,
+        )
+        if (
+            task_index is None
+            or joint_name not in self._isaaclab_controlled_joint_names
+        ):
+            return False
+        col = self._isaaclab_controlled_joint_names.index(joint_name)
+        self._wrist_override[env_id] = (task_index, col, float(target))
+        return True
+
+    def _wrist_target_clamped(self, env_id: int, col: int, target: float) -> float:
+        joint_id = self._isaaclab_controlled_joint_ids[col]
+        lim = self._asset.data.joint_pos_limits[env_id, joint_id]
+        return float(min(max(target, float(lim[0])), float(lim[1])))
+
+    # ==================== Action Application ====================
+
+    def apply_actions(self) -> None:
+        """Apply the computed joint positions based on the inverse kinematics solution.
+
+        IK solve is only performed every ``cfg.decimation`` steps.  Cached
+        joint-position targets are applied every step so the robot never
+        stalls.
+        """
+        self._step_count += 1
+
+        if len(self._env_ids) == 0:
+            return
+
+        # Recompute IK only on decimation steps (matches process_actions gate)
+        if (self._step_count - 1) % self.cfg.decimation == 0:
+            ik_joint_positions = self._compute_ik_solutions()
+            if os.environ.get("MMB_PINK_LOG"):
+                with open(os.environ["MMB_PINK_LOG"], "a") as _f:
+                    _f.write("OUT %d q=%s sol=%s\n" % (self._step_count, [round(float(v), 5) for v in self._asset.data.joint_pos[0, self._controlled_joint_ids].tolist()], [round(float(v), 5) for v in ik_joint_positions[0].tolist()]))
+
+            if self.hand_joint_dim > 0:
+                all_joint_positions = torch.cat(
+                    (ik_joint_positions, self._target_hand_joint_positions), dim=1
+                )
+            else:
+                all_joint_positions = ik_joint_positions
+
+            self._processed_actions[self._env_ids] = all_joint_positions
+
+        for env_id, (_, col, target) in self._wrist_override.items():
+            self._processed_actions[env_id, col] = self._wrist_target_clamped(
+                env_id, col, target
+            )
+
+        # Always apply cached targets + gravity compensation
+        if self.cfg.enable_gravity_compensation:
+            self._apply_gravity_compensation()
+
+        self._asset.set_joint_position_target(
+            self._processed_actions[self._env_ids],
+            self._controlled_joint_ids,
+            self._env_ids,
+        )
+
+    def _apply_gravity_compensation(self) -> None:
+        """Apply gravity compensation to arm joints for valid envs only."""
+        if not self._asset.cfg.spawn.rigid_props.disable_gravity:
+            # Get gravity compensation forces using cached tensor for valid envs
+            all_forces = self._asset.root_physx_view.get_gravity_compensation_forces()
+            if self._asset.is_fixed_base:
+                if getattr(self.cfg, "fixed_base_gravity_compensation", False):
+                    # PhysX returns (count, max_dofs) for fixed-base articulations
+                    # (no root block), so the joint ids index directly — the same
+                    # indexing IsaacLab's own pink / OSC actions use.
+                    gravity = all_forces[self._env_ids][
+                        :, self._controlled_joint_ids_tensor
+                    ]
+                else:  # legacy: no compensation on fixed-base robots
+                    gravity = torch.zeros_like(
+                        all_forces[self._env_ids][:, self._controlled_joint_ids_tensor]
+                    )
+            else:
+                # If floating base, then need to skip the first 6 joints (base)
+                gravity = all_forces[self._env_ids][
+                    :,
+                    self._controlled_joint_ids_tensor
+                    + self._physx_floating_joint_indices_offset,
+                ]
+
+            # Apply gravity compensation to arm joints for valid envs
+            self._asset.set_joint_effort_target(
+                gravity, self._controlled_joint_ids, self._env_ids
+            )
+
+    def _solve_ik(self, ik_controller, current_joint_pos) -> torch.Tensor:
+        """One IK solve of one env: the iterative solver when the controller asks for more than one QP, else the historical call."""
+        if getattr(ik_controller.cfg, "ik_iterations", 1) > 1:
+            return ik_controller.compute(
+                current_joint_pos,
+                self._sim_dt,
+                control_dt=self._sim_dt * self.cfg.decimation,
+            )
+        # Keep the historical single-QP call and its 5 ms dt exactly.
+        return ik_controller.compute(current_joint_pos, self._sim_dt)
+
+    def _compute_ik_solutions(self) -> torch.Tensor:
+        """Compute IK solutions for valid environments only.
+
+        Returns:
+            IK joint positions tensor for valid environments.
+        """
+        ik_solutions = []
+
+        for env_index in self._env_ids.tolist():
+            ik_controller = self._ik_controllers[env_index]
+            # Get current joint positions for this environment
+            current_joint_pos = self._asset.data.joint_pos.cpu().numpy()[env_index]
+
+            ov = self._wrist_override.get(int(env_index))
+            if ov is None:
+                joint_pos_des = self._solve_ik(ik_controller, current_joint_pos)
+            else:
+                # The task objects are shared by every env's controller, so
+                # the cost is swapped only around this env's solve.
+                task_index, col, target = ov
+                task = ik_controller.cfg.variable_input_tasks[task_index]
+                saved = task.orientation_cost.copy()
+                free = saved.copy()
+                free[2] = 0.0
+                task.set_orientation_cost(free)
+                try:
+                    joint_pos_des = self._solve_ik(ik_controller, current_joint_pos)
+                finally:
+                    task.set_orientation_cost(saved)
+                joint_pos_des = joint_pos_des.clone()
+                joint_pos_des[col] = self._wrist_target_clamped(
+                    int(env_index), col, target
+                )
+            ik_solutions.append(joint_pos_des)
+
+        return torch.stack(ik_solutions)
+
+    # ==================== Reset ====================
+
+    def reset(self, env_ids: Sequence[int] | None = None) -> None:
+        """Reset the action term for specified environments.
+
+        Args:
+            env_ids: A list of environment IDs to reset. If None, all environments are reset.
+        """
+        self._raw_actions[env_ids] = torch.zeros(self.action_dim, device=self.device)
+        self._last_passed_action[env_ids] = torch.zeros(
+            self.action_dim, device=self.device
+        )
+        # Episode boundary: drop the NaN=HOLD anchors — the next NaN tick
+        # re-captures from the freshly reset pose (the yaml park).
+        self._nan_anchor[env_ids] = float("nan")
+        # Also clear ``_processed_actions`` (the cached IK solution applied
+        # in ``apply_actions`` every step). Without this, after a reset the
+        # next NaN-action tick has ``valid_actions=None`` and
+        # ``process_actions`` returns early — but ``apply_actions`` still
+        # ships the *previous* episode's IK joint targets to PhysX,
+        # snapping the arms back to whatever pose ended the last attempt.
+        # Replace with the robot's current joint positions ("hold here")
+        # so the freshly-reset default joint pose is what gets commanded.
+        if env_ids is None:
+            env_ids_tensor = torch.arange(self.num_envs, device=self.device)
+        elif isinstance(env_ids, torch.Tensor):
+            env_ids_tensor = env_ids.to(self.device, dtype=torch.long)
+        else:
+            env_ids_tensor = torch.tensor(
+                list(env_ids), device=self.device, dtype=torch.long
+            )
+        if env_ids_tensor.numel() > 0:
+            cur_joint_pos = self._asset.data.joint_pos[env_ids_tensor][
+                :, self._controlled_joint_ids
+            ]
+            self._processed_actions[env_ids_tensor] = cur_joint_pos
+            for e in env_ids_tensor.tolist():
+                self._wrist_override.pop(int(e), None)
+
+
+class PinkDualDifferentialInverseKinematicsAction(PinkInverseKinematicsAction):
+    """Hybrid Pink IK + per-arm differential IK action.
+
+    On decimation fires this runs the full Pink IK solve (via the parent
+    implementation). Between fires it falls back to per-arm Jacobian tracking
+    toward the *same* target, matching the computation in
+    :class:`DualDifferentialInverseKinematicsAction`. Hand-joint targets are
+    refreshed only on Pink fires; PhysX retains them between steps.
+
+    The action layout is identical to the parent: ``[right_pose(7), left_pose(7)]``
+    for dual-arm robots, because the Pink ``variable_input_tasks`` must be
+    ordered right-first, left-second.
+    """
+
+    cfg: "pink_actions_cfg.PinkDualDifferentialInverseKinematicsActionCfg"
+
+    def __init__(
+        self,
+        cfg: "pink_actions_cfg.PinkDualDifferentialInverseKinematicsActionCfg",
+        env,
+    ):
+        super().__init__(cfg, env)
+        self._setup_dual_diff()
+
+    def _setup_dual_diff(self) -> None:
+        cfg = self.cfg
+
+        self._right_joint_ids, _ = self._asset.find_joints(cfg.right_joint_names)
+        self._left_joint_ids, _ = self._asset.find_joints(cfg.left_joint_names)
+
+        r_ids, r_names = self._asset.find_bodies(cfg.right_body_name)
+        l_ids, l_names = self._asset.find_bodies(cfg.left_body_name)
+        if len(r_ids) != 1 or len(l_ids) != 1:
+            raise ValueError(
+                f"Expected one body match each: right={r_names}, left={l_names}"
+            )
+        self._right_body_idx = r_ids[0]
+        self._left_body_idx = l_ids[0]
+
+        self._right_ref_idx = None
+        self._left_ref_idx = None
+        if cfg.right_command_reference_body_name is not None:
+            ids, _ = self._asset.find_bodies(cfg.right_command_reference_body_name)
+            if len(ids) != 1:
+                raise ValueError(
+                    f"right_command_reference_body_name resolved to {len(ids)} bodies."
+                )
+            self._right_ref_idx = ids[0]
+        if cfg.left_command_reference_body_name is not None:
+            ids, _ = self._asset.find_bodies(cfg.left_command_reference_body_name)
+            if len(ids) != 1:
+                raise ValueError(
+                    f"left_command_reference_body_name resolved to {len(ids)} bodies."
+                )
+            self._left_ref_idx = ids[0]
+
+        if self._asset.is_fixed_base:
+            self._right_jacobi_body_idx = self._right_body_idx - 1
+            self._left_jacobi_body_idx = self._left_body_idx - 1
+            self._right_jacobi_joint_ids = self._right_joint_ids
+            self._left_jacobi_joint_ids = self._left_joint_ids
+        else:
+            self._right_jacobi_body_idx = self._right_body_idx
+            self._left_jacobi_body_idx = self._left_body_idx
+            self._right_jacobi_joint_ids = [i + 6 for i in self._right_joint_ids]
+            self._left_jacobi_joint_ids = [i + 6 for i in self._left_joint_ids]
+
+        self._right_offset_pos, self._right_offset_rot = self._build_diff_offset(
+            cfg.right_body_offset
+        )
+        self._left_offset_pos, self._left_offset_rot = self._build_diff_offset(
+            cfg.left_body_offset
+        )
+
+        self._right_diff_ik = DifferentialIKController(
+            cfg=cfg.diff_ik_controller, num_envs=self.num_envs, device=self.device
+        )
+        self._left_diff_ik = DifferentialIKController(
+            cfg=cfg.diff_ik_controller, num_envs=self.num_envs, device=self.device
+        )
+
+        if self._num_frame_tasks != 2:
+            raise ValueError(
+                "PinkDualDifferentialInverseKinematicsAction expects exactly 2 "
+                f"FrameTask entries in variable_input_tasks, got {self._num_frame_tasks}."
+            )
+
+    def _build_diff_offset(self, offset_cfg):
+        if offset_cfg is None:
+            return None, None
+        pos = torch.tensor(offset_cfg.pos, device=self.device).repeat(self.num_envs, 1)
+        rot = torch.tensor(offset_cfg.rot, device=self.device).repeat(self.num_envs, 1)
+        return pos, rot
+
+    # --- per-arm geometry helpers (mirror DualDifferentialInverseKinematicsAction)
+
+    def _arm_pose_in_ref(
+        self,
+        env_ids,
+        body_idx,
+        ref_idx,
+        offset_pos,
+        offset_rot,
+    ):
+        ee_pos_w = self._asset.data.body_pos_w[env_ids, body_idx]
+        ee_quat_w = self._asset.data.body_quat_w[env_ids, body_idx]
+        if ref_idx is not None:
+            ref_pos_w = self._asset.data.body_pos_w[env_ids, ref_idx]
+            ref_quat_w = self._asset.data.body_quat_w[env_ids, ref_idx]
+        else:
+            ref_pos_w = self._asset.data.root_pos_w[env_ids]
+            ref_quat_w = self._asset.data.root_quat_w[env_ids]
+        ee_pos_b, ee_quat_b = math_utils.subtract_frame_transforms(
+            ref_pos_w, ref_quat_w, ee_pos_w, ee_quat_w
+        )
+        if offset_pos is not None:
+            ee_pos_b, ee_quat_b = math_utils.combine_frame_transforms(
+                ee_pos_b, ee_quat_b, offset_pos[env_ids], offset_rot[env_ids]
+            )
+        return ee_pos_b, ee_quat_b
+
+    def _arm_jacobian(
+        self,
+        env_ids,
+        jacobi_body_idx,
+        jacobi_joint_ids,
+        ref_idx,
+        offset_pos,
+        offset_rot,
+    ):
+        jac_w = self._asset.root_physx_view.get_jacobians()[
+            :, jacobi_body_idx, :, jacobi_joint_ids
+        ][env_ids]
+        if ref_idx is not None:
+            ref_quat = self._asset.data.body_quat_w[env_ids, ref_idx]
+        else:
+            ref_quat = self._asset.data.root_quat_w[env_ids]
+        R = math_utils.matrix_from_quat(math_utils.quat_inv(ref_quat))
+        jac = jac_w.clone()
+        jac[:, 0:3, :] = torch.bmm(R, jac[:, 0:3, :])
+        jac[:, 3:, :] = torch.bmm(R, jac[:, 3:, :])
+        if offset_pos is not None:
+            skew = math_utils.skew_symmetric_matrix(offset_pos[env_ids])
+            jac[:, 0:3, :] += torch.bmm(-skew, jac[:, 3:, :])
+            jac[:, 3:, :] = torch.bmm(
+                math_utils.matrix_from_quat(offset_rot[env_ids]), jac[:, 3:, :]
+            )
+        return jac
+
+    def _transform_command_to_ref(self, command, env_ids, ref_idx):
+        """Convert an env-local world-frame pose command into the arm's ref frame."""
+        if ref_idx is not None:
+            base_pos_w = self._asset.data.body_pos_w[env_ids, ref_idx]
+            base_quat_w = self._asset.data.body_quat_w[env_ids, ref_idx]
+        else:
+            base_pos_w = self._asset.data.root_pos_w[env_ids]
+            base_quat_w = self._asset.data.root_quat_w[env_ids]
+        env_origins = getattr(self._env.scene, "env_origins", None)
+        positions_w = command[:, 0:3]
+        if env_origins is not None:
+            positions_w = positions_w + env_origins[env_ids]
+        if self.cfg.diff_ik_controller.command_type == "position":
+            pos_b, _ = math_utils.subtract_frame_transforms(
+                base_pos_w, base_quat_w, positions_w, None
+            )
+            return pos_b
+        quats_w = command[:, 3:7]
+        pos_b, quat_b = math_utils.subtract_frame_transforms(
+            base_pos_w, base_quat_w, positions_w, quats_w
+        )
+        return torch.cat((pos_b, quat_b), dim=1)
+
+    # --- overridden apply ---------------------------------------------------
+
+    def apply_actions(self) -> None:
+        self._step_count += 1
+
+        if len(self._env_ids) == 0:
+            return
+
+        is_pink_fire = (self._step_count - 1) % max(1, self.cfg.decimation) == 0
+
+        if is_pink_fire:
+            ik_joint_positions = self._compute_ik_solutions()
+            if self.hand_joint_dim > 0:
+                all_joint_positions = torch.cat(
+                    (ik_joint_positions, self._target_hand_joint_positions), dim=1
+                )
+            else:
+                all_joint_positions = ik_joint_positions
+            self._processed_actions[self._env_ids] = all_joint_positions
+
+            if self.cfg.enable_gravity_compensation:
+                self._apply_gravity_compensation()
+
+            self._asset.set_joint_position_target(
+                self._processed_actions[self._env_ids],
+                self._controlled_joint_ids,
+                self._env_ids,
+            )
+        else:
+            self._apply_dual_diff_ik()
+
+    def _apply_dual_diff_ik(self) -> None:
+        """Per-arm Jacobian tracking toward the stored Pink pose target."""
+        raw = self._raw_actions[self._env_ids]
+        right_cmd = raw[:, : self.pose_dim]
+        left_cmd = raw[:, self.pose_dim : 2 * self.pose_dim]
+
+        # Pink interprets actions in env-local world frame when relative_to_base is False.
+        if not self.cfg.relative_to_base:
+            right_cmd = self._transform_command_to_ref(
+                right_cmd, self._env_ids, self._right_ref_idx
+            )
+            left_cmd = self._transform_command_to_ref(
+                left_cmd, self._env_ids, self._left_ref_idx
+            )
+
+        r_pos, r_quat = self._arm_pose_in_ref(
+            self._env_ids,
+            self._right_body_idx,
+            self._right_ref_idx,
+            self._right_offset_pos,
+            self._right_offset_rot,
+        )
+        l_pos, l_quat = self._arm_pose_in_ref(
+            self._env_ids,
+            self._left_body_idx,
+            self._left_ref_idx,
+            self._left_offset_pos,
+            self._left_offset_rot,
+        )
+
+        self._right_diff_ik.set_command(right_cmd, r_pos, r_quat)
+        self._left_diff_ik.set_command(left_cmd, l_pos, l_quat)
+
+        r_q = self._asset.data.joint_pos[
+            self._env_ids.unsqueeze(1), self._right_joint_ids
+        ]
+        l_q = self._asset.data.joint_pos[
+            self._env_ids.unsqueeze(1), self._left_joint_ids
+        ]
+        r_jac = self._arm_jacobian(
+            self._env_ids,
+            self._right_jacobi_body_idx,
+            self._right_jacobi_joint_ids,
+            self._right_ref_idx,
+            self._right_offset_pos,
+            self._right_offset_rot,
+        )
+        l_jac = self._arm_jacobian(
+            self._env_ids,
+            self._left_jacobi_body_idx,
+            self._left_jacobi_joint_ids,
+            self._left_ref_idx,
+            self._left_offset_pos,
+            self._left_offset_rot,
+        )
+
+        r_q_des = self._right_diff_ik.compute(r_pos, r_quat, r_jac, r_q)
+        l_q_des = self._left_diff_ik.compute(l_pos, l_quat, l_jac, l_q)
+
+        r_lo = self._asset.data.joint_pos_limits[0, self._right_joint_ids, 0]
+        r_hi = self._asset.data.joint_pos_limits[0, self._right_joint_ids, 1]
+        l_lo = self._asset.data.joint_pos_limits[0, self._left_joint_ids, 0]
+        l_hi = self._asset.data.joint_pos_limits[0, self._left_joint_ids, 1]
+        r_q_des = torch.clamp(r_q_des, min=r_lo, max=r_hi)
+        l_q_des = torch.clamp(l_q_des, min=l_lo, max=l_hi)
+
+        if self.cfg.enable_gravity_compensation:
+            self._apply_gravity_compensation()
+
+        self._asset.set_joint_position_target(
+            r_q_des, self._right_joint_ids, self._env_ids
+        )
+        self._asset.set_joint_position_target(
+            l_q_des, self._left_joint_ids, self._env_ids
+        )

debug/536/n052_sawyer/pinklog_armlock_2026-11-09/pinkana.py

@@ -0,0 +1,56 @@
+import sys, re, csv, math
+log, csvf = sys.argv[1], sys.argv[2]
+t0, t1 = int(sys.argv[3]), int(sys.argv[4])
+step = int(sys.argv[5]) if len(sys.argv) > 5 else 1
+fl = lambda s: [float(x) for x in s.split(',')] if s.strip() else []
+ins = []; outs = {}
+for line in open(log):
+    if line.startswith('IN '):
+        m = re.match(r'IN (\d+) raw=\[(.*?)\] valid=\[(.*?)\] envs=\[(.*?)\]', line)
+        ins.append((int(m.group(1)), fl(m.group(2)), m.group(3)))
+    elif line.startswith('OUT '):
+        m = re.match(r'OUT (\d+) q=\[(.*?)\] sol=\[(.*?)\]', line)
+        outs[int(m.group(1))] = (fl(m.group(2)), fl(m.group(3)))
+rows = list(csv.DictReader(open(csvf)))
+J = ['right_j%d' % i for i in range(7)]
+def qcsv(r, p='q_'): return [float(r[p + j]) for j in J]
+# align: for each csv row find OUT index whose q matches best (search window)
+okeys = sorted(outs)
+def best(qc, lo, hi):
+    b = None
+    for k in range(max(1, lo), min(hi, okeys[-1]) + 1):
+        if k not in outs: continue
+        d = max(abs(a - c) for a, c in zip(outs[k][0], qc))
+        if b is None or d < b[1]: b = (k, d)
+    return b
+r0 = rows[0]
+b = best(qcsv(r0), 1, okeys[-1])
+off = b[0] - 4 * int(r0['tick'])
+print('align first csv tick', r0['tick'], '-> OUT', b, 'offset(OUT-4*tick)=', off)
+for r in rows[::200]:
+    t = int(r['tick']); k = 4 * t + off
+    bb = best(qcsv(r), k - 40, k + 40)
+    print('  check tick', t, 'pred OUT', k, 'best', bb)
+nan_ticks = [k for k, raw, e in ins if any(math.isnan(x) for x in raw)]
+print('NaN IN at OUT idx:', nan_ticks[:5], '...', nan_ticks[-5:], 'n=', len(nan_ticks))
+inmap = {k: (raw, e) for k, raw, e in ins}
+prev = None
+print('tick phase | IN raw pos (xyz) quat | dRaw_pos_mm | eef csv | base xy yaw | sol-q max(rad) | qtgt_csv - sol(last) max | per-joint sol delta vs prev tick')
+psol = None
+for r in rows:
+    t = int(r['tick'])
+    if t < t0 or t > t1 or (t - t0) % step: continue
+    k = 4 * t + off
+    kin = max([kk for kk in inmap if kk <= k + 4] or [0])
+    raw, e = inmap.get(kin, ([float('nan')] * 7, '?'))
+    d = '' if prev is None else '%.2f' % (1000 * math.dist(raw[:3], prev[:3])) if not any(math.isnan(x) for x in raw[:3] + prev[:3]) else 'nan'
+    prev = raw
+    sl = outs.get(k + 3) or outs.get(k)
+    sm = max(abs(a - c) for a, c in zip(sl[1], sl[0])) if sl else float('nan')
+    qt = qcsv(r, 'qtgt_')
+    dq = max(abs(a - c) for a, c in zip(qt, sl[1])) if sl else float('nan')
+    jd = '' if psol is None or not sl else ' '.join('%+.4f' % (a - c) for a, c in zip(sl[1], psol))
+    psol = sl[1] if sl else None
+    print('%d %-16s | %s %s | %s | eef %.3f %.3f %.3f | base %.3f %.3f %.3f | %.4f | %.4f | %s' % (
+        t, r['phase'][-16:], ' '.join('%.4f' % x for x in raw[:3]), ' '.join('%.3f' % x for x in raw[3:]), d,
+        float(r['eef_x']), float(r['eef_y']), float(r['eef_z']), float(r['base_x']), float(r['base_y']), float(r['base_yaw']), sm, dq, jd))

debug/536/n052_sawyer/pinklog_armlock_2026-11-09/pinkj.py

@@ -0,0 +1,22 @@
+import sys, re, csv
+log, csvf, t0, t1 = sys.argv[1], sys.argv[2], int(sys.argv[3]), int(sys.argv[4])
+off = -1907
+fl = lambda s: [float(x) for x in s.split(',')]
+outs = {}
+for line in open(log):
+    if line.startswith('OUT '):
+        m = re.match(r'OUT (\d+) q=\[(.*?)\] sol=\[(.*?)\]', line)
+        outs[int(m.group(1))] = (fl(m.group(2)), fl(m.group(3)))
+J = ['right_j%d' % i for i in range(7)]
+for r in csv.DictReader(open(csvf)):
+    t = int(r['tick'])
+    if t < t0 or t > t1: continue
+    k = 4 * t + off
+    q = [float(r['q_' + j]) for j in J]; qt = [float(r['qtgt_' + j]) for j in J]
+    print(t, r['phase'][-6:])
+    for kk in range(k, k + 4):
+        if kk in outs:
+            print('   OUT%d q  ' % kk, ' '.join('%+.4f' % x for x in outs[kk][0]))
+            print('   OUT%d sol' % kk, ' '.join('%+.4f' % x for x in outs[kk][1]))
+    print('   csv q    ', ' '.join('%+.4f' % x for x in q))
+    print('   csv qtgt ', ' '.join('%+.4f' % x for x in qt))

debug/536/n052_sawyer/pinklog_armlock_2026-11-09/pinklog_close_per_tick.txt

@@ -0,0 +1,64 @@
+align first csv tick 2838 -> OUT (9445, 4.484291076689573e-06) offset(OUT-4*tick)= -1907
+  check tick 2838 pred OUT 9445 best (9445, 4.484291076689573e-06)
+  check tick 3038 pred OUT 10245 best (10245, 4.548530578629695e-06)
+  check tick 3238 pred OUT 11045 best (11045, 4.772872924885263e-06)
+  check tick 3438 pred OUT 11845 best (11845, 4.700317382866359e-06)
+  check tick 3638 pred OUT 12645 best (12645, 4.18144226088657e-06)
+NaN IN at OUT idx: [5424, 5428, 5432, 5436, 5440] ... [14996, 15000, 15004, 15008, 15012] n= 81
+tick phase | IN raw pos (xyz) quat | dRaw_pos_mm | eef csv | base xy yaw | sol-q max(rad) | qtgt_csv - sol(last) max | per-joint sol delta vs prev tick
+2905 take_item:down:0 | -0.1119 1.1007 0.7490 -0.490 0.500 -0.503 -0.507 |  | eef -0.112 1.101 0.760 | base 0.091 2.079 -1.573 | 0.0098 | 0.0037 | 
+2906 take_item:down:0 | -0.1119 1.1006 0.7475 -0.491 0.500 -0.502 -0.506 | 1.49 | eef -0.112 1.101 0.758 | base 0.091 2.079 -1.573 | 0.0098 | 0.0037 | -0.0013 -0.0000 +0.0037 +0.0019 -0.0012 -0.0015 +0.0020
+2907 take_item:down:0 | -0.1119 1.1005 0.7460 -0.492 0.500 -0.502 -0.505 | 1.49 | eef -0.112 1.101 0.757 | base 0.091 2.079 -1.573 | 0.0098 | 0.0037 | -0.0013 -0.0000 +0.0037 +0.0019 -0.0012 -0.0015 +0.0019
+2908 take_item:down:0 | -0.1119 1.1005 0.7445 -0.493 0.500 -0.502 -0.505 | 1.49 | eef -0.112 1.101 0.755 | base 0.091 2.079 -1.573 | 0.0098 | 0.0037 | -0.0013 -0.0000 +0.0037 +0.0019 -0.0011 -0.0015 +0.0019
+2909 take_item:down:0 | -0.1120 1.1004 0.7430 -0.494 0.500 -0.502 -0.504 | 1.49 | eef -0.112 1.101 0.754 | base 0.091 2.079 -1.573 | 0.0098 | 0.0037 | -0.0013 +0.0000 +0.0037 +0.0019 -0.0011 -0.0015 +0.0019
+2910 take_item:down:0 | -0.1120 1.1003 0.7415 -0.495 0.500 -0.501 -0.503 | 1.49 | eef -0.112 1.101 0.752 | base 0.091 2.079 -1.573 | 0.0098 | 0.0037 | -0.0013 +0.0000 +0.0037 +0.0019 -0.0011 -0.0015 +0.0019
+2911 take_item:down:0 | -0.1120 1.1003 0.7401 -0.496 0.500 -0.501 -0.503 | 1.49 | eef -0.112 1.101 0.751 | base 0.091 2.079 -1.573 | 0.0098 | 0.0037 | -0.0013 +0.0000 +0.0037 +0.0019 -0.0011 -0.0015 +0.0019
+2912 take_item:down:0 | -0.1120 1.1002 0.7386 -0.497 0.500 -0.501 -0.502 | 1.49 | eef -0.112 1.101 0.749 | base 0.091 2.079 -1.573 | 0.0098 | 0.0037 | -0.0013 +0.0000 +0.0037 +0.0019 -0.0011 -0.0015 +0.0019
+2913 take_item:down:0 | -0.1121 1.1001 0.7371 -0.498 0.500 -0.501 -0.501 | 1.49 | eef -0.112 1.101 0.748 | base 0.091 2.079 -1.573 | 0.0098 | 0.0037 | -0.0014 +0.0000 +0.0037 +0.0019 -0.0011 -0.0015 +0.0019
+2914 take_item:down:0 | -0.1121 1.1001 0.7356 -0.499 0.500 -0.500 -0.501 | 1.49 | eef -0.112 1.101 0.746 | base 0.091 2.079 -1.573 | 0.0098 | 0.0037 | -0.0014 +0.0000 +0.0037 +0.0019 -0.0011 -0.0015 +0.0019
+2915 take_item:down:0 | -0.1121 1.1000 0.7341 -0.500 0.500 -0.500 -0.500 | 1.49 | eef -0.112 1.101 0.745 | base 0.091 2.079 -1.573 | 0.0098 | 0.0037 | -0.0014 +0.0000 +0.0037 +0.0019 -0.0011 -0.0015 +0.0019
+2916 take_item:down:0 | -0.1121 1.1000 0.7341 -0.500 0.500 -0.500 -0.500 | 0.00 | eef -0.112 1.100 0.743 | base 0.091 2.079 -1.573 | 0.0098 | 0.0037 | -0.0014 +0.0001 +0.0037 +0.0018 -0.0011 -0.0015 +0.0019
+2917 take_item:down:0 | nan nan nan nan nan nan nan | nan | eef -0.112 1.100 0.742 | base 0.091 2.079 -1.573 | 0.0098 | 0.0038 | -0.0014 -0.0009 +0.0038 +0.0011 -0.0022 -0.0009 +0.0018
+2918                  | nan nan nan nan nan nan nan | nan | eef -0.112 1.100 0.740 | base 0.091 2.079 -1.573 | 0.0098 | 0.0038 | -0.0015 -0.0010 +0.0039 +0.0010 -0.0024 -0.0008 +0.0018
+2919 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.112 1.100 0.739 | base 0.091 2.079 -1.573 | 0.0098 | 0.0039 | -0.0015 -0.0011 +0.0039 +0.0009 -0.0026 -0.0007 +0.0018
+2920 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.112 1.100 0.738 | base 0.091 2.079 -1.573 | 0.0098 | 0.0039 | -0.0015 -0.0012 +0.0039 +0.0008 -0.0027 -0.0006 +0.0018
+2921 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.112 1.100 0.736 | base 0.091 2.079 -1.573 | 0.0098 | 0.0039 | -0.0015 -0.0013 +0.0040 +0.0007 -0.0028 -0.0006 +0.0018
+2922 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.112 1.100 0.736 | base 0.091 2.079 -1.573 | 0.0098 | 0.0038 | -0.0010 -0.0009 +0.0012 +0.0014 -0.0038 +0.0017 +0.0015
+2923 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.113 1.103 0.734 | base 0.091 2.079 -1.573 | 0.0098 | 0.0034 | -0.0004 -0.0000 -0.0029 +0.0034 -0.0028 +0.0025 +0.0005
+2924 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.116 1.107 0.732 | base 0.091 2.079 -1.573 | 0.0098 | 0.0044 | -0.0007 +0.0006 -0.0044 +0.0036 -0.0015 +0.0024 -0.0001
+2925 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.119 1.112 0.730 | base 0.091 2.079 -1.573 | 0.0098 | 0.0054 | -0.0009 +0.0010 -0.0054 +0.0036 -0.0004 +0.0021 -0.0007
+2926 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.124 1.117 0.728 | base 0.091 2.079 -1.573 | 0.0150 | 0.0065 | -0.0004 +0.0003 -0.0065 +0.0064 +0.0011 +0.0001 -0.0021
+2927 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.129 1.123 0.726 | base 0.091 2.079 -1.573 | 0.0174 | 0.0211 | -0.0035 -0.0040 -0.0211 +0.0203 -0.0201 +0.0167 -0.0079
+2928 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.139 1.143 0.717 | base 0.091 2.079 -1.573 | 0.0184 | 0.0522 | -0.0086 +0.0040 -0.0384 +0.0400 -0.0522 +0.0357 -0.0114
+2929 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.155 1.176 0.703 | base 0.091 2.079 -1.573 | 0.0227 | 0.0624 | -0.0169 +0.0046 -0.0403 +0.0407 -0.0624 +0.0422 -0.0098
+2930 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.173 1.215 0.691 | base 0.091 2.079 -1.573 | 0.0227 | 0.0635 | -0.0223 +0.0040 -0.0405 +0.0407 -0.0635 +0.0414 -0.0064
+2931 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.192 1.257 0.683 | base 0.091 2.079 -1.573 | 0.0227 | 0.0640 | -0.0253 -0.0001 -0.0407 +0.0406 -0.0640 +0.0426 -0.0065
+2932 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.207 1.300 0.680 | base 0.091 2.079 -1.573 | 0.0227 | 0.0645 | -0.0263 -0.0044 -0.0408 +0.0403 -0.0645 +0.0412 -0.0065
+2933 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.220 1.345 0.680 | base 0.091 2.080 -1.573 | 0.0227 | 0.0653 | -0.0257 -0.0075 -0.0410 +0.0399 -0.0653 +0.0472 -0.0064
+2934 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.227 1.389 0.684 | base 0.091 2.080 -1.573 | 0.0227 | 0.0639 | -0.0220 -0.0076 -0.0408 +0.0395 -0.0639 +0.0472 -0.0065
+2935 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.228 1.432 0.689 | base 0.091 2.080 -1.573 | 0.0227 | 0.0629 | -0.0189 -0.0079 -0.0407 +0.0363 -0.0628 +0.0494 -0.0067
+2936 take_item:close  | nan nan nan nan nan nan nan | nan | eef -0.226 1.470 0.697 | base 0.091 2.080 -1.573 | 0.0227 | 0.0603 | -0.0148 -0.0084 -0.0404 +0.0314 -0.0603 +0.0432 -0.0067
+2937 take_item:close  | -0.2129 1.5288 0.7202 -0.841 0.052 -0.234 -0.485 | nan | eef -0.221 1.505 0.707 | base 0.091 2.080 -1.573 | 0.0227 | 0.0593 | -0.0111 -0.0083 -0.0402 +0.0245 -0.0593 +0.0528 -0.0070
+2938                  | -0.2117 1.5234 0.7219 -0.839 0.058 -0.238 -0.487 | 5.83 | eef -0.214 1.534 0.719 | base 0.091 2.080 -1.573 | 0.0174 | 0.0582 | -0.0069 +0.0007 -0.0401 +0.0175 -0.0582 +0.0543 +0.0166
+2939 take_item:lift   | -0.2104 1.5180 0.7236 -0.836 0.063 -0.242 -0.488 | 5.83 | eef -0.206 1.560 0.732 | base 0.091 2.080 -1.573 | 0.0227 | 0.0564 | -0.0041 -0.0050 -0.0399 +0.0137 -0.0564 +0.0492 -0.0207
+2940 take_item:lift   | -0.2091 1.5125 0.7253 -0.834 0.069 -0.246 -0.489 | 5.83 | eef -0.197 1.582 0.746 | base 0.091 2.080 -1.573 | 0.0227 | 0.0315 | -0.0041 -0.0027 -0.0315 +0.0127 -0.0190 +0.0127 -0.0080
+2941 take_item:lift   | -0.2078 1.5071 0.7270 -0.831 0.075 -0.250 -0.491 | 5.83 | eef -0.194 1.594 0.752 | base 0.091 2.080 -1.573 | 0.0227 | 0.0119 | -0.0066 -0.0084 -0.0045 -0.0019 +0.0083 -0.0118 -0.0087
+2942 take_item:lift   | -0.2066 1.5017 0.7287 -0.829 0.081 -0.254 -0.492 | 5.83 | eef -0.197 1.595 0.754 | base 0.091 2.080 -1.573 | 0.0227 | 0.0087 | -0.0022 -0.0020 +0.0000 -0.0046 +0.0066 -0.0068 -0.0087
+2943 take_item:lift   | -0.2053 1.4963 0.7304 -0.826 0.087 -0.258 -0.493 | 5.83 | eef -0.199 1.594 0.754 | base 0.091 2.080 -1.573 | 0.0227 | 0.0084 | +0.0016 +0.0028 -0.0045 -0.0023 -0.0024 +0.0065 -0.0084
+2944 take_item:lift   | -0.2040 1.4908 0.7321 -0.824 0.093 -0.262 -0.494 | 5.83 | eef -0.198 1.595 0.756 | base 0.091 2.080 -1.573 | 0.0227 | 0.0087 | +0.0018 +0.0012 -0.0041 +0.0014 +0.0042 -0.0054 -0.0087
+2945 take_item:lift   | -0.2027 1.4854 0.7338 -0.821 0.099 -0.266 -0.496 | 5.83 | eef -0.198 1.594 0.756 | base 0.091 2.080 -1.573 | 0.0227 | 0.0088 | +0.0015 +0.0006 +0.0013 -0.0019 +0.0063 -0.0063 -0.0088
+2946 take_item:lift   | -0.2015 1.4800 0.7355 -0.818 0.105 -0.270 -0.497 | 5.83 | eef -0.199 1.592 0.756 | base 0.091 2.080 -1.573 | 0.0227 | 0.0082 | +0.0039 +0.0016 -0.0081 +0.0014 -0.0045 +0.0072 -0.0082
+2947 take_item:lift   | -0.2002 1.4745 0.7372 -0.815 0.111 -0.274 -0.498 | 5.83 | eef -0.197 1.593 0.757 | base 0.091 2.080 -1.573 | 0.0227 | 0.0088 | +0.0019 -0.0016 +0.0003 -0.0006 +0.0058 -0.0060 -0.0088
+2948 take_item:lift   | -0.1989 1.4691 0.7389 -0.812 0.117 -0.278 -0.499 | 5.83 | eef -0.198 1.592 0.756 | base 0.091 2.080 -1.573 | 0.0227 | 0.0082 | +0.0037 +0.0009 -0.0066 +0.0000 -0.0044 +0.0067 -0.0082
+2949 take_item:lift   | -0.1976 1.4637 0.7406 -0.810 0.123 -0.282 -0.500 | 5.83 | eef -0.196 1.593 0.758 | base 0.091 2.080 -1.573 | 0.0227 | 0.0088 | +0.0020 -0.0019 -0.0012 +0.0002 +0.0052 -0.0058 -0.0088
+2950 take_item:lift   | -0.1964 1.4583 0.7423 -0.807 0.129 -0.286 -0.501 | 5.83 | eef -0.196 1.592 0.758 | base 0.091 2.080 -1.573 | 0.0227 | 0.0084 | +0.0023 -0.0021 +0.0006 -0.0030 -0.0006 +0.0039 -0.0084
+2951 take_item:lift   | -0.1951 1.4528 0.7440 -0.803 0.135 -0.290 -0.502 | 5.83 | eef -0.196 1.592 0.760 | base 0.091 2.080 -1.573 | 0.0227 | 0.0085 | +0.0036 +0.0006 -0.0078 +0.0023 +0.0013 -0.0034 -0.0085
+2952 take_item:lift   | -0.1938 1.4474 0.7457 -0.800 0.141 -0.294 -0.503 | 5.83 | eef -0.195 1.592 0.760 | base 0.091 2.080 -1.573 | 0.0227 | 0.0088 | +0.0019 -0.0028 +0.0011 -0.0015 +0.0061 -0.0062 -0.0088
+2953 take_item:lift   | -0.1925 1.4420 0.7474 -0.797 0.147 -0.298 -0.504 | 5.83 | eef -0.196 1.590 0.760 | base 0.091 2.080 -1.573 | 0.0227 | 0.0082 | +0.0038 +0.0003 -0.0068 -0.0003 -0.0052 +0.0060 -0.0082
+2954 take_item:lift   | -0.1912 1.4365 0.7491 -0.794 0.154 -0.302 -0.505 | 5.83 | eef -0.194 1.592 0.762 | base 0.091 2.080 -1.573 | 0.0224 | 0.0084 | +0.0020 -0.0027 -0.0014 +0.0001 +0.0052 -0.0058 -0.0084
+2955 take_item:lift   | -0.1900 1.4311 0.7507 -0.791 0.160 -0.305 -0.506 | 5.83 | eef -0.194 1.591 0.763 | base 0.091 2.080 -1.573 | 0.0214 | 0.0073 | +0.0023 -0.0027 +0.0005 -0.0031 -0.0010 +0.0034 -0.0073
+2956 take_item:lift   | -0.1887 1.4257 0.7524 -0.787 0.166 -0.309 -0.507 | 5.83 | eef -0.193 1.591 0.765 | base 0.091 2.080 -1.573 | 0.0216 | 0.0083 | +0.0035 +0.0003 -0.0082 +0.0020 +0.0010 -0.0036 -0.0083
+2957 take_item:lift   | -0.1874 1.4203 0.7541 -0.784 0.172 -0.313 -0.508 | 5.83 | eef -0.193 1.591 0.766 | base 0.091 2.080 -1.573 | 0.0206 | 0.0073 | +0.0018 -0.0033 +0.0009 -0.0016 +0.0061 -0.0062 -0.0073
+2958 take_item:lift   | -0.1861 1.4148 0.7558 -0.780 0.178 -0.317 -0.509 | 5.83 | eef -0.194 1.589 0.766 | base 0.091 2.080 -1.573 | 0.0208 | 0.0076 | +0.0037 +0.0003 -0.0070 -0.0007 -0.0056 +0.0055 -0.0076
+2959 take_item:lift   | -0.1849 1.4094 0.7575 -0.777 0.184 -0.321 -0.510 | 5.83 | eef -0.192 1.591 0.768 | base 0.091 2.080 -1.573 | 0.0202 | 0.0074 | +0.0019 -0.0029 -0.0016 -0.0002 +0.0052 -0.0058 -0.0074
+2960 take_item:lift   | -0.1836 1.4040 0.7592 -0.773 0.190 -0.325 -0.510 | 5.83 | eef -0.192 1.590 0.769 | base 0.091 2.080 -1.573 | 0.0192 | 0.0068 | +0.0019 -0.0032 +0.0019 -0.0025 +0.0064 -0.0064 -0.0068

debug/536/n052_sawyer/pinklog_armlock_2026-11-09/pinklog_per_physics_step_2926_2928.txt

@@ -0,0 +1,33 @@
+2926 :close
+   OUT9797 q   -0.2437 -0.6982 -0.5979 +1.2014 +0.4359 +1.2346 -0.6254
+   OUT9797 sol -0.2386 -0.6984 -0.5882 +1.1917 +0.4468 +1.2246 -0.6261
+   OUT9798 q   -0.2449 -0.6981 -0.5995 +1.2030 +0.4350 +1.2354 -0.6254
+   OUT9798 sol -0.2388 -0.6983 -0.5897 +1.1932 +0.4471 +1.2245 -0.6266
+   OUT9799 q   -0.2460 -0.6980 -0.6011 +1.2046 +0.4342 +1.2362 -0.6254
+   OUT9799 sol -0.2389 -0.6982 -0.5913 +1.1948 +0.4475 +1.2244 -0.6271
+   OUT9800 q   -0.2472 -0.6979 -0.6029 +1.2063 +0.4326 +1.2378 -0.6254
+   OUT9800 sol -0.2389 -0.6981 -0.5931 +1.1965 +0.4477 +1.2247 -0.6277
+   csv q     -0.2437 -0.6982 -0.5979 +1.2014 +0.4359 +1.2346 -0.6254
+   csv qtgt  -0.2385 -0.6984 -0.5867 +1.1901 +0.4465 +1.2246 -0.6256
+2927 :close
+   OUT9801 q   -0.2483 -0.6976 -0.6059 +1.2088 +0.4293 +1.2412 -0.6253
+   OUT9801 sol -0.2396 -0.6981 -0.5961 +1.1990 +0.4467 +1.2263 -0.6292
+   OUT9802 q   -0.2492 -0.6973 -0.6105 +1.2129 +0.4244 +1.2460 -0.6254
+   OUT9802 sol -0.2405 -0.6989 -0.6007 +1.2031 +0.4419 +1.2287 -0.6308
+   OUT9803 q   -0.2501 -0.6968 -0.6166 +1.2188 +0.4180 +1.2518 -0.6255
+   OUT9803 sol -0.2414 -0.7003 -0.6068 +1.2090 +0.4354 +1.2344 -0.6330
+   OUT9804 q   -0.2511 -0.6961 -0.6241 +1.2265 +0.4101 +1.2588 -0.6257
+   OUT9804 sol -0.2424 -0.7022 -0.6143 +1.2167 +0.4275 +1.2414 -0.6356
+   csv q     -0.2483 -0.6977 -0.6059 +1.2088 +0.4293 +1.2412 -0.6253
+   csv qtgt  -0.2390 -0.6981 -0.5932 +1.1965 +0.4477 +1.2247 -0.6277
+2928 :close
+   OUT9805 q   -0.2524 -0.6953 -0.6329 +1.2360 +0.4008 +1.2667 -0.6261
+   OUT9805 sol -0.2437 -0.7019 -0.6231 +1.2262 +0.4182 +1.2493 -0.6386
+   OUT9806 q   -0.2542 -0.6942 -0.6425 +1.2462 +0.3884 +1.2756 -0.6268
+   OUT9806 sol -0.2455 -0.7008 -0.6327 +1.2364 +0.4058 +1.2582 -0.6415
+   OUT9807 q   -0.2567 -0.6929 -0.6525 +1.2564 +0.3731 +1.2846 -0.6276
+   OUT9807 sol -0.2480 -0.6995 -0.6427 +1.2466 +0.3906 +1.2672 -0.6443
+   OUT9808 q   -0.2597 -0.6915 -0.6625 +1.2665 +0.3579 +1.2945 -0.6286
+   OUT9808 sol -0.2510 -0.6982 -0.6527 +1.2567 +0.3753 +1.2771 -0.6470
+   csv q     -0.2524 -0.6953 -0.6329 +1.2360 +0.4008 +1.2667 -0.6261
+   csv qtgt  -0.2424 -0.7021 -0.6143 +1.2167 +0.4275 +1.2414 -0.6356

debug/536/n052_sawyer/pinklog_armlock_2026-11-09/pinklog_s101_pink.log.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:543dafcbd6d2a96107494aaa6d2cf6f9f2cbd0e7baa8341f6e9df38014988944
+size 606493

debug/536/n052_sawyer/step_2026-11-09/README.md

@@ -0,0 +1,13 @@
+# N052 sawyer_umi on 536: step-by-step around take_item:close (seed 101, 2026-11-09 19:15 CST)
+
+By @nova (536 collector). Files in this folder: `closeup_s101_take_item_ticks.csv` (the per-tick recorder, both UMI fingers + gripper links), `closeup_s101_run.log`, `closeup_s101_grasp_slow4x.mp4` (fixed camera on the scale, 4x slow motion of the grasp), and `step_table_s101.txt` (the table below for every tick).
+
+## What the ticks show (rows = control ticks; qtgt = articulation joint position target, q = measured)
+- **Before the hand touches the die:** during the last ticks of `take_item:down` the pose target (`arm_tgt`) is constant, but the joint targets move every tick (right_j2 +0.004 rad/tick). The hand sinks about 1 mm/tick (eef z 0.742 -> 0.735).
+- **`take_item:close` (gripper only, no arm pose sent):** the joint targets keep moving the same way. From the first finger contact (tick 4 of close) the drift speeds up to about 0.04 rad/tick. qtgt stays **ahead** of q by 0.02-0.03 rad, so the controller is driving the arm, and the arm is not being pushed by the die. eef y moves 1.100 -> 1.304 m in 13 ticks, z sinks 4 cm.
+- **The grip:** the left pad is pushed back to its open stop (q 0.0208) and the right pad nearly closes (-0.0035). The die is then clamped between a rigid stop and the other pad, and the contact force rises to about 340-430 N per pad. That force is set by the arm, not by the 20 N gripper drive.
+- **Pink `nan_hold` (forced on through an env switch in a test copy):** the first close tick changes, then the same drift continues, so the moving targets do not come only from the NaN fallback.
+- **Joint limits:** every joint is far from its limit (j0 -0.23..-0.34, j1 -0.69..-0.72, j2 -0.59..-0.80, j3 1.17..1.40, j4 0.47..0.15, j5 1.22..1.43, j6 -0.62..-0.65), so the PINK_JOINT_MARGIN clamp is not the cause.
+
+## Suspect
+The arm's joint targets come from the IK (Pink) layer, or from something else writing them, and they keep changing with a fixed or absent pose target. The next check is to log, per tick, the pose that reaches `PinkInverseKinematicsAction.process_actions` (after `_handle_nan_actions`) and the joint targets `_compute_ik_solutions` returns. Note that an all-NaN arm row is skipped (`_env_ids` empty), so `apply_actions` returns early. In that case the targets should stay frozen and gravity compensation is not applied that tick, yet the recorded targets keep moving.

debug/536/n052_sawyer/step_2026-11-09/closeup_s101_grasp_slow4x.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e542e486edaff7b9559ed777448ce8116c05f87793b39d967ec526deac169237
+size 3158095

debug/536/n052_sawyer/step_2026-11-09/closeup_s101_run.log

@@ -0,0 +1,672 @@
+host gpu046 step_gpus=2 start 2026-11-09 19:06:06
+0, 2179 MiB, 46068 MiB
+1, 11632 MiB, 46068 MiB
+2, 14968 MiB, 46068 MiB
+[memory] DR spawn jitter: {'seed': 101, 'spawn_jitter': 0.3, 'start': [-3.4252, -2.0984, 0.2512], 'nominal_start': [-3.3, -2.3, 0.0]}
+Warp DeprecationWarning: The namespace `warp.torch` will soon be removed from the public API. It can still be accessed from `warp._src.torch` but might be changed or removed without notice.
+Warp DeprecationWarning: The namespace `warp.context` will soon be removed from the public API. It can still be accessed from `warp._src.context` but might be changed or removed without notice.
+Warp 1.12.1 initialized:
+   CUDA Toolkit 12.9, Driver 12.8
+   Devices:
+     "cpu"      : "x86_64"
+     "cuda:0"   : "NVIDIA L40S" (44 GiB, sm_89, mempool enabled)
+   Kernel cache:
+     /mnt/scratch/xc536_atomic/n052saw_closeup_s101/xdg/warp
+[WARN][AppLauncher]: There are no arguments attached to the ArgumentParser object. If you have your own arguments, please load your own arguments before calling the `AppLauncher.add_app_launcher_args` method. This allows the method to check the validity of the arguments and perform checks for argument names.
+[INFO][AppLauncher]: Using device: cuda:0
+[INFO][AppLauncher]: Loading experience file: /mnt/scratch/xc536_atomic/n052saw_closeup_s101/ptree/src/magicsim/Launch/magicsim.python.kit
+[Info] [carb] Logging to file: /home/luhr/tactilelab/MagicSim/.venv/lib/python3.11/site-packages/isaacsim/kit/logs/Kit/Isaac-Sim/5.1/kit_20261109_190620.log
+[1.041s] Failed to solve some dependencies locally, syncing with extension registry...
+[1.043s] [ext: omni.kit.async_engine-0.0.3] startup
+[1.058s] [ext: omni.client.lib-1.1.0] startup
+[1.193s] [ext: omni.kit.registry.nucleus-0.0.0] startup
+[6.204s] Pulling extension: `omni.kit.pip_archive-0.0.0` from the registry.
+[12.502s] Extension: `omni.kit.pip_archive-0.0.0` was installed.
+[12.503s] Pulling extension: `isaacsim.asset.importer.urdf-2.4.31` from the registry.
+[15.001s] Extension: `isaacsim.asset.importer.urdf-2.4.31` was installed.
+[15.001s] Pulling extension: `omni.flowusd.ui-107.1.2` from the registry.
+[15.300s] Extension: `omni.flowusd.ui-107.1.2` was installed.
+[15.301s] Pulling extension: `omni.flowusd.bundle-1.1.2` from the registry.
+[15.689s] Extension: `omni.flowusd.bundle-1.1.2` was installed.
+[15.709s] [ext: omni.metrics.core-0.0.3] startup
+[15.714s] [ext: omni.blobkey-1.1.2] startup
+[15.717s] [ext: omni.stats-1.0.1] startup
+[17.463s] [ext: omni.datastore-0.0.0] startup
+[17.520s] [ext: omni.client-1.3.0] startup
+[17.524s] [ext: omni.ujitso.default-1.0.0] startup
+[17.533s] [ext: omni.hsscclient-1.1.2] startup
+[17.606s] [ext: omni.gpu_foundation.shadercache.vulkan-1.0.0] startup
+[17.651s] [ext: omni.assets.plugins-0.0.0] startup
+[17.665s] [ext: omni.gpu_foundation-0.0.0] startup
+[17.800s] [ext: carb.windowing.plugins-1.0.0] startup
+2026-11-10T00:06:38Z [17,458ms] [Warning] [carb.windowing-glfw.plugin] GLFW initialization failed.
+2026-11-10T00:06:38Z [17,458ms] [Warning] [carb] Failed to startup plugin carb.windowing-glfw.plugin (interfaces: [carb::windowing::IGLContext v1.0],[carb::windowing::IWindowing v1.5]) (impl: carb.windowing-glfw.plugin)
+[17.808s] [ext: omni.kit.renderer.init-0.0.0] startup
+2026-11-10T00:06:38Z [17,500ms] [Warning] [omni.platforminfo.plugin] failed to open the default display.  Can't verify X Server version.
+2026-11-10T00:06:38Z [17,510ms] [Warning] [carb.cudainterop.plugin] CUDA_VISIBLE_DEVICES environment variable is set.
+2026-11-10T00:06:38Z [17,510ms] [Warning] [carb.cudainterop.plugin] Note CUDA device enumeration and Omniverse device enumeration are different.
+2026-11-10T00:06:38Z [17,510ms] [Warning] [carb.cudainterop.plugin] Setting CUDA_VISIBLE_DEVICES can lead to undesired behavior or crashes.
+
+|---------------------------------------------------------------------------------------------|
+| Driver Version: 570.148.08    | Graphics API: Vulkan
+|=============================================================================================|
+| GPU | Name                             | Active | LDA | GPU Memory | Vendor-ID | LUID       |
+|     |                                  |        |     |            | Device-ID | UUID       |
+|     |                                  |        |     |            | Bus-ID    |            |
+| 0   | NVIDIA L40S                      | Yes: 0 |     | 46068   MB | 10de      | 0          |
+|     |                                  |        |     |            | 26b9      | 62e0bf1f.. |
+|     |                                  |        |     |            | ab        |            |
+| OS: 24.04.2 LTS (Noble Numbat) ubuntu, Version: 24.04.2, Kernel: 5.14.0-570.134.1.el9_6.x86_64
+| Processor: INTEL(R) XEON(R) GOLD 6526Y
+| Cores: 32 | Logical Cores: 32
+| Total Memory (MB): 515701 | Free Memory: 341338
+| Total Page/Swap (MB): 0 | Free Page/Swap: 0
+2026-11-10T00:06:39Z [19,157ms] [Warning] [gpu.foundation.plugin] ECC is enabled on physical device 0
+[20.141s] [ext: omni.kit.pipapi-0.0.0] startup
+[20.189s] [ext: omni.materialx.libs-1.0.7] startup
+[20.248s] [ext: omni.kit.pip_archive-0.0.0] startup
+[20.249s] [ext: omni.usd.config-1.0.6] startup
+[20.281s] [ext: omni.gpucompute.plugins-0.0.0] startup
+[20.286s] [ext: omni.usd.libs-1.0.1] startup
+[20.735s] [ext: omni.mdl-56.0.3] startup
+[20.894s] [ext: omni.iray.libs-0.0.0] startup
+[21.108s] [ext: omni.mdl.neuraylib-0.2.12] startup
+[21.123s] [ext: omni.kit.usd.mdl-1.1.5] startup
+[21.427s] [ext: omni.kit.loop-isaac-1.3.7] startup
+[21.518s] [ext: omni.kit.test-2.0.1] startup
+[21.898s] [ext: omni.services.pip_archive-0.16.0] startup
+[21.900s] [ext: omni.kit.telemetry-0.5.2] startup
+[22.034s] [ext: omni.appwindow-1.1.10] startup
+2026-11-10T00:06:42Z [21,701ms] [Warning] [carb.windowing-glfw.plugin] GLFW initialization failed.
+2026-11-10T00:06:42Z [21,704ms] [Warning] [carb] Failed to startup plugin carb.windowing-glfw.plugin (interfaces: [carb::windowing::IGLContext v1.0],[carb::windowing::IWindowing v1.5]) (impl: carb.windowing-glfw.plugin)
+[22.070s] [ext: omni.kit.renderer.core-1.1.0] startup
+[22.269s] [ext: omni.kit.renderer.capture-0.0.0] startup
+[22.285s] [ext: omni.kit.renderer.imgui-2.0.5] startup
+[23.007s] [ext: omni.ui-2.27.1] startup
+[23.054s] [ext: omni.kit.mainwindow-1.0.3] startup
+[23.063s] [ext: carb.audio-0.1.0] startup
+[23.081s] [ext: omni.uiaudio-1.0.0] startup
+[23.086s] [ext: omni.kit.uiapp-0.0.0] startup
+[23.087s] [ext: omni.usd.schema.metrics.assembler-107.3.1] startup
+[23.108s] [ext: omni.usd.schema.audio-0.0.0] startup
+[23.420s] [ext: omni.usd.schema.omnigraph-1.0.0] startup
+[23.461s] [ext: omni.usd_resolver-1.0.0] startup
+[23.475s] [ext: omni.usd.core-1.5.3] startup
+[23.489s] [ext: omni.usd.schema.render_settings.rtx-0.0.0] startup
+[23.554s] [ext: omni.usd.schema.omni_lens_distortion-0.0.0] startup
+[23.556s] [ext: omni.usd.schema.omniscripting-1.0.0] startup
+[23.567s] [ext: omni.usd.schema.anim-0.0.0] startup
+[23.630s] [ext: omni.usd.schema.physx-107.3.26] startup
+[23.711s] [ext: isaacsim.robot.schema-3.6.0] startup
+[23.733s] [ext: omni.anim.navigation.schema-107.3.3] startup
+[23.743s] [ext: omni.usd.schema.scene.visualization-2.0.2] startup
+[23.746s] [ext: omni.anim.behavior.schema-107.3.0] startup
+[23.756s] [ext: omni.usd.schema.semantics-0.0.0] startup
+[23.764s] [ext: omni.usd.schema.geospatial-0.0.0] startup
+[23.771s] [ext: omni.usd.schema.omni_sensors-0.0.0] startup
+[23.773s] [ext: omni.anim.graph.schema-107.3.3] startup
+[23.789s] [ext: omni.activity.core-1.0.3] startup
+[23.795s] [ext: omni.graph.exec-0.9.6] startup
+[23.797s] [ext: omni.kit.actions.core-1.0.0] startup
+[23.807s] [ext: omni.kit.exec.core-0.13.4] startup
+[23.815s] [ext: omni.kit.commands-1.4.10] startup
+[23.835s] [ext: usdrt.scenegraph-7.6.1] startup
+[23.963s] [ext: omni.resourcemonitor-107.0.1] startup
+[23.973s] [ext: omni.kit.audiodeviceenum-1.0.2] startup
+[23.980s] [ext: omni.hydra.usdrt_delegate-7.5.1] startup
+[24.011s] [ext: omni.kit.window.popup_dialog-2.0.24] startup
+[24.033s] [ext: omni.timeline-1.0.14] startup
+[24.042s] [ext: omni.kit.widget.nucleus_connector-2.0.1] startup
+[24.054s] [ext: omni.kit.usd_undo-0.1.8] startup
+[24.058s] [ext: omni.hydra.scene_delegate-0.3.4] startup
+[24.072s] [ext: omni.usd-1.13.10] startup
+[24.172s] [ext: omni.kit.asset_converter-5.0.17] startup
+[24.523s] [ext: omni.usd.schema.flow-107.1.1] startup
+[24.554s] [ext: omni.index.libs-380600.8087.0] startup
+[24.554s] [ext: omni.kit.clipboard-1.0.5] startup
+[24.572s] [ext: omni.index-1.0.1] startup
+[24.586s] [ext: omni.kit.helper.file_utils-0.1.9] startup
+[24.631s] [ext: omni.kit.menu.core-1.1.2] startup
+[24.668s] [ext: omni.kit.widget.path_field-2.0.11] startup
+[24.706s] [ext: omni.kit.widget.options_menu-1.1.6] startup
+[24.741s] [ext: omni.kit.widget.filebrowser-2.12.3] startup
+[24.781s] [ext: omni.kit.widget.context_menu-1.2.5] startup
+[24.789s] [ext: omni.kit.widget.browser_bar-2.0.10] startup
+[24.794s] [ext: omni.ui.scene-1.11.5] startup
+[24.815s] [ext: omni.kit.widget.options_button-1.0.3] startup
+[24.820s] [ext: omni.kit.notification_manager-1.0.10] startup
+[24.830s] [ext: omni.hydra.rtx.shadercache.vulkan-1.0.0] startup
+[25.024s] [ext: omni.kit.window.filepicker-2.13.4] startup
+[25.126s] [ext: omni.volume-0.5.2] startup
+[25.155s] [ext: omni.ujitso.client-0.0.0] startup
+[25.157s] [ext: omni.kit.usd.layers-2.2.11] startup
+[25.208s] [ext: omni.hydra.rtx-1.0.0] startup
+2026-11-10T00:06:45Z [25,104ms] [Warning] [omni.log] Source: omni.hydra was already registered.
+[25.614s] [ext: omni.kit.window.file_importer-1.1.18] startup
+[25.623s] [ext: omni.kit.menu.utils-2.0.5] startup
+[25.668s] [ext: omni.kit.context_menu-1.8.6] startup
+[25.682s] [ext: omni.kit.raycast.query-1.1.0] startup
+[25.699s] [ext: omni.kit.material.library-2.0.7] startup
+[25.739s] [ext: omni.hydra.engine.stats-1.0.3] startup
+[25.751s] [ext: omni.kit.viewport.scene_camera_model-1.0.6] startup
+[25.765s] [ext: omni.kit.hydra_texture-1.4.6] startup
+[25.778s] [ext: omni.kit.widget.searchable_combobox-1.0.6] startup
+[25.786s] [ext: omni.kit.window.drop_support-1.0.5] startup
+[25.789s] [ext: omni.kit.widget.viewport-107.1.3] startup
+[25.816s] [ext: omni.kit.viewport.registry-104.0.6] startup
+[25.819s] [ext: omni.kit.viewport.legacy_gizmos-1.0.19] startup
+[25.826s] [ext: omni.kit.widget.settings-1.2.6] startup
+[25.843s] [ext: omni.kit.viewport.window-107.2.0] startup
+[25.915s] [ext: omni.kit.window.preferences-1.8.0] startup
+[25.956s] [ext: omni.kit.widget.toolbar-2.0.1] startup
+[25.985s] [ext: omni.kit.viewport.utility-1.1.2] startup
+[25.990s] [ext: omni.kit.manipulator.transform-107.0.0] startup
+[26.024s] [ext: omni.kit.manipulator.tool.snap-1.5.13] startup
+[26.046s] [ext: omni.kit.manipulator.selector-1.1.3] startup
+[26.051s] [ext: omni.kit.viewport.manipulator.transform-107.0.4] startup
+[26.062s] [ext: omni.kit.manipulator.viewport-107.0.1] startup
+[26.067s] [ext: omni.kit.widget.prompt-1.0.7] startup
+[26.072s] [ext: omni.inspect-1.0.2] startup
+[26.079s] [ext: omni.kit.manipulator.prim.core-107.0.8] startup
+[26.113s] [ext: omni.kit.primitive.mesh-1.0.17] startup
+[26.134s] [ext: omni.kit.window.file_exporter-1.0.33] startup
+[26.139s] [ext: omni.fabric.commands-1.1.6] startup
+[26.158s] [ext: omni.rtx.window.settings-0.6.19] startup
+[26.176s] [ext: omni.ocio-0.1.1] startup
+[26.180s] [ext: omni.graph.core-2.184.5] startup
+[26.201s] [ext: omni.kit.manipulator.prim.usd-107.0.3] startup
+[26.207s] [ext: omni.kit.manipulator.prim.fabric-107.0.4] startup
+[26.214s] [ext: omni.rtx.settings.core-0.6.5] startup
+[26.236s] [ext: isaacsim.core.deprecation_manager-0.2.7] startup
+[26.240s] [ext: omni.kit.manipulator.prim-107.0.0] startup
+[26.240s] [ext: omni.debugdraw-0.1.4] startup
+[26.250s] [ext: omni.physics-107.3.26] startup
+[26.263s] [ext: isaacsim.core.version-2.0.6] startup
+[26.266s] [ext: omni.kvdb-107.3.26] startup
+[26.275s] [ext: omni.usdphysics-107.3.26] startup
+[26.283s] [ext: omni.physx.foundation-107.3.26] startup
+[26.298s] [ext: omni.localcache-107.3.26] startup
+[26.305s] [ext: omni.convexdecomposition-107.3.26] startup
+[26.317s] [ext: omni.kit.property.adapter.core-1.0.2] startup
+[26.328s] [ext: omni.physx.cooking-107.3.26] startup
+[26.345s] [ext: omni.kit.property.adapter.usd-1.0.2] startup
+[26.352s] [ext: omni.kit.property.adapter.fabric-1.0.3] startup
+[26.361s] [ext: omni.physx-107.3.26] startup
+[26.416s] [ext: omni.kit.stage_template.core-1.1.22] startup
+[26.422s] [ext: omni.kit.widget.filter-1.1.4] startup
+[26.427s] [ext: omni.kit.widget.searchfield-1.1.8] startup
+[26.436s] [ext: omni.kit.hotkeys.core-1.3.10] startup
+[26.440s] [ext: omni.kit.window.file-2.0.5] startup
+[26.461s] [ext: omni.kit.window.content_browser_registry-0.0.6] startup
+[26.466s] [ext: omni.kit.widget.stage-3.1.4] startup
+[26.534s] [ext: omni.kit.widget.highlight_label-1.0.3] startup
+[26.539s] [ext: omni.kit.window.content_browser-3.1.3] startup
+[26.566s] [ext: omni.kit.viewport.menubar.core-107.2.1] startup
+[26.632s] [ext: omni.kit.viewport.actions-107.0.2] startup
+[26.645s] [ext: omni.kit.window.property-1.12.1] startup
+[26.661s] [ext: omni.kit.widget.layers-1.8.6] startup
+[26.728s] [ext: omni.kit.viewport.menubar.display-107.0.3] startup
+[26.738s] [ext: omni.kit.property.usd-4.5.12] startup
+[26.842s] [ext: omni.kit.manipulator.selection-106.0.1] startup
+[26.865s] [ext: omni.physx.commands-107.3.26] startup
+[26.925s] [ext: omni.usdphysics.ui-107.3.26] startup
+[27.035s] [ext: omni.physics.stageupdate-107.3.26] startup
+[27.043s] [ext: isaacsim.storage.native-1.5.1] startup
+[27.054s] [ext: omni.usd.metrics.assembler-107.3.1] startup
+[27.191s] [ext: omni.physx.ui-107.3.26] startup
+[27.375s] [ext: omni.physics.physx-107.3.26] startup
+2026-11-10T00:06:47Z [27,029ms] [Warning] [carb] Acquiring non optional plugin interface which is not listed as dependency: [omni::physx::IPhysxBenchmarks v1.0] (plugin: <default plugin>), by client: omni.physics.physx.plugin. Add it to CARB_PLUGIN_IMPL_DEPS() macro of a client.
+[27.380s] [ext: omni.kit.numpy.common-0.1.3] startup
+[27.386s] [ext: omni.kit.stagerecorder.core-107.0.3] startup
+[27.397s] [ext: omni.usd.metrics.assembler.physics-107.3.26] startup
+[27.412s] [ext: omni.warp.core-1.8.2] startup
+[27.421s] [ext: omni.isaac.dynamic_control-2.0.7] startup
+2026-11-10T00:06:47Z [27,092ms] [Warning] [omni.isaac.dynamic_control] omni.isaac.dynamic_control is deprecated as of Isaac Sim 4.5. No action is needed from end-users.
+[27.444s] [ext: semantics.schema.property-2.0.1] startup
+2026-11-10T00:06:47Z [27,108ms] [Warning] [pxr.Semantics] pxr.Semantics is deprecated - please use Semantics instead
+[27.465s] [ext: omni.kit.stagerecorder.ui-107.0.1] startup
+[27.485s] [ext: omni.physics.tensors-107.3.26] startup
+[27.511s] [ext: omni.kit.stagerecorder.bundle-105.0.2] startup
+[27.511s] [ext: isaacsim.core.utils-3.5.1] startup
+[27.530s] [ext: omni.kit.viewport.menubar.camera-107.0.6] startup
+[27.559s] [ext: isaacsim.app.about-2.0.11] startup
+[27.564s] [ext: omni.physx.tensors-107.3.26] startup
+[27.578s] [ext: omni.kit.manipulator.camera-106.0.4] startup
+[27.606s] [ext: omni.kit.widget.text_editor-1.1.1] startup
+[27.614s] [ext: omni.graph.tools-1.79.2] startup
+[27.782s] [ext: omni.kit.widget.stage_icons-1.0.8] startup
+[27.789s] [ext: omni.ui_query-1.1.8] startup
+[27.794s] [ext: omni.kit.window.extensions-1.4.27] startup
+[27.840s] [ext: omni.graph-1.141.2] startup
+[28.044s] [ext: omni.kit.window.stage-2.6.1] startup
+[28.119s] [ext: omni.graph.ui-1.101.6] startup
+[28.234s] [ext: omni.graph.action_core-1.1.7] startup
+[28.244s] [ext: omni.kit.widget.zoombar-1.0.6] startup
+[28.251s] [ext: omni.kit.menu.stage-1.2.7] startup
+[28.256s] [ext: omni.graph.action_nodes-1.50.4] startup
+[28.273s] [ext: omni.kit.browser.core-2.3.13] startup
+[29.001s] [ext: omni.kit.usd.collect-2.4.5] startup
+[29.024s] [ext: omni.kit.ui_test-1.3.7] startup
+[29.041s] [ext: omni.graph.action-1.130.0] startup
+[29.045s] [ext: omni.kit.browser.folder.core-1.10.9] startup
+[29.084s] [ext: omni.kit.usdz_export-1.0.9] startup
+[29.094s] [ext: omni.graph.visualization.nodes-2.1.3] startup
+[29.117s] [ext: omni.kit.tool.collect-2.2.18] startup
+[29.188s] [ext: omni.kit.tool.asset_importer-4.3.2] startup
+[29.244s] [ext: isaacsim.gui.components-1.2.1] startup
+[29.282s] [ext: isaacsim.examples.browser-0.2.1] startup
+[29.303s] [ext: isaacsim.asset.importer.urdf-2.4.31] startup
+[29.470s] [ext: isaacsim.core.simulation_manager-1.4.4] startup
+[29.685s] [ext: semantics.schema.editor-2.0.2] startup
+[29.702s] [ext: omni.kit.viewport.menubar.render-107.0.10] startup
+[29.718s] [ext: omni.kit.viewport.menubar.settings-107.0.3] startup
+[29.742s] [ext: isaacsim.core.cloner-1.4.10] startup
+[29.752s] [ext: omni.graph.image.core-0.6.1] startup
+[29.757s] [ext: omni.kit.ui.actions-1.0.5] startup
+[29.764s] [ext: omni.graph.image.nodes-1.3.1] startup
+[29.771s] [ext: omni.kit.widget.graph-2.0.0] startup
+[29.820s] [ext: omni.syntheticdata-0.6.13] startup
+[29.884s] [ext: omni.graph.scriptnode-1.50.0] startup
+[29.902s] [ext: isaacsim.core.prims-0.6.1] startup
+[30.341s] [ext: omni.kit.graph.delegate.default-1.2.3] startup
+[30.349s] [ext: omni.graph.ui_nodes-1.50.5] startup
+[30.376s] [ext: omni.graph.nodes-1.170.10] startup
+[30.828s] [ext: omni.kit.browser.sample-1.4.9] startup
+[30.846s] [ext: omni.kit.widget.material_preview-1.0.16] startup
+[30.866s] [ext: omni.kit.graph.usd.commands-1.3.1] startup
+[30.877s] [ext: omni.kit.graph.editor.core-1.5.3] startup
+[30.908s] [ext: omni.kit.stage_templates-2.0.0] startup
+[30.945s] [ext: omni.warp-1.8.2] startup
+[30.969s] [ext: omni.kit.window.material_graph-1.9.1] startup
+[31.083s] [ext: omni.videoencoding-0.1.2] startup
+[31.110s] [ext: isaacsim.core.experimental.utils-0.3.0] startup
+[31.129s] [ext: isaacsim.test.docstring-1.1.0] startup
+[31.210s] [ext: omni.replicator.core-1.12.27] startup
+2026-11-10T00:06:52Z [31,377ms] [Warning] [omni.graph.core.plugin] Found duplicate of category 'Replicator' - was 'Annotators', adding 'Fabric Reader'
+2026-11-10T00:06:52Z [31,377ms] [Warning] [omni.graph.core.plugin] Category 'Replicator' not accepted on node type 'omni.replicator.core.FabricReader' in extension 'omni.replicator.core'
+2026-11-10T00:06:52Z [31,381ms] [Warning] [omni.replicator.core.scripts.extension] No material configuration file, adding configuration to material settings directly.
+[31.734s] [ext: isaacsim.core.api-4.8.0] startup
+[32.080s] [ext: isaacsim.core.experimental.prims-0.8.1] startup
+[32.385s] [ext: isaacsim.core.nodes-3.4.3] startup
+[32.425s] [ext: isaacsim.robot.surface_gripper-3.3.1] startup
+[32.549s] [ext: omni.kit.widget.imageview-1.0.4] startup
+[32.668s] [ext: isaacsim.robot.manipulators-3.3.6] startup
+[32.893s] [ext: omni.kit.window.quicksearch-2.4.4] startup
+[33.190s] [ext: omni.hydra.scene_api-0.1.2] startup
+[33.237s] [ext: omni.kit.property.geometry-2.0.4] startup
+[33.295s] [ext: omni.kit.property.camera-1.0.10] startup
+[33.332s] [ext: omni.kit.property.material-1.11.9] startup
+[33.541s] [ext: omni.kit.property.transform-1.5.13] startup
+[33.595s] [ext: omni.kit.property.render-1.2.1] startup
+[33.606s] [ext: omni.kit.property.light-1.0.12] startup
+[33.713s] [ext: omni.kit.property.audio-1.0.16] startup
+[33.749s] [ext: isaacsim.sensors.physics-0.4.3] startup
+[33.794s] [ext: omni.kit.property.bundle-1.4.1] startup
+[33.915s] [ext: isaacsim.robot.policy.examples-4.1.11] startup
+[34.005s] [ext: isaacsim.robot.wheeled_robots-4.0.24] startup
+[34.172s] [ext: omni.kit.viewport.menubar.lighting-107.3.1] startup
+[34.365s] [ext: omni.kit.window.console-1.1.4] startup
+[34.424s] [ext: omni.anim.asset-107.3.0] startup
+[34.448s] [ext: omni.usd.metrics.assembler.ui-107.3.1] startup
+[34.497s] [ext: omni.kit.menu.create-2.0.1] startup
+[34.503s] [ext: isaacsim.core.throttling-2.2.2] startup
+[34.512s] [ext: omni.kit.window.script_editor-2.0.1] startup
+[34.530s] [ext: isaacsim.gui.property-1.1.3] startup
+[34.542s] [ext: omni.curve.creator-107.0.1] startup
+[34.550s] [ext: omni.scene.visualization.core-107.0.2] startup
+[34.567s] [ext: omni.anim.retarget.core-107.3.3] startup
+[34.594s] [ext: omni.kit.scripting-107.3.2] startup
+[34.619s] [ext: omni.kit.quicklayout-1.0.8] startup
+[34.625s] [ext: omni.anim.navigation.core-107.3.8] startup
+[34.654s] [ext: omni.curve.manipulator-107.0.4] startup
+[34.726s] [ext: omni.anim.graph.core-107.3.4] startup
+[34.766s] [ext: omni.metropolis.utils-0.1.20] startup
+[34.853s] [ext: omni.anim.navigation.ui-107.3.6] startup
+2026-11-10T00:06:55Z [34,711ms] [Warning] [omni.graph.core._impl._registration.register_python_ogn] Python import process in omni.anim.graph.core failed - 'NoneType' object is not subscriptable. Aborting Python node registration
+[35.112s] [ext: omni.flowusd-107.1.8] startup
+2026-11-10T00:06:55Z [34,784ms] [Warning] [omni.stageupdate.plugin] Deprecated: direct use of IStageUpdate callbacks is deprecated. Use IStageUpdate::getStageUpdate instead.
+[35.148s] [ext: omni.sensors.nv.common-3.0.0] startup
+[35.326s] [ext: omni.anim.people-0.7.9] startup
+[35.346s] [ext: omni.anim.navigation.bundle-107.3.3] startup
+[35.355s] [ext: omni.sensors.net-1.0.0] startup
+[35.472s] [ext: omni.sensors.nv.materials-2.0.0] startup
+[35.540s] [ext: isaacsim.anim.robot-0.0.15] startup
+[35.548s] [ext: isaacsim.util.debug_draw-3.1.0] startup
+[35.566s] [ext: isaacsim.replicator.incident-0.1.28] startup
+[35.657s] [ext: omni.kit.mesh.raycast-107.0.1] startup
+[35.674s] [ext: omni.sensors.nv.wpm-3.0.0] startup
+[35.692s] [ext: omni.sensors.nv.ids-2.0.0] startup
+[35.703s] [ext: isaacsim.replicator.behavior-1.1.16] startup
+[35.709s] [ext: isaacsim.sensors.physx-2.3.2] startup
+[35.734s] [ext: omni.kit.window.status_bar-0.1.9] startup
+[35.744s] [ext: omni.sensors.nv.lidar-3.0.0] startup
+[35.809s] [ext: omni.sensors.nv.radar-3.0.0] startup
+[35.834s] [ext: omni.kit.property.physx-107.3.26] startup
+[35.903s] [ext: omni.physx.demos-107.3.26] startup
+[36.066s] [ext: isaacsim.sensors.camera-1.3.6] startup
+[36.097s] [ext: omni.kit.selection-0.1.6] startup
+[36.112s] [ext: isaacsim.sensors.rtx-15.8.4] startup
+[36.142s] [ext: omni.asset_validator.core-1.1.6] startup
+[36.329s] [ext: omni.physx.vehicle-107.3.26] startup
+[36.447s] [ext: isaacsim.asset.browser-1.3.23] startup
+/home/luhr/tactilelab/MagicSim/.venv/lib/python3.11/site-packages/isaacsim/extscache/omni.services.pip_archive-0.16.0+107.0.3.lx64.cp311/pip_prebundle/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.5.0) or chardet (7.6.0)/charset_normalizer (3.3.2) doesn't match a supported version!
+  warnings.warn(
+[36.695s] [ext: isaacsim.asset.importer.mjcf-2.5.13] startup
+[36.730s] [ext: isaacsim.gui.menu-2.4.4] startup
+[36.765s] [ext: isaacsim.replicator.agent.core-0.7.28] startup
+[36.811s] [ext: omni.physx.asset_validator-107.3.26] startup
+[36.844s] [ext: omni.physx.camera-107.3.26] startup
+[36.871s] [ext: omni.physx.cct-107.3.26] startup
+[36.907s] [ext: omni.physx.graph-107.3.26] startup
+[37.128s] [ext: omni.kit.graph.widget.variables-2.1.0] startup
+[37.579s] [ext: omni.ramp-107.0.1] startup
+[37.712s] [ext: omni.kit.graph.delegate.modern-1.10.9] startup
+[37.877s] [ext: omni.graph.bundle.action-2.30.0] startup
+[37.881s] [ext: omni.graph.io-1.30.1] startup
+2026-11-10T00:06:58Z [37,537ms] [Warning] [omni.graph.io._impl.extension] omni.graph.io is deprecated. Nodes in this extension have been moved to omni.graph.nodes. Update any dependencies accordingly.
+[37.887s] [ext: omni.kit.window.toolbar-2.0.0] startup
+[37.894s] [ext: isaacsim.simulation_app-2.12.2] startup
+[37.896s] [ext: omni.genproc.core-107.0.3] startup
+[37.926s] [ext: omni.flowusd.ui-107.1.2] startup
+2026-11-10T00:06:58Z [37,631ms] [Warning] [omni.flowusd.ui.scripts.widgets] Flow schema not found. Flow properties not ordered
+[37.981s] [ext: isaacsim.asset.gen.omap-2.0.29] startup
+[38.230s] [ext: omni.vdb_timesample_editor-0.2.0] startup
+[38.250s] [ext: omni.physx.supportui-107.3.26] startup
+[38.309s] [ext: omni.physx.telemetry-107.3.26] startup
+[38.318s] [ext: omni.anim.graph.ui-107.3.3] startup
+[38.407s] [ext: isaaclab-0.54.3] startup
+[38.483s] [ext: isaaclab_contrib-0.0.2] startup
+[38.486s] [ext: omni.flowusd.bundle-1.1.2] startup
+[38.487s] [ext: omni.anim.curve.core-1.3.1] startup
+[38.541s] [ext: omni.physx.bundle-107.3.26] startup
+[38.543s] [ext: omni.anim.graph.bundle-107.3.3] startup
+[38.564s] [ext: omni.replicator.replicator_yaml-2.0.11] startup
+[38.590s] [ext: isaaclab_assets-0.2.4] startup
+[39.715s] [ext: isaaclab_tasks-0.11.13] startup
+[40.531s] [ext: isaaclab_rl-0.4.7] startup
+[40.534s] [ext: isaaclab_mimic-1.0.16] startup
+[40.536s] [ext: omni.kit.menu.common-2.0.1] startup
+[40.543s] [ext: magicsim.python-2.3.2] startup
+[40.547s] Simulation App Starting
+Warning: Possible version incompatibility. Attempting to load omni::fabric::IStageReaderWriter with version v0.14 against v0.13.
+2026-11-10T00:07:01Z [41,198ms] [Warning] [carb.audio.device] audio device is misconfigured or broken {deviceIndex = 0, name = 'default'} (The device is likely misconfigured, check your $HOME/.asoundrc)
+2026-11-10T00:07:01Z [41,198ms] [Warning] [carb.audio.output] failed to retrieve the capabilities for device 0 {result = eDeviceLost (2)}
+2026-11-10T00:07:01Z [41,198ms] [Warning] [carb.audio.context] failed to set the requested output during context creation.  Using a null streamer instead {result = eDeviceLost (2)}
+2026-11-10T00:07:02Z [41,297ms] [Warning] [omni.fabric.plugin] Warning: attribute viewPickingId not found for bucket id 9
+
+[41.764s] app ready
+[47.908s] Simulation App Startup Complete
+2026-11-10T00:07:26Z [65,740ms] [Warning] [matplotlib.font_manager] Matplotlib is building the font cache; this may take a moment.
+2026-11-09 19:07:31.382 | INFO     | magicsim.Env.Utils.file:info:52 - MemoryStation: RecordManager output dir: /mnt/scratch/xc536_atomic/n052saw_closeup_s101/rec/memorystage/000
+2026-11-09T19:07:31.382397-0500 MemoryStation: RecordManager output dir: /mnt/scratch/xc536_atomic/n052saw_closeup_s101/rec/memorystage/000
+2026-11-09T19:07:31.383139-0500 MemoryStation: record: magicsim recorder -> /mnt/scratch/xc536_atomic/n052saw_closeup_s101/rec/memorystage/000/magicsim
+2026-11-09T19:07:31.383250-0500 MemoryStation: record: sim_info recorder -> /mnt/scratch/xc536_atomic/n052saw_closeup_s101/rec/memorystage/000/sim_info
+2026-11-09 19:07:31.383 | INFO     | magicsim.Env.Utils.file:info:52 - MemoryStation: record: magicsim recorder -> /mnt/scratch/xc536_atomic/n052saw_closeup_s101/rec/memorystage/000/magicsim
+2026-11-09 19:07:31.383 | INFO     | magicsim.Env.Utils.file:info:52 - MemoryStation: record: sim_info recorder -> /mnt/scratch/xc536_atomic/n052saw_closeup_s101/rec/memorystage/000/sim_info
+[INFO]: Parsing configuration from: magicsim.Env.Environment.Isaac.IsaacRLEnv:IsaacRLEnvCfg
+
+======================================================================================
+[INFO][IsaacLab]: Logging to file: /tmp/isaaclab/logs/isaaclab_2026-11-09_19-07-31.log
+19:07:31 [simulation_context.py] WARNING: The `enable_external_forces_every_iteration` parameter in the PhysxCfg is set to False. If you are experiencing noisy velocities, consider enabling this flag. You may need to slightly increase the number of velocity iterations (setting it to 1 or 2 rather than 0), together with this flag, to improve the accuracy of velocity updates.
+[INFO]: Base environment:
+	Environment device    : cpu
+	Environment seed      : 101
+	Physics step-size     : 0.005
+	Rendering step-size   : 0.005
+	Environment step-size : 0.02
+19:07:31 [direct_rl_env.py] WARNING: The render interval (1) is smaller than the decimation (4). Multiple render calls will happen for each environment step.If this is not intended, set the render interval to be equal to the decimation.
+2026-11-10T00:07:33Z [72,385ms] [Warning] [isaacsim.sensors.camera.camera] 'verticalAperture' (1.5290800094604493) and 'horizontalAperture' (2.0954999923706055) are inconsistent with the pixel resolution aspect ratio (1.3333333333333333). Setting 'verticalAperture' to 1.571624994277954 to ensure square pixels.
+[NavManager] NavMeshVolume @ /NavMeshVolume: center=(0.000, 0.000, 1.250) scale=(30.800, 30.800, 3.500) offsets=(0.0, 0.0, 0)
+19:07:33 [interactive_scene.py] WARNING: Collision filtering can only be automatically enabled when replicate_physics=True and using GPU simulation. Please call scene.filter_collisions(global_prim_paths) to filter collisions across environments.
+[PhysXMonitor] suppressed 5 repeated msgs for [Warning] [carb.windowing-glfw.plugin] GLFW initialization failed.
+2026-11-10T00:07:33Z [72,644ms] [Warning] [carb.windowing-glfw.plugin] GLFW initialization failed.
+[PhysXMonitor] suppressed 5 repeated msgs for [Warning] [carb] Failed to startup plugin carb.windowing-glfw.plugin (interfaces: [carb::windowing::IGLContext v1.0],[carb::windowing::IWindowing v1.5]) (impl: carb.windowing-glfw.plugin)
+2026-11-10T00:07:33Z [72,644ms] [Warning] [carb] Failed to startup plugin carb.windowing-glfw.plugin (interfaces: [carb::windowing::IGLContext v1.0],[carb::windowing::IWindowing v1.5]) (impl: carb.windowing-glfw.plugin)
+[NavManager] NavMeshVolume @ /NavMeshVolume: center=(0.000, 0.000, 1.500) scale=(11.000, 11.000, 3.300) offsets=(0.0, 0.0, 0)
+[INFO]: Time taken for scene creation : 1.948619 seconds
+[INFO]: Scene manager:  <class InteractiveScene>
+	Number of environments: 1
+	Environment spacing   : 14
+	Source prim name      : /World/envs/env_0
+	Global prim paths     : []
+	Replicate physics     : False
+[INFO]: Starting the simulation. This may take a few seconds. Please wait...
+2026-11-10T00:07:35Z [74,951ms] [Warning] [omni.physx.cooking.plugin] ConvexDecompositionTask: failed to cook GPU-compatible mesh, collision detection will fall back to CPU. Collisions with particles and deformables will not work with this mesh. Prim /__Prototype_2/base_link_collisions_mesh_1/World/mesh
+[INFO]: Time taken for simulation start : 8.526572 seconds
+[INFO]: Completed setting up the environment...
+[INFO] Flow rendering enabled for fire/smoke/fluid effects
+       - /rtx/flow/enabled: True
+       - /rtx/flow/pathTracingEnabled: True
+       - /rtx/flow/rayTracedReflectionsEnabled: True
+       - /rtx/flow/rayTracedTranslucencyEnabled: True
+[INFO] effective render cfg: rendering_mode=performance antialiasing_mode=None enable_translucency=False enable_reflections=False enable_global_illumination=False enable_shadows=None enable_ambient_occlusion=None enable_direct_lighting=None enable_dl_denoiser=False enable_dlssg=None dlss_mode=None samples_per_pixel=None
+[INFO] effective render carb: /rtx/rendermode=RaytracedLighting /isaaclab/rendering/rendering_mode= /rtx/post/aa/op=3 /rtx/post/dlss/execMode=0 /rtx/translucency/enabled=False /rtx/reflections/enabled=False /rtx/indirectDiffuse/enabled=False /rtx/directLighting/enabled=True /rtx/directLighting/sampledLighting/enabled=False /rtx/shadows/enabled=True /rtx/ambientOcclusion/enabled=False /rtx-transient/dlssg/enabled=False /rtx-transient/dldenoiser/enabled=False /rtx/flow/enabled=True
+2026-11-10T00:07:42Z [81,520ms] [Warning] [magicsim.Env.Scene.Object.Background] `/omni/replicator/RTSubframes` must be > 3 to avoid blank textures while randomizing dome light texture. RTSubframes has been automatically increased from None to 3
+2026-11-10T00:07:49Z [88,664ms] [Error] [omni.physx.plugin] PhysicsUSD: Parse collision - triangle mesh collision (approximation None/MeshSimplification) cannot be a part of a dynamic body, falling back to convexHull approximation: /World/envs/env_0/dynamic/item_red/item_red_1.For dynamic collision please use approximations : convex hull, convex decomposition, box, sphere or SDF approximation.
+2026-11-10T00:07:49Z [88,888ms] [Error] [omni.physx.plugin] PhysicsUSD: Parse collision - triangle mesh collision (approximation None/MeshSimplification) cannot be a part of a dynamic body, falling back to convexHull approximation: /World/envs/env_0/dynamic/item_blue/item_blue_1.For dynamic collision please use approximations : convex hull, convex decomposition, box, sphere or SDF approximation.
+2026-11-10T00:08:27Z [126,484ms] [Warning] [omni.fabric.plugin] USD->Fabric: Unhandled array type string[]
+2026-11-10T00:08:27Z [126,487ms] [Warning] [usdrt.population.plugin] [UsdNoticeHandler] Unhandled attribute type VtArray<std::string> (prim attribute: omni:rtx:material:db:flattener:reflection_roughness_constant)
+2026-11-10T00:08:27Z [126,490ms] [Warning] [usdrt.population.plugin] [UsdNoticeHandler] Unhandled attribute type VtArray<std::string> (prim attribute: omni:rtx:material:db:flattener:ior_constant)
+2026-11-10T00:08:27Z [126,491ms] [Warning] [usdrt.population.plugin] [UsdNoticeHandler] Unhandled attribute type VtArray<std::string> (prim attribute: omni:rtx:material:db:flattener:transmittance_color)
+2026-11-10T00:08:27Z [126,583ms] [Warning] [omni.fabric.plugin] getAttributeCount called on non-existent path /World/envs/env_0/Robot_0/right_l6/visuals/right_l6_visuals_mesh_0/World
+2026-11-10T00:08:27Z [126,584ms] [Warning] [omni.fabric.plugin] getTypes called on non-existent path /World/envs/env_0/Robot_0/right_l6/visuals/right_l6_visuals_mesh_0/World
+2026-11-10T00:09:24Z [184,080ms] [Warning] [omni.syntheticdata.plugin] OgnSdPostRenderVarToHost : rendervar copy from texture directly to host buffer is counter-performant. Please use copy from texture to device buffer first.
+Scene Reset Finished
+[DualIKInstance 0] Creating LOCKED base IK solver...
+Warp DeprecationWarning: The symbol `warp.torch.device_from_torch` will soon be removed from the public API. Use `warp.device_from_torch` instead.
+[DualIKInstance 0] Creating FREE base IK solver...
+[DualIKServer] Initialized with 1 instance(s), batch_size=1, microbatch_wait_ms=400.0, num_seeds_locked=20, num_seeds_free=50, paired=True, planner_devices=['cuda:0']
+[DualMotionGenInstance 0] Creating LOCKED base MotionPlanner...
+[DualMotionGenInstance 0] Creating FREE base MotionPlanner...
+[DualMotionGenServer] Initialized with 1 instance(s), batch_size=1, microbatch_wait_ms=400.0, planner_devices=['cuda:0']
+pm:  <PlannerManager> total_action_dim=16
++--------------------------------------------------------------------------------+
+|                        Planner Slots (total input: 16)                         |
++--------------------+------+-------------+-----------+-------------+------------+
+| Robot              | Slot | Planner     | Input Dim | Input Slice | Output Dim |
+| RidgebackSawyerUmi | base | nav         |         8 |       [0:8] |          3 |
+| RidgebackSawyerUmi | arm  | passthrough |         7 |      [8:15] |          7 |
+| RidgebackSawyerUmi | eef  | passthrough |         1 |     [15:16] |          1 |
+[PhysXMonitor] suppressed 1 repeated msgs for 
+
+[tick-csv] contact reports subscribed for {'gripper_base': '/World/envs/env_0/Robot_0/right_gripper_base', 'electric_gripper_base': '/World/envs/env_0/Robot_0/right_electric_gripper_base', 'left': '/World/envs/env_0/Robot_0/right_gripper_l_finger', 'right': '/World/envs/env_0/Robot_0/right_gripper_r_finger', 'gripper_tip': '/World/envs/env_0/Robot_0/right_gripper_tip'}
+[tick-csv] writing /mnt/scratch/xc536_atomic/n052saw_closeup_s101/out/take_item_ticks.csv (fingers {'gripper_base': 'right_gripper_base', 'gripper_tip': 'right_gripper_tip', 'electric_gripper_base': 'right_electric_gripper_base', 'left': 'right_gripper_l_finger', 'right': 'right_gripper_r_finger'}, finger joints [], tick_dt 0.02)
+2026-11-10T00:10:25Z [244,763ms] [Warning] [carb] Client gpu.foundation.plugin has acquired [gpu::unstable::IMemoryBudgetManagerFactory v0.1] 100 times. Consider accessing this interface with carb::getCachedInterface() (Performance warning)
+Creating atomic skill: NavToPose for env 0
+action:  (0, tensor([ 0.0000, -0.5500,  1.5708]))
+[NavTo GP] env=0 robot=RidgebackSawyerUmi _build_nav_path target=[0.0, -0.550000011920929, 1.5707963705062866]
+[memory] tick 100 wall=272s t=2.0s base=(-3.22,-2.07,0.69) station=S skill=NavToPose/0 dwb_cmd=[0.3, -0.194, 0.8] raw_cmd=[0.3, -0.194, 0.8] rootz=0.06 jp=[0.205, -0.021, 0.434] jv=[0.352, -0.052, 0.8] eff=[205.9, 97.4, 6.8] vt=[0.353, -0.051, 0.8] base_quat=[0.666, 0.238, -0.666, 0.238] base_ang_vel=[-0.0, -0.0, 0.8] link_quat=[0.666, 0.238, -0.666, 0.238] eef=[-3.284, -2.246, 0.583] arm_q=[3.0, -1.0, 0.001, 1.999, 0.0, 1.4, 0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=29 path=[[-3.43, -2.1, 1.57], [-2.94, -1.88, 1.57], [-2.55, -1.71, 1.57], [-2.21, -1.56, 1.57], [-1.8, -1.39, 1.57], [-1.53, -1.27, 1.57], [-1.12, -1.09, 1.57], [-0.83, -0.96, 1.57], [-0.44, -0.79, 1.57], [-0.11, -0.65, 1.57], [0.0, -0.55, 1.57]] tgt=[0.0, -0.55, 1.57] public={'demo_running': False, 'demo_done': False}
+[memory] tick 200 wall=303s t=4.0s base=(-2.49,-1.75,1.48) station=S skill=NavToPose/0 dwb_cmd=[0.213, -0.3, 0.102] raw_cmd=[0.213, -0.3, 0.102] rootz=0.06 jp=[0.994, 0.107, 1.231] jv=[0.354, 0.1, 0.102] eff=[-18.7, 38.9, -1.6] vt=[0.354, 0.1, 0.102] base_quat=[0.522, 0.477, -0.522, 0.477] base_ang_vel=[-0.0, -0.0, 0.102] link_quat=[0.522, 0.477, -0.522, 0.477] eef=[-2.406, -1.916, 0.583] arm_q=[3.001, -1.0, 0.0, 2.0, -0.0, 1.4, -0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=29 path=[[-3.43, -2.1, 1.57], [-2.94, -1.88, 1.57], [-2.55, -1.71, 1.57], [-2.21, -1.56, 1.57], [-1.8, -1.39, 1.57], [-1.53, -1.27, 1.57], [-1.12, -1.09, 1.57], [-0.83, -0.96, 1.57], [-0.44, -0.79, 1.57], [-0.11, -0.65, 1.57], [0.0, -0.55, 1.57]] tgt=[0.0, -0.55, 1.57] public={'demo_running': False, 'demo_done': False}
+[memory] tick 300 wall=358s t=6.0s base=(-1.88,-1.43,1.56) station=S skill=NavToPose/0 dwb_cmd=[0.147, -0.3, 0.012] raw_cmd=[0.147, -0.3, 0.012] rootz=0.06 jp=[1.666, 0.263, 1.311] jv=[0.328, 0.065, 0.012] eff=[-1.3, 4.1, -0.1] vt=[0.328, 0.065, 0.012] base_quat=[0.502, 0.498, -0.502, 0.498] base_ang_vel=[-0.0, 0.0, 0.012] link_quat=[0.502, 0.498, -0.502, 0.498] eef=[-1.78, -1.591, 0.583] arm_q=[3.0, -1.0, 0.0, 2.0, 0.0, 1.4, -0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=29 path=[[-3.43, -2.1, 1.57], [-2.94, -1.88, 1.57], [-2.55, -1.71, 1.57], [-2.21, -1.56, 1.57], [-1.8, -1.39, 1.57], [-1.53, -1.27, 1.57], [-1.12, -1.09, 1.57], [-0.83, -0.96, 1.57], [-0.44, -0.79, 1.57], [-0.11, -0.65, 1.57], [0.0, -0.55, 1.57]] tgt=[0.0, -0.55, 1.57] public={'demo_running': False, 'demo_done': False}
+[memory] tick 400 wall=394s t=8.0s base=(-1.28,-1.16,1.57) station=S skill=NavToPose/0 dwb_cmd=[0.129, -0.294, 0.002] raw_cmd=[0.129, -0.294, 0.002] rootz=0.06 jp=[2.313, 0.377, 1.319] jv=[0.317, 0.052, 0.002] eff=[-0.2, 0.8, -0.0] vt=[0.317, 0.052, 0.002] base_quat=[0.5, 0.5, -0.5, 0.5] base_ang_vel=[-0.0, -0.0, 0.002] link_quat=[0.5, 0.5, -0.5, 0.5] eef=[-1.181, -1.319, 0.583] arm_q=[3.0, -1.0, -0.0, 2.0, 0.0, 1.4, -0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=29 path=[[-3.43, -2.1, 1.57], [-2.94, -1.88, 1.57], [-2.55, -1.71, 1.57], [-2.21, -1.56, 1.57], [-1.8, -1.39, 1.57], [-1.53, -1.27, 1.57], [-1.12, -1.09, 1.57], [-0.83, -0.96, 1.57], [-0.44, -0.79, 1.57], [-0.11, -0.65, 1.57], [0.0, -0.55, 1.57]] tgt=[0.0, -0.55, 1.57] public={'demo_running': False, 'demo_done': False}
+[memory] tick 500 wall=424s t=10.0s base=(-0.69,-0.90,1.57) station=S skill=NavToPose/0 dwb_cmd=[0.129, -0.3, 0.002] raw_cmd=[0.129, -0.3, 0.002] rootz=0.06 jp=[2.949, 0.48, 1.322] jv=[0.322, 0.051, 0.002] eff=[137.4, -34.4, 8.7] vt=[0.322, 0.051, 0.002] base_quat=[0.499, 0.501, -0.499, 0.501] base_ang_vel=[-0.0, 0.0, 0.002] link_quat=[0.499, 0.501, -0.499, 0.501] eef=[-0.59, -1.06, 0.583] arm_q=[3.0, -1.0, -0.0, 2.0, -0.0, 1.4, -0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=29 path=[[-3.43, -2.1, 1.57], [-2.94, -1.88, 1.57], [-2.55, -1.71, 1.57], [-2.21, -1.56, 1.57], [-1.8, -1.39, 1.57], [-1.53, -1.27, 1.57], [-1.12, -1.09, 1.57], [-0.83, -0.96, 1.57], [-0.44, -0.79, 1.57], [-0.11, -0.65, 1.57], [0.0, -0.55, 1.57]] tgt=[0.0, -0.55, 1.57] public={'demo_running': False, 'demo_done': False}
+[memory] tick 600 wall=454s t=12.0s base=(-0.16,-0.65,1.57) station=S skill=NavToPose/0 dwb_cmd=[0.076, -0.146, 0.004] raw_cmd=[0.076, -0.146, 0.004] rootz=0.06 jp=[3.52, 0.596, 1.32] jv=[0.161, 0.039, 0.004] eff=[-44.4, -228.2, 0.5] vt=[0.16, 0.037, 0.004] base_quat=[0.5, 0.5, -0.5, 0.5] base_ang_vel=[-0.0, 0.0, 0.004] link_quat=[0.5, 0.5, -0.5, 0.5] eef=[-0.065, -0.806, 0.583] arm_q=[3.001, -1.0, 0.0, 2.0, -0.0, 1.4, -0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=29 path=[[-3.43, -2.1, 1.57], [-2.94, -1.88, 1.57], [-2.55, -1.71, 1.57], [-2.21, -1.56, 1.57], [-1.8, -1.39, 1.57], [-1.53, -1.27, 1.57], [-1.12, -1.09, 1.57], [-0.83, -0.96, 1.57], [-0.44, -0.79, 1.57], [-0.11, -0.65, 1.57], [0.0, -0.55, 1.57]] tgt=[0.0, -0.55, 1.57] public={'demo_running': False, 'demo_done': False}
+Global planner NavTo of env 0 is finished or truncated, truncated: 0
+Atomic skill of env 0 is finished or truncated
+[memory] tick 700 wall=477s t=14.0s base=(-0.08,-0.60,1.57) station=A skill=None/None dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[3.612, 0.619, 1.32] jv=[0.0, 0.0, -0.0] eff=[-0.0, -0.0, 0.0] vt=[0.0, 0.0, 0.0] base_quat=[0.5, 0.5, -0.5, 0.5] base_ang_vel=[-0.0, -0.0, -0.0] link_quat=[0.5, 0.5, -0.5, 0.5] eef=[0.018, -0.761, 0.583] arm_q=[3.001, -1.0, -0.0, 2.001, -0.0, 1.4, -0.0] plan=None/None steps=None locked=None gp=NoneType nseg=0 public={'demo_running': True, 'demo_done': False}
+[memory] tick 800 wall=496s t=16.0s base=(-0.08,-0.60,1.57) station=A skill=None/None dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[3.612, 0.619, 1.32] jv=[0.0, 0.0, -0.0] eff=[-0.0, -0.0, 0.0] vt=[0.0, 0.0, 0.0] base_quat=[0.5, 0.5, -0.5, 0.5] base_ang_vel=[-0.0, -0.0, -0.0] link_quat=[0.5, 0.5, -0.5, 0.5] eef=[0.018, -0.761, 0.583] arm_q=[3.001, -1.0, -0.0, 2.001, -0.0, 1.4, -0.0] plan=None/None steps=None locked=None gp=NoneType nseg=0 public={'demo_running': True, 'demo_done': False}
+[memory] tick 900 wall=517s t=18.0s base=(-0.08,-0.60,1.57) station=A skill=None/None dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[3.612, 0.619, 1.32] jv=[0.0, 0.0, -0.0] eff=[-0.0, -0.0, 0.0] vt=[0.0, 0.0, 0.0] base_quat=[0.5, 0.5, -0.5, 0.5] base_ang_vel=[-0.0, -0.0, -0.0] link_quat=[0.5, 0.5, -0.5, 0.5] eef=[0.018, -0.761, 0.583] arm_q=[3.001, -1.0, -0.0, 2.001, -0.0, 1.4, -0.0] plan=None/None steps=None locked=None gp=NoneType nseg=0 public={'demo_running': True, 'demo_done': False}
+[memory] tick 1000 wall=551s t=20.0s base=(-0.08,-0.60,1.57) station=A skill=None/None dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[3.612, 0.619, 1.32] jv=[0.0, 0.0, -0.0] eff=[-0.0, -0.0, 0.0] vt=[0.0, 0.0, 0.0] base_quat=[0.5, 0.5, -0.5, 0.5] base_ang_vel=[-0.0, -0.0, -0.0] link_quat=[0.5, 0.5, -0.5, 0.5] eef=[0.018, -0.761, 0.583] arm_q=[3.001, -1.0, -0.0, 2.001, -0.0, 1.4, -0.0] plan=None/None steps=None locked=None gp=NoneType nseg=0 public={'demo_running': True, 'demo_done': False}
+Creating atomic skill: NavToPose for env 0
+action:  (0, tensor([ 0.0000,  2.0500, -1.5708]))
+[NavTo GP] env=0 robot=RidgebackSawyerUmi _build_nav_path target=[0.0, 2.049999952316284, -1.5707963705062866]
+[memory] tick 1100 wall=589s t=22.0s base=(0.15,-0.64,2.05) station=A skill=NavToPose/0 dwb_cmd=[-0.222, -0.3, 0.8] raw_cmd=[-0.222, -0.3, 0.8] rootz=0.06 jp=[3.822, 0.528, 1.802] jv=[0.341, -0.15, 0.8] eff=[168.1, 307.1, 8.5] vt=[0.342, -0.148, 0.8] base_quat=[0.366, 0.605, -0.366, 0.605] base_ang_vel=[-0.0, 0.0, 0.8] link_quat=[0.366, 0.605, -0.366, 0.605] eef=[0.307, -0.734, 0.583] arm_q=[3.001, -1.001, -0.0, 2.0, -0.0, 1.4, 0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=40 path=[[-0.08, -0.6, -1.57], [0.63, -0.52, -1.57], [1.09, -0.35, -1.57], [1.37, 0.38, -1.57], [1.46, 0.83, -1.57], [1.26, 1.31, -1.57], [1.05, 1.82, -1.57], [0.44, 2.07, -1.57], [0.0, 2.05, -1.57]] tgt=[0.0, 2.05, -1.57] public={'demo_running': False, 'demo_done': True}
+[memory] tick 1200 wall=620s t=24.0s base=(0.83,-0.61,-2.63) station=A skill=NavToPose/0 dwb_cmd=[-0.3, 0.063, 0.8] raw_cmd=[-0.3, 0.063, 0.8] rootz=0.06 jp=[4.493, 0.388, 3.401] jv=[0.307, 0.013, 0.8] eff=[-25.9, 241.8, 7.7] vt=[0.306, 0.015, 0.8] base_quat=[0.178, -0.684, -0.178, -0.684] base_ang_vel=[-0.0, 0.0, 0.8] link_quat=[0.178, -0.684, -0.178, -0.684] eef=[0.922, -0.442, 0.583] arm_q=[3.001, -1.001, -0.0, 2.0, -0.0, 1.4, 0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=40 path=[[-0.08, -0.6, -1.57], [0.63, -0.52, -1.57], [1.09, -0.35, -1.57], [1.37, 0.38, -1.57], [1.46, 0.83, -1.57], [1.26, 1.31, -1.57], [1.05, 1.82, -1.57], [0.44, 2.07, -1.57], [0.0, 2.05, -1.57]] tgt=[0.0, 2.05, -1.57] public={'demo_running': False, 'demo_done': True}
+[memory] tick 1300 wall=650s t=26.0s base=(1.20,-0.14,-1.68) station=A skill=NavToPose/0 dwb_cmd=[-0.3, 0.073, 0.127] raw_cmd=[-0.3, 0.073, 0.127] rootz=0.06 jp=[4.964, 0.744, 4.352] jv=[0.175, 0.255, 0.127] eff=[-17.5, 18.5, -0.1] vt=[0.175, 0.255, 0.127] base_quat=[0.472, -0.527, -0.472, -0.527] base_ang_vel=[-0.0, -0.0, 0.127] link_quat=[0.472, -0.527, -0.472, -0.527] eef=[1.118, 0.026, 0.583] arm_q=[3.001, -1.001, 0.0, 2.001, -0.0, 1.4, 0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=40 path=[[-0.08, -0.6, -1.57], [0.63, -0.52, -1.57], [1.09, -0.35, -1.57], [1.37, 0.38, -1.57], [1.46, 0.83, -1.57], [1.26, 1.31, -1.57], [1.05, 1.82, -1.57], [0.44, 2.07, -1.57], [0.0, 2.05, -1.57]] tgt=[0.0, 2.05, -1.57] public={'demo_running': False, 'demo_done': True}
+[memory] tick 1400 wall=679s t=28.0s base=(1.41,0.45,-1.58) station=A skill=NavToPose/0 dwb_cmd=[-0.3, 0.071, 0.016] raw_cmd=[-0.3, 0.071, 0.016] rootz=0.06 jp=[5.313, 1.265, 4.452] jv=[0.147, 0.271, 0.016] eff=[-231.6, 63.6, 14.7] vt=[0.146, 0.272, 0.016] base_quat=[0.498, -0.502, -0.498, -0.502] base_ang_vel=[-0.0, -0.0, 0.016] link_quat=[0.498, -0.502, -0.498, -0.502] eef=[1.31, 0.609, 0.583] arm_q=[3.0, -1.0, -0.0, 2.001, -0.0, 1.4, 0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=40 path=[[-0.08, -0.6, -1.57], [0.63, -0.52, -1.57], [1.09, -0.35, -1.57], [1.37, 0.38, -1.57], [1.46, 0.83, -1.57], [1.26, 1.31, -1.57], [1.05, 1.82, -1.57], [0.44, 2.07, -1.57], [0.0, 2.05, -1.57]] tgt=[0.0, 2.05, -1.57] public={'demo_running': False, 'demo_done': True}
+[memory] tick 1500 wall=710s t=30.0s base=(1.36,1.01,-1.57) station=A skill=NavToPose/0 dwb_cmd=[-0.3, -0.066, 0.0] raw_cmd=[-0.3, -0.066, 0.0] rootz=0.06 jp=[5.406, 1.822, 4.46] jv=[0.011, 0.307, 0.0] eff=[0.1, 1.1, 0.4] vt=[0.011, 0.307, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[-0.0, 0.0, 0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[1.26, 1.17, 0.583] arm_q=[3.002, -1.002, -0.001, 2.002, -0.0, 1.4, 0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=40 path=[[-0.08, -0.6, -1.57], [0.63, -0.52, -1.57], [1.09, -0.35, -1.57], [1.37, 0.38, -1.57], [1.46, 0.83, -1.57], [1.26, 1.31, -1.57], [1.05, 1.82, -1.57], [0.44, 2.07, -1.57], [0.0, 2.05, -1.57]] tgt=[0.0, 2.05, -1.57] public={'demo_running': False, 'demo_done': True}
+[memory] tick 1600 wall=740s t=32.0s base=(1.13,1.61,-1.57) station=A skill=NavToPose/0 dwb_cmd=[-0.273, -0.119, -0.0] raw_cmd=[-0.273, -0.119, -0.0] rootz=0.06 jp=[5.333, 2.459, 4.46] jv=[-0.048, 0.295, -0.0] eff=[209.5, -130.1, -14.2] vt=[-0.047, 0.294, -0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, -0.0, -0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[1.031, 1.77, 0.583] arm_q=[3.001, -1.0, 0.001, 2.0, 0.0, 1.4, 0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=40 path=[[-0.08, -0.6, -1.57], [0.63, -0.52, -1.57], [1.09, -0.35, -1.57], [1.37, 0.38, -1.57], [1.46, 0.83, -1.57], [1.26, 1.31, -1.57], [1.05, 1.82, -1.57], [0.44, 2.07, -1.57], [0.0, 2.05, -1.57]] tgt=[0.0, 2.05, -1.57] public={'demo_running': False, 'demo_done': True}
+[memory] tick 1700 wall=772s t=34.0s base=(0.63,1.98,-1.57) station=A skill=NavToPose/0 dwb_cmd=[-0.118, -0.3, -0.003] raw_cmd=[-0.118, -0.3, -0.003] rootz=0.06 jp=[4.938, 2.943, 4.46] jv=[-0.261, 0.189, -0.003] eff=[0.7, 0.7, 0.0] vt=[-0.261, 0.189, -0.003] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, 0.0, -0.003] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[0.528, 2.14, 0.583] arm_q=[3.001, -1.0, 0.0, 2.0, -0.0, 1.4, 0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=40 path=[[-0.08, -0.6, -1.57], [0.63, -0.52, -1.57], [1.09, -0.35, -1.57], [1.37, 0.38, -1.57], [1.46, 0.83, -1.57], [1.26, 1.31, -1.57], [1.05, 1.82, -1.57], [0.44, 2.07, -1.57], [0.0, 2.05, -1.57]] tgt=[0.0, 2.05, -1.57] public={'demo_running': False, 'demo_done': True}
+[memory] tick 1800 wall=803s t=36.0s base=(0.16,2.10,-1.57) station=B skill=NavToPose/0 dwb_cmd=[0.035, -0.154, -0.001] raw_cmd=[0.035, -0.154, -0.001] rootz=0.06 jp=[4.519, 3.172, 4.459] jv=[-0.158, 0.005, -0.001] eff=[0.5, 0.2, -0.7] vt=[-0.158, 0.005, -0.001] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[-0.0, 0.0, -0.001] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[0.066, 2.257, 0.583] arm_q=[3.001, -1.0, 0.0, 2.0, -0.0, 1.4, 0.0] plan=None/None steps=0 locked=None gp=NavTo nseg=40 path=[[-0.08, -0.6, -1.57], [0.63, -0.52, -1.57], [1.09, -0.35, -1.57], [1.37, 0.38, -1.57], [1.46, 0.83, -1.57], [1.26, 1.31, -1.57], [1.05, 1.82, -1.57], [0.44, 2.07, -1.57], [0.0, 2.05, -1.57]] tgt=[0.0, 2.05, -1.57] public={'demo_running': False, 'demo_done': True}
+Global planner NavTo of env 0 is finished or truncated, truncated: 0
+Atomic skill of env 0 is finished or truncated
+Creating atomic skill: Move for env 0
+Atomic skill of env 0 is finished or truncated
+[memory] event move_target {'label': 'take_item:pre', 'mode': -1, 'pos': [-0.1121, 1.1, 0.8541], 'quat': [0.5, -0.5, 0.5, 0.5], 'base': [0.0914, 2.0792, 0.06, -1.5727]}
+Creating atomic skill: Move for env 0
+Submit IK env_id= 0 robot= RidgebackSawyerUmi target= tensor([[-0.1121,  1.1000,  0.8541,  0.5000, -0.5000,  0.5000,  0.5000]])  lock_base=True (fixed base check)
+[DualIKServer][debug] Instance 0 [locked] Solved batch [0] (G=1) success=[True] goalset_index=[0] pos_err_max=0.0000m rot_err_max=0.0191rad dt=0.2518s
+IK (fixed base) SUCCESS -> using locked base MotionGen env_id= 0 robot= RidgebackSawyerUmi target= tensor([[-0.1121,  1.1000,  0.8541,  0.5000, -0.5000,  0.5000,  0.5000]])
+Submit MotionGen env_id= 0 robot= RidgebackSawyerUmi target= tensor([[-0.1121,  1.1000,  0.8541,  0.5000, -0.5000,  0.5000,  0.5000]])  lock_base=True
+[DualMotionGenServer][debug] Instance 0 [locked] plan_pose envs=[0] success=tensor([True], device='cuda:0') pos_err_max=0.0000m rot_err_max=0.0189rad dt=3.1426s
+MotionGen result success env_id= 0 robot= RidgebackSawyerUmi target= tensor([[-0.1121,  1.1000,  0.8541,  0.5000, -0.5000,  0.5000,  0.5000]])
+[MobileMoveL] waypoint 20/181 waits: eef 0 position off by 0.0625 (limit 0.0500) (cur - target [-0.062, 0.007, -0.004])
+[MobileMoveL] waypoint 21/181 waits: eef 0 position off by 0.0650 (limit 0.0500) (cur - target [-0.064, 0.013, -0.003])
+[MobileMoveL] waypoint 22/181 waits: eef 0 position off by 0.0667 (limit 0.0500) (cur - target [-0.065, 0.016, -0.003])
+[memory] tick 1900 wall=832s t=38.0s base=(0.09,2.08,-1.57) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.445, 3.172, 4.459] jv=[0.0, 0.0, -0.0] eff=[-0.3, -0.6, 0.6] vt=[0.0, -0.0, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, -0.0, -0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[0.055, 2.242, 0.59] arm_q=[2.926, -1.01, -0.086, 1.994, 0.018, 1.398, 0.14] arm_tgt=[-0.112, 1.1, 0.854] plan=22/181 steps=38 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+[MobileMoveL] waypoint 23/181 waits: eef 0 position off by 0.0704 (limit 0.0500) (cur - target [-0.067, 0.02, -0.003])
+[MobileMoveL] waypoint 24/181 waits: eef 0 position off by 0.0756 (limit 0.0500) (cur - target [-0.071, 0.025, -0.003])
+[MobileMoveL] waypoint 25/181 waits: eef 0 position off by 0.0817 (limit 0.0500) (cur - target [-0.076, 0.03, -0.003])
+[MobileMoveL] waypoint 26/181 waits: eef 0 position off by 0.0885 (limit 0.0500) (cur - target [-0.08, 0.037, -0.004])
+[MobileMoveL] waypoint 27/181 waits: eef 0 position off by 0.0955 (limit 0.0500) (cur - target [-0.085, 0.044, -0.004])
+[MobileMoveL] waypoint 28/181 waits: eef 0 position off by 0.1025 (limit 0.0500) (cur - target [-0.089, 0.051, -0.004])
+[MobileMoveL] waypoint 29/181 waits: eef 0 position off by 0.1095 (limit 0.0500) (cur - target [-0.092, 0.059, -0.004])
+[MobileMoveL] waypoint 30/181 waits: eef 0 position off by 0.1163 (limit 0.0500) (cur - target [-0.095, 0.067, -0.004])
+[MobileMoveL] waypoint 31/181 waits: eef 0 position off by 0.1231 (limit 0.0500) (cur - target [-0.097, 0.075, -0.004])
+[MobileMoveL] waypoint 32/181 waits: eef 0 position off by 0.1297 (limit 0.0500) (cur - target [-0.099, 0.084, -0.004])
+[MobileMoveL] waypoint 33/181 waits: eef 0 position off by 0.1361 (limit 0.0500) (cur - target [-0.1, 0.092, -0.004])
+[MobileMoveL] waypoint 34/181 waits: eef 0 position off by 0.1424 (limit 0.0500) (cur - target [-0.1, 0.101, -0.004])
+[memory] tick 2000 wall=854s t=40.0s base=(0.09,2.08,-1.57) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.445, 3.172, 4.459] jv=[-0.0, -0.0, -0.0] eff=[0.7, 0.7, 1.0] vt=[0.0, -0.0, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, -0.0, -0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[0.28, 2.143, 0.634] arm_q=[2.593, -1.104, -0.458, 2.047, 0.113, 1.308, 0.417] arm_tgt=[-0.112, 1.1, 0.854] plan=34/181 steps=138 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+[MobileMoveL] waypoint 35/181 waits: eef 0 position off by 0.1484 (limit 0.0500) (cur - target [-0.1, 0.11, -0.003])
+[MobileMoveL] waypoint 36/181 waits: eef 0 position off by 0.1543 (limit 0.0500) (cur - target [-0.099, 0.119, -0.003])
+[MobileMoveL] waypoint 37/181 waits: eef 0 position off by 0.1599 (limit 0.0500) (cur - target [-0.097, 0.127, -0.003])
+[MobileMoveL] waypoint 38/181 waits: eef 0 position off by 0.1652 (limit 0.0500) (cur - target [-0.094, 0.136, -0.002])
+[MobileMoveL] waypoint 39/181 waits: eef 0 position off by 0.1702 (limit 0.0500) (cur - target [-0.091, 0.144, -0.001])
+[MobileMoveL] waypoint 40/181 waits: eef 0 position off by 0.1748 (limit 0.0500) (cur - target [-0.087, 0.152, -0.0])
+[MobileMoveL] waypoint 41/181 waits: eef 0 position off by 0.1790 (limit 0.0500) (cur - target [-0.083, 0.159, 0.001])
+[MobileMoveL] waypoint 42/181 waits: eef 0 position off by 0.1826 (limit 0.0500) (cur - target [-0.078, 0.165, 0.002])
+[MobileMoveL] waypoint 43/181 waits: eef 0 position off by 0.1858 (limit 0.0500) (cur - target [-0.072, 0.171, 0.003])
+[MobileMoveL] waypoint 44/181 waits: eef 0 position off by 0.1884 (limit 0.0500) (cur - target [-0.067, 0.176, 0.005])
+[MobileMoveL] waypoint 45/181 waits: eef 0 position off by 0.1903 (limit 0.0500) (cur - target [-0.061, 0.18, 0.006])
+[MobileMoveL] waypoint 46/181 waits: eef 0 position off by 0.1916 (limit 0.0500) (cur - target [-0.054, 0.184, 0.008])
+[MobileMoveL] waypoint 47/181 waits: eef 0 position off by 0.1922 (limit 0.0500) (cur - target [-0.048, 0.186, 0.01])
+[memory] tick 2100 wall=877s t=42.0s base=(0.09,2.08,-1.57) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.445, 3.172, 4.459] jv=[0.0, -0.0, -0.0] eff=[-0.2, 0.1, 1.2] vt=[0.0, -0.0, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, -0.0, -0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[0.392, 1.931, 0.715] arm_q=[2.261, -1.309, -0.829, 2.043, 0.081, 1.204, 0.711] arm_tgt=[-0.112, 1.1, 0.854] plan=47/181 steps=238 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+[MobileMoveL] waypoint 48/181 waits: eef 0 position off by 0.1921 (limit 0.0500) (cur - target [-0.042, 0.187, 0.012])
+[MobileMoveL] waypoint 49/181 waits: eef 0 position off by 0.1913 (limit 0.0500) (cur - target [-0.036, 0.187, 0.014])
+[MobileMoveL] waypoint 50/181 waits: eef 0 position off by 0.1899 (limit 0.0500) (cur - target [-0.031, 0.187, 0.016])
+[MobileMoveL] waypoint 51/181 waits: eef 0 position off by 0.1878 (limit 0.0500) (cur - target [-0.026, 0.185, 0.018])
+[MobileMoveL] waypoint 52/181 waits: eef 0 position off by 0.1851 (limit 0.0500) (cur - target [-0.022, 0.183, 0.021])
+[MobileMoveL] waypoint 53/181 waits: eef 0 position off by 0.1818 (limit 0.0500) (cur - target [-0.018, 0.179, 0.023])
+[MobileMoveL] waypoint 54/181 waits: eef 0 position off by 0.1780 (limit 0.0500) (cur - target [-0.016, 0.176, 0.025])
+[MobileMoveL] waypoint 55/181 waits: eef 0 position off by 0.1739 (limit 0.0500) (cur - target [-0.014, 0.171, 0.027])
+[MobileMoveL] waypoint 56/181 waits: eef 0 position off by 0.1696 (limit 0.0500) (cur - target [-0.013, 0.167, 0.029])
+[MobileMoveL] waypoint 57/181 waits: eef 0 position off by 0.1652 (limit 0.0500) (cur - target [-0.014, 0.162, 0.031])
+[MobileMoveL] waypoint 58/181 waits: eef 0 position off by 0.1608 (limit 0.0500) (cur - target [-0.015, 0.157, 0.032])
+[MobileMoveL] waypoint 59/181 waits: eef 0 position off by 0.1566 (limit 0.0500) (cur - target [-0.018, 0.152, 0.033])
+[memory] tick 2200 wall=900s t=44.0s base=(0.09,2.08,-1.57) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.445, 3.172, 4.459] jv=[-0.0, -0.0, -0.0] eff=[3.3, 0.8, 4.2] vt=[0.0, -0.0, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, -0.0, -0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[0.319, 1.679, 0.815] arm_q=[1.93, -1.559, -1.2, 1.856, -0.154, 1.238, 0.788] arm_tgt=[-0.112, 1.1, 0.854] plan=59/181 steps=338 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+[MobileMoveL] waypoint 60/181 waits: eef 0 position off by 0.1529 (limit 0.0500) (cur - target [-0.021, 0.148, 0.034])
+[MobileMoveL] waypoint 61/181 waits: eef 0 position off by 0.1498 (limit 0.0500) (cur - target [-0.024, 0.144, 0.034])
+[MobileMoveL] waypoint 62/181 waits: eef 0 position off by 0.1474 (limit 0.0500) (cur - target [-0.025, 0.142, 0.033])
+[MobileMoveL] waypoint 63/181 waits: eef 0 position off by 0.1458 (limit 0.0500) (cur - target [-0.025, 0.14, 0.032])
+[MobileMoveL] waypoint 64/181 waits: eef 0 position off by 0.1446 (limit 0.0500) (cur - target [-0.024, 0.139, 0.031])
+[MobileMoveL] waypoint 65/181 waits: eef 0 position off by 0.1440 (limit 0.0500) (cur - target [-0.023, 0.139, 0.031])
+[MobileMoveL] waypoint 66/181 waits: eef 0 position off by 0.1440 (limit 0.0500) (cur - target [-0.022, 0.139, 0.03])
+[MobileMoveL] waypoint 67/181 waits: eef 0 position off by 0.1444 (limit 0.0500) (cur - target [-0.019, 0.14, 0.029])
+[MobileMoveL] waypoint 68/181 waits: eef 0 position off by 0.1452 (limit 0.0500) (cur - target [-0.014, 0.142, 0.028])
+[MobileMoveL] waypoint 69/181 waits: eef 0 position off by 0.1463 (limit 0.0500) (cur - target [-0.009, 0.144, 0.026])
+[MobileMoveL] waypoint 70/181 waits: eef 0 position off by 0.1475 (limit 0.0500) (cur - target [-0.005, 0.145, 0.025])
+[MobileMoveL] waypoint 71/181 waits: eef 0 position off by 0.1489 (limit 0.0500) (cur - target [-0.001, 0.147, 0.024])
+[MobileMoveL] waypoint 72/181 waits: eef 0 position off by 0.1504 (limit 0.0500) (cur - target [0.003, 0.149, 0.023])
+[memory] tick 2300 wall=923s t=46.0s base=(0.09,2.08,-1.57) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.445, 3.172, 4.459] jv=[-0.0, 0.0, -0.0] eff=[0.6, -0.1, 2.0] vt=[0.0, -0.0, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, -0.0, -0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[0.225, 1.561, 0.872] arm_q=[1.597, -1.424, -1.453, 1.75, -0.023, 1.118, 0.532] arm_tgt=[-0.112, 1.1, 0.854] plan=72/181 steps=438 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+[MobileMoveL] waypoint 73/181 waits: eef 0 position off by 0.1520 (limit 0.0500) (cur - target [0.007, 0.15, 0.022])
+[MobileMoveL] waypoint 74/181 waits: eef 0 position off by 0.1537 (limit 0.0500) (cur - target [0.01, 0.152, 0.022])
+[MobileMoveL] waypoint 75/181 waits: eef 0 position off by 0.1555 (limit 0.0500) (cur - target [0.014, 0.153, 0.021])
+[MobileMoveL] waypoint 76/181 waits: eef 0 position off by 0.1574 (limit 0.0500) (cur - target [0.018, 0.155, 0.02])
+[MobileMoveL] waypoint 77/181 waits: eef 0 position off by 0.1593 (limit 0.0500) (cur - target [0.022, 0.157, 0.02])
+[MobileMoveL] waypoint 78/181 waits: eef 0 position off by 0.1614 (limit 0.0500) (cur - target [0.025, 0.158, 0.019])
+[MobileMoveL] waypoint 79/181 waits: eef 0 position off by 0.1634 (limit 0.0500) (cur - target [0.029, 0.16, 0.019])
+[MobileMoveL] waypoint 80/181 waits: eef 0 position off by 0.1655 (limit 0.0500) (cur - target [0.033, 0.161, 0.018])
+[MobileMoveL] waypoint 81/181 waits: eef 0 position off by 0.1675 (limit 0.0500) (cur - target [0.038, 0.162, 0.018])
+[MobileMoveL] waypoint 82/181 waits: eef 0 position off by 0.1695 (limit 0.0500) (cur - target [0.042, 0.163, 0.017])
+[MobileMoveL] waypoint 83/181 waits: eef 0 position off by 0.1715 (limit 0.0500) (cur - target [0.046, 0.164, 0.017])
+[MobileMoveL] waypoint 84/181 waits: eef 0 position off by 0.1734 (limit 0.0500) (cur - target [0.05, 0.165, 0.016])
+[memory] tick 2400 wall=946s t=48.0s base=(0.09,2.08,-1.57) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.445, 3.172, 4.459] jv=[-0.0, 0.0, -0.0] eff=[0.4, -0.4, 0.9] vt=[0.0, -0.0, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, -0.0, -0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[0.188, 1.495, 0.903] arm_q=[1.266, -1.303, -1.433, 1.673, 0.033, 1.042, 0.349] arm_tgt=[-0.112, 1.1, 0.854] plan=84/181 steps=538 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+[MobileMoveL] waypoint 85/181 waits: eef 0 position off by 0.1753 (limit 0.0500) (cur - target [0.055, 0.166, 0.016])
+[MobileMoveL] waypoint 86/181 waits: eef 0 position off by 0.1770 (limit 0.0500) (cur - target [0.059, 0.166, 0.015])
+[MobileMoveL] waypoint 87/181 waits: eef 0 position off by 0.1786 (limit 0.0500) (cur - target [0.063, 0.166, 0.015])
+[MobileMoveL] waypoint 88/181 waits: eef 0 position off by 0.1801 (limit 0.0500) (cur - target [0.068, 0.166, 0.014])
+[MobileMoveL] waypoint 89/181 waits: eef 0 position off by 0.1815 (limit 0.0500) (cur - target [0.072, 0.166, 0.013])
+[MobileMoveL] waypoint 90/181 waits: eef 0 position off by 0.1826 (limit 0.0500) (cur - target [0.077, 0.165, 0.013])
+[MobileMoveL] waypoint 91/181 waits: eef 0 position off by 0.1836 (limit 0.0500) (cur - target [0.081, 0.164, 0.012])
+[MobileMoveL] waypoint 92/181 waits: eef 0 position off by 0.1843 (limit 0.0500) (cur - target [0.085, 0.163, 0.011])
+[MobileMoveL] waypoint 93/181 waits: eef 0 position off by 0.1848 (limit 0.0500) (cur - target [0.089, 0.161, 0.011])
+[MobileMoveL] waypoint 94/181 waits: eef 0 position off by 0.1851 (limit 0.0500) (cur - target [0.093, 0.16, 0.01])
+[MobileMoveL] waypoint 95/181 waits: eef 0 position off by 0.1852 (limit 0.0500) (cur - target [0.097, 0.157, 0.009])
+[MobileMoveL] waypoint 96/181 waits: eef 0 position off by 0.1849 (limit 0.0500) (cur - target [0.101, 0.155, 0.008])
+[MobileMoveL] waypoint 97/181 waits: eef 0 position off by 0.1844 (limit 0.0500) (cur - target [0.104, 0.152, 0.007])
+[memory] tick 2500 wall=970s t=50.0s base=(0.09,2.08,-1.57) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.445, 3.172, 4.459] jv=[0.0, -0.0, 0.0] eff=[-0.2, 0.0, -0.1] vt=[0.0, -0.0, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[-0.0, 0.0, 0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[0.167, 1.408, 0.899] arm_q=[0.936, -1.074, -1.397, 1.654, 0.265, 0.95, 0.035] arm_tgt=[-0.112, 1.1, 0.854] plan=97/181 steps=638 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+[MobileMoveL] waypoint 98/181 waits: eef 0 position off by 0.1837 (limit 0.0500) (cur - target [0.107, 0.149, 0.006])
+[MobileMoveL] waypoint 99/181 waits: eef 0 position off by 0.1826 (limit 0.0500) (cur - target [0.109, 0.146, 0.005])
+[MobileMoveL] waypoint 100/181 waits: eef 0 position off by 0.1812 (limit 0.0500) (cur - target [0.112, 0.143, 0.004])
+[MobileMoveL] waypoint 101/181 waits: eef 0 position off by 0.1795 (limit 0.0500) (cur - target [0.113, 0.139, 0.003])
+[MobileMoveL] waypoint 102/181 waits: eef 0 position off by 0.1776 (limit 0.0500) (cur - target [0.115, 0.135, 0.002])
+[MobileMoveL] waypoint 103/181 waits: eef 0 position off by 0.1752 (limit 0.0500) (cur - target [0.116, 0.131, 0.001])
+[MobileMoveL] waypoint 104/181 waits: eef 0 position off by 0.1726 (limit 0.0500) (cur - target [0.117, 0.127, 0.0])
+[MobileMoveL] waypoint 105/181 waits: eef 0 position off by 0.1697 (limit 0.0500) (cur - target [0.117, 0.123, -0.001])
+[MobileMoveL] waypoint 106/181 waits: eef 0 position off by 0.1664 (limit 0.0500) (cur - target [0.117, 0.118, -0.002])
+[MobileMoveL] waypoint 107/181 waits: eef 0 position off by 0.1628 (limit 0.0500) (cur - target [0.117, 0.113, -0.003])
+[MobileMoveL] waypoint 108/181 waits: eef 0 position off by 0.1589 (limit 0.0500) (cur - target [0.116, 0.108, -0.004])
+[MobileMoveL] waypoint 109/181 waits: eef 0 position off by 0.1547 (limit 0.0500) (cur - target [0.115, 0.103, -0.006])
+[memory] tick 2600 wall=991s t=52.0s base=(0.09,2.08,-1.57) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.445, 3.172, 4.459] jv=[0.0, -0.0, -0.0] eff=[-0.1, 0.1, 0.2] vt=[0.0, -0.0, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, -0.0, -0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[0.107, 1.301, 0.872] arm_q=[0.606, -0.821, -1.35, 1.634, 0.642, 0.92, -0.4] arm_tgt=[-0.112, 1.1, 0.854] plan=109/181 steps=738 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+[MobileMoveL] waypoint 110/181 waits: eef 0 position off by 0.1502 (limit 0.0500) (cur - target [0.113, 0.098, -0.007])
+[MobileMoveL] waypoint 111/181 waits: eef 0 position off by 0.1454 (limit 0.0500) (cur - target [0.111, 0.093, -0.008])
+[MobileMoveL] waypoint 112/181 waits: eef 0 position off by 0.1404 (limit 0.0500) (cur - target [0.109, 0.088, -0.009])
+[MobileMoveL] waypoint 113/181 waits: eef 0 position off by 0.1351 (limit 0.0500) (cur - target [0.106, 0.083, -0.01])
+[MobileMoveL] waypoint 114/181 waits: eef 0 position off by 0.1296 (limit 0.0500) (cur - target [0.103, 0.078, -0.011])
+[MobileMoveL] waypoint 115/181 waits: eef 0 position off by 0.1238 (limit 0.0500) (cur - target [0.099, 0.073, -0.012])
+[MobileMoveL] waypoint 116/181 waits: eef 0 position off by 0.1179 (limit 0.0500) (cur - target [0.096, 0.068, -0.013])
+[MobileMoveL] waypoint 117/181 waits: eef 0 position off by 0.1120 (limit 0.0500) (cur - target [0.092, 0.063, -0.014])
+[MobileMoveL] waypoint 118/181 waits: eef 0 position off by 0.1062 (limit 0.0500) (cur - target [0.087, 0.059, -0.014])
+[MobileMoveL] waypoint 119/181 waits: eef 0 position off by 0.1003 (limit 0.0500) (cur - target [0.083, 0.054, -0.014])
+[MobileMoveL] waypoint 120/181 waits: eef 0 position off by 0.0944 (limit 0.0500) (cur - target [0.079, 0.05, -0.014])
+[MobileMoveL] waypoint 121/181 waits: eef 0 position off by 0.0886 (limit 0.0500) (cur - target [0.074, 0.047, -0.013])
+[MobileMoveL] waypoint 122/181 waits: eef 0 position off by 0.0828 (limit 0.0500) (cur - target [0.07, 0.043, -0.013])
+[memory] tick 2700 wall=1011s t=54.0s base=(0.09,2.08,-1.57) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.445, 3.172, 4.459] jv=[-0.0, -0.0, -0.0] eff=[0.0, 0.2, 1.4] vt=[0.0, -0.0, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, -0.0, -0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[0.005, 1.206, 0.85] arm_q=[0.276, -0.632, -1.296, 1.485, 0.878, 1.062, -0.745] arm_tgt=[-0.112, 1.1, 0.854] plan=122/181 steps=838 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+[MobileMoveL] waypoint 123/181 waits: eef 0 position off by 0.0770 (limit 0.0500) (cur - target [0.065, 0.039, -0.012])
+[MobileMoveL] waypoint 124/181 waits: eef 0 position off by 0.0713 (limit 0.0500) (cur - target [0.06, 0.036, -0.012])
+[MobileMoveL] waypoint 125/181 waits: eef 0 position off by 0.0656 (limit 0.0500) (cur - target [0.056, 0.033, -0.011])
+[MobileMoveL] waypoint 126/181 waits: eef 0 position off by 0.0600 (limit 0.0500) (cur - target [0.051, 0.029, -0.01])
+[MobileMoveL] waypoint 127/181 waits: eef 0 position off by 0.0544 (limit 0.0500) (cur - target [0.047, 0.026, -0.009])
+[MobileMoveL] waypoint 128/181 waits: eef 0 position off by 0.0511 (limit 0.0500) (cur - target [0.044, 0.025, -0.008])
+[MobileMoveL] waypoint 129/181 waits: eef 0 position off by 0.0509 (limit 0.0500) (cur - target [0.044, 0.025, -0.008])
+[MobileMoveL] waypoint 130/181 waits: eef 0 position off by 0.0504 (limit 0.0500) (cur - target [0.044, 0.024, -0.008])
+[MobileMoveL] waypoint 132/181 waits: eef 0 position off by 0.0501 (limit 0.0500) (cur - target [0.043, 0.024, -0.007])
+[memory] tick 2800 wall=1031s t=56.0s base=(0.09,2.08,-1.57) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.445, 3.172, 4.459] jv=[-0.0, -0.0, -0.0] eff=[0.4, 0.5, 1.8] vt=[0.0, -0.0, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, -0.0, -0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[-0.092, 1.122, 0.843] arm_q=[-0.055, -0.635, -1.033, 1.164, 0.761, 1.228, -0.906] arm_tgt=[-0.112, 1.1, 0.854] plan=170/181 steps=938 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+Global planner MobileMoveL of env 0 is finished or truncated, truncated: 0
+Atomic skill of env 0 is finished or truncated
+[memory] event move_target {'label': 'take_item:down:0', 'mode': 4, 'pos': [-0.1121, 1.1, 0.7341], 'quat': [0.5, -0.5, 0.5, 0.5], 'base': [0.0914, 2.0792, 0.06, -1.5727]}
+Creating atomic skill: Move for env 0
+[tick-csv] window opens at tick 2837 (phase take_item:down:0)
+[memory] tick 2900 wall=1051s t=58.0s base=(0.09,2.08,-1.57) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.445, 3.172, 4.459] jv=[-0.0, 0.0, -0.0] eff=[0.0, -0.2, 0.7] vt=[0.0, -0.0, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, -0.0, -0.0] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[-0.111, 1.102, 0.766] arm_q=[-0.205, -0.693, -0.662, 1.139, 0.489, 1.247, -0.671] arm_tgt=[-0.112, 1.1, 0.734] plan=0/None steps=64 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+Global planner MobileMoveL of env 0 is finished or truncated, truncated: 0
+Atomic skill of env 0 is finished or truncated
+[memory] event settle_err {'label': 'take_item:down', 'k': 0, 'err_mm': [-0.2, -0.3, -4.6], 'norm_mm': 4.6}
+Creating atomic skill: Move for env 0
+Atomic skill of env 0 is finished or truncated
+[memory] event move_target {'label': 'take_item:lift', 'mode': 4, 'pos': [-0.1121, 1.1, 0.8541], 'quat': [0.5, -0.5, 0.5, 0.5], 'base': [0.0913, 2.0797, 0.06, -1.5728]}
+Creating atomic skill: Move for env 0
+[memory] tick 3000 wall=1072s t=60.0s base=(0.09,2.08,-1.57) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.445, 3.173, 4.459] jv=[0.001, -0.001, 0.004] eff=[-144.6, 94.6, -403.3] vt=[0.0, -0.0, 0.0] base_quat=[0.5, -0.5, -0.5, -0.5] base_ang_vel=[0.0, -0.0, 0.004] link_quat=[0.5, -0.5, -0.5, -0.5] eef=[-0.156, 1.684, 1.026] arm_q=[-0.456, -0.667, -1.708, 1.32, -0.881, 1.925, -1.178] arm_tgt=[-0.112, 1.1, 0.854] plan=0/None steps=62 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+[MobileMoveL] final arrival waits: eef 0 position off by 0.4615 (limit 0.0060) (cur - target [-0.367, -0.034, -0.278])
+[memory] tick 3100 wall=1094s t=62.0s base=(0.09,2.08,-1.58) station=B skill=Move/0 dwb_cmd=[0.0, 0.0, 0.0] raw_cmd=[0.0, 0.0, 0.0] rootz=0.06 jp=[4.449, 3.175, 4.453] jv=[0.004, -0.002, -0.0] eff=[-439.8, 164.5, 40.2] vt=[0.0, -0.0, 0.0] base_quat=[0.498, -0.502, -0.498, -0.502] base_ang_vel=[0.0, -0.0, -0.0] link_quat=[0.498, -0.502, -0.498, -0.502] eef=[-0.44, 2.431, 0.373] arm_q=[-3.05, 0.764, -3.043, 0.598, -2.862, 2.976, -1.056] arm_tgt=[-0.112, 1.1, 0.854] plan=0/None steps=162 locked=True gp=MobileMoveL nseg=0 public={'demo_running': False, 'demo_done': True}
+Resetting env tensor([0]) with seed None and options None
+Global planner MobileMoveL of env 0 is finished or truncated, truncated: 2
+Atomic skill of env 0 is finished or truncated
+Task of env 0 is finished or truncated, state: truncated: env truncated
+[memory] episode 1 done: full_success=False terminated=False reason='item_dropped' t=62.06s
+[chase] video: /mnt/scratch/xc536_atomic/n052saw_closeup_s101/rec/cams/cam1/video.mp4 (1551 frames, every 2 steps)
+MEMORY_RESULT {"task_id": "N052", "config": "/mnt/scratch/xc536_atomic/n052saw_closeup_s101/ptree/src/magicsim/Collect/Task/MemoryStation/Conf/memory_n052_easy.yaml", "success": false, "planner_seeds": "default", "dr": {"seed": 101, "spawn_jitter": 0.3, "start": [-3.4252, -2.0984, 0.2512], "nominal_start": [-3.3, -2.3, 0.0]}, "output_dir": "/mnt/scratch/xc536_atomic/n052saw_closeup_s101/rec/memorystage/000", "chase_video": "/mnt/scratch/xc536_atomic/n052saw_closeup_s101/rec/cams/cam1/video.mp4", "steps": 3103, "elapsed_s": 1115.253, "episodes_run": 1, "episodes_full_success": 0}
+[memory] record finalize: nothing_to_enqueue
+[PhysXMonitor] suppressed 2 repeated msgs for [Warning] [carb.windowing-glfw.plugin] GLFW initialization failed.
+2026-11-10T00:24:50Z [1,109,388ms] [Warning] [carb.windowing-glfw.plugin] GLFW initialization failed.
+[PhysXMonitor] suppressed 2 repeated msgs for [Warning] [carb] Failed to startup plugin carb.windowing-glfw.plugin (interfaces: [carb::windowing::IGLContext v1.0],[carb::windowing::IWindowing v1.5]) (impl: carb.windowing-glfw.plugin)
+2026-11-10T00:24:50Z [1,109,389ms] [Warning] [carb] Failed to startup plugin carb.windowing-glfw.plugin (interfaces: [carb::windowing::IGLContext v1.0],[carb::windowing::IWindowing v1.5]) (impl: carb.windowing-glfw.plugin)
+2026-11-10T00:24:50Z [1,109,510ms] [Warning] [omni.usd] Unexpected reference count of 2 for UsdStage 'anon:0x55644ad4c2d0:World0.usd' while being closed in UsdContext (this may indicate it is still resident in memory).
+total 62796
+drwxr-xr-x.  2 xc536 xc536       69 Nov  9 19:24 .
+drwxr-xr-x. 11 xc536 xc536      128 Nov  9 19:07 ..
+-rw-r--r--.  1 xc536 xc536    20703 Nov  9 19:24 report.json
+-rw-r--r--.  1 xc536 xc536   697121 Nov  9 19:24 take_item_ticks.csv
+-rw-r--r--.  1 xc536 xc536 63577131 Nov  9 19:24 video.mp4
+INFO:    Terminating fuse-overlayfs after timeout
+INFO:    Timeouts can be caused by a running background process
+exit 0 2026-11-09 19:24:54

debug/536/n052_sawyer/step_2026-11-09/step_table_s101.txt

@@ -0,0 +1,18 @@
+close rows 82 100 gp cols: ['gp_state', 'gp_step_count', 'interp_substep', 'interp_steps', 'planner_mode', 'arm_tgt_x', 'arm_tgt_y', 'arm_tgt_z']
+79 2916 take_item:do running 4 arm_tgt 1.0999 | eef y 1.100 z 0.742 | qtgt-qtgt0: +0.000 +0.000 +0.000 +0.000 +0.000 +0.000 +0.000 | qtgt-q: -0.003 -0.000 +0.009 +0.004 -0.003 -0.003 +0.005 | fingers 0.0208 -0.0208 Ld 0
+80 2917 take_item:do running 4 arm_tgt 1.0999 | eef y 1.100 z 0.740 | qtgt-qtgt0: -0.001 -0.001 +0.004 +0.001 -0.002 -0.001 +0.002 | qtgt-q: -0.003 -0.001 +0.009 +0.004 -0.004 -0.003 +0.004 | fingers 0.0208 -0.0208 Ld 0
+81 2918    arm_tgt  | eef y 1.100 z 0.739 | qtgt-qtgt0: -0.003 -0.002 +0.008 +0.002 -0.005 -0.002 +0.004 | qtgt-q: -0.003 -0.002 +0.009 +0.003 -0.004 -0.003 +0.004 | fingers 0.0208 -0.0208 Ld 0
+82 2919 take_item:cl running close arm_tgt  | eef y 1.100 z 0.738 | qtgt-qtgt0: -0.004 -0.003 +0.012 +0.003 -0.007 -0.002 +0.005 | qtgt-q: -0.003 -0.003 +0.009 +0.002 -0.005 -0.002 +0.004 | fingers 0.0188 -0.0188 Ld 0
+83 2920 take_item:cl running close arm_tgt  | eef y 1.100 z 0.736 | qtgt-qtgt0: -0.006 -0.004 +0.015 +0.004 -0.010 -0.003 +0.007 | qtgt-q: -0.003 -0.003 +0.009 +0.002 -0.005 -0.002 +0.004 | fingers 0.0168 -0.0168 Ld 0
+84 2921 take_item:cl running close arm_tgt  | eef y 1.100 z 0.736 | qtgt-qtgt0: -0.007 -0.005 +0.019 +0.005 -0.013 -0.004 +0.009 | qtgt-q: -0.003 -0.004 +0.009 +0.002 -0.006 -0.002 +0.004 | fingers 0.0148 -0.0148 Ld 0
+85 2922 take_item:cl running close arm_tgt  | eef y 1.100 z 0.735 | qtgt-qtgt0: -0.009 -0.007 +0.023 +0.005 -0.016 -0.004 +0.011 | qtgt-q: -0.003 -0.004 +0.009 +0.001 -0.006 -0.002 +0.004 | fingers 0.0131 -0.0128 Ld 3
+86 2923 take_item:cl running close arm_tgt  | eef y 1.103 z 0.734 | qtgt-qtgt0: -0.010 -0.008 +0.025 +0.006 -0.020 -0.002 +0.012 | qtgt-q: -0.003 -0.004 +0.010 -0.001 -0.003 -0.004 +0.004 | fingers 0.0147 -0.0111 Ld 52
+87 2924 take_item:cl running close arm_tgt  | eef y 1.107 z 0.732 | qtgt-qtgt0: -0.010 -0.008 +0.022 +0.010 -0.022 +0.000 +0.013 | qtgt-q: -0.001 -0.002 +0.011 -0.005 +0.002 -0.006 +0.003 | fingers 0.0165 -0.0093 Ld 54
+88 2925 take_item:cl running close arm_tgt  | eef y 1.112 z 0.731 | qtgt-qtgt0: -0.011 -0.008 +0.017 +0.013 -0.024 +0.003 +0.013 | qtgt-q: +0.002 -0.001 +0.011 -0.008 +0.006 -0.008 +0.001 | fingers 0.0182 -0.0075 Ld 57
+89 2926 take_item:cl running close arm_tgt  | eef y 1.116 z 0.729 | qtgt-qtgt0: -0.012 -0.007 +0.011 +0.016 -0.024 +0.005 +0.012 | qtgt-q: +0.005 +0.001 +0.011 -0.011 +0.010 -0.010 -0.000 | fingers 0.0200 -0.0058 Ld 59
+90 2927 take_item:cl running close arm_tgt  | eef y 1.122 z 0.728 | qtgt-qtgt0: -0.012 -0.007 +0.004 +0.022 -0.023 +0.005 +0.010 | qtgt-q: +0.009 +0.001 +0.013 -0.012 +0.018 -0.016 -0.002 | fingers 0.0208 -0.0042 Ld 113
+91 2928 take_item:cl running close arm_tgt  | eef y 1.144 z 0.721 | qtgt-qtgt0: -0.016 -0.011 -0.020 +0.041 -0.044 +0.024 +0.002 | qtgt-q: +0.010 -0.003 +0.020 -0.019 +0.027 -0.026 -0.010 | fingers 0.0208 -0.0037 Ld 332
+92 2929 take_item:cl running close arm_tgt  | eef y 1.179 z 0.711 | qtgt-qtgt0: -0.027 -0.016 -0.060 +0.080 -0.098 +0.060 -0.011 | qtgt-q: +0.013 -0.006 +0.020 -0.020 +0.033 -0.027 -0.018 | fingers 0.0208 -0.0036 Ld 409
+93 2930 take_item:cl running close arm_tgt  | eef y 1.218 z 0.703 | qtgt-qtgt0: -0.045 -0.019 -0.100 +0.121 -0.161 +0.099 -0.020 | qtgt-q: +0.014 -0.006 +0.020 -0.020 +0.033 -0.028 -0.021 | fingers 0.0208 -0.0035 Ld 426
+94 2931 take_item:cl running close arm_tgt  | eef y 1.260 z 0.699 | qtgt-qtgt0: -0.069 -0.023 -0.141 +0.161 -0.224 +0.141 -0.026 | qtgt-q: +0.015 -0.005 +0.020 -0.020 +0.033 -0.028 -0.021 | fingers 0.0208 -0.0035 Ld 426
+95 2932 take_item:cl running close arm_tgt  | eef y 1.304 z 0.698 | qtgt-qtgt0: -0.096 -0.031 -0.181 +0.202 -0.287 +0.181 -0.032 | qtgt-q: +0.016 -0.004 +0.020 -0.020 +0.033 -0.027 -0.021 | fingers 0.0208 -0.0035 Ld 430