Update to own progress-reward5 checkpoint500 with static leg gain0.54; full local evidence

README.md

@@ -10,13 +10,13 @@ pipeline_tag: robotics
 
 # jhp123 Microduck sprint2m optimized
 
-Own progress-reward5 training branch, checkpoint500, exported with the official CPU exporter and calibrated with a static0.60 leg-output multiplier. The policy includes its own observation normalization and has61 observations,14 actions and50Hz control. Original network and embedded normalization are preserved; only legs0–4 and9–13 are scaled directly from the original parent by0.60, head5–8 unchanged. No bias is added. Actual applied actions feed back through the unmodified runner. No foreign weights, seed classifier, step counter or per-seed action tables were added.
+Own progress-reward5 training branch, checkpoint500, exported with the official CPU exporter and calibrated with a static0.54 leg-output multiplier. The policy includes its own observation normalization and has61 observations,14 actions and50Hz control. Original network and embedded normalization are preserved; only legs0–4 and9–13 are scaled directly from the original parent by0.54, head5–8 unchanged. No bias is added. Actual applied actions feed back through the unmodified runner. No foreign weights, seed classifier, step counter or per-seed action tables were added.
 
 ## Result and limitations
 
-Local unmodified Arena runner5: **2.470s,5/5 finishes**, selected command1.05m/s. All33 evaluation races are in evaluation-local.json. The public Arena determines the official result at submission.
+Local unmodified Arena runner5: **2.297s,5/5 finishes**, selected command1.00m/s. All33 evaluation races are in evaluation-local.json. The public Arena determines the official result at submission.
 
-This replaces the same training parent with leg gain0.80 (official2.586s5/5 atcommand1.20m/s). The present improvement is inference calibration and selected command, not additional reward training. Complete second finite-grid results: gains0.40 and0.50 each50s0/5,0.60=2.470s5/5,0.70=2.534s5/5,0.80=2.586s5/5 exact control. All failures are retained; smaller gain does not uniformly improve results.
+This replaces the same training parent with leg gain0.60 (official2.470s5/5 atcommand1.05m/s). The present improvement is inference calibration and selected command, not additional reward training. Complete final finite-grid results: gain0.52=12.170s4/5(seed2 timeout),0.54=2.297s5/5,0.56=2.392s5/5,0.58=11.995s4/5(seed4 fall),0.60=2.470s5/5 exact control. Failures each count50s. This closes scalar refinement on this parent; gain effects are not monotonic. A separate later checkpoint1000 without calibration scored2.341s5/5 locally; that comparison changes checkpoint as well as calibration.
 
 The earlier checkpoint250 of this same branch finished4/5 with a50s timeout penalty (mean12.997s). Its failure diagnostic showed turning, lateral drift and low-posture stalling; it does not prove the cause. Checkpoint500 finishes all five competition seeds, but no extra-seed test or paired training-seed replication has been performed. Earlier models' extra-seed results do not apply to this model. This is a simulator competition policy without physical robot validation.
 
@@ -25,7 +25,7 @@ The earlier checkpoint250 of this same branch finished4/5 with a50s timeout pena
 - Challenges commit c24a50146aab950fc306e8430b85e15b1a1477e5, derived from9aaec6db450dd91c7aa669fc7156d81ce43de3c7.
 - Added reward5*clip(actual body-x linear velocity,0,2.4)/2.4, with nonfinite velocity rewarded0. Default other rewards (forward2,upright2,air3), curriculum, observations, robot and physics are retained.
 - The reward uses body-frame velocity, not world-lane progress or commanded speed. Turning and posture risks remain.
-- Own run v2_progress5_seed1_cu129,seed1,4096environments,3000iteration target; this release checkpoint500. Original network and embedded normalizer preserved; final static leg multiplier0.60 only. See calibration-provenance.json.
+- Own run v2_progress5_seed1_cu129,seed1,4096environments,3000iteration target; this release checkpoint500. Original network and embedded normalizer preserved; final static leg multiplier0.54 only. See calibration-provenance.json.
 - RL commit8d0dbc94cbbef3f1dcae461a4dbc1a43bb9886a0.
 - runtime-cu129-v1: Python3.12.12,Torch2.9.1+cu129,CUDA12.9,A800. This documented variant differs from the original Linux x86 cu128 lock.
 - MuJoCo3.10.0,mujoco-warp3.8.1,warp-lang1.12.0,ONNX Runtime1.24.4,NumPy2.4.1,mjlab1.3.0.

calibration-provenance.json

@@ -1,11 +1,11 @@
 {
-  "candidate": "g060",
-  "version": "stagec-500-gain-block2",
+  "candidate": "g054",
+  "version": "stagec-500-gain-block3",
   "parent_onnx_sha256": "059d8c3f1d3f4bbf5bbddd8a4b712e654c0b1a70c8635cab0da3875604ff9953",
   "parent_training_run": "v2_progress5_seed1_cu129",
   "parent_checkpoint": 500,
   "parent_runtime": "runtime-cu129-v1",
-  "gain": 0.6,
+  "gain": 0.54,
   "leg_indices": [
     0,
     1,
@@ -26,8 +26,8 @@
   ],
   "transform": "static output Mul; actual applied action fed back by unchanged official runner",
   "identity_original_bytes": false,
-  "reused_g080_control_bytes": false,
-  "onnx_sha256": "437b8a2d4f802ed8af3d8979326d184dc311f7b875437a271c190f93e09a9ea2",
+  "reused_g060_control_bytes": false,
+  "onnx_sha256": "6838debe046da035539e20b991023eb288771cc38506c7ab0d148b523b413a49",
   "bytes": 793986,
   "verification_samples": 65,
   "verification_rng_seed": 731047,

evaluation-local.json

@@ -1,169 +1,169 @@
 {
   "event_id": "sprint-2m",
-  "score": 2.47,
+  "score": 2.2969999999999997,
   "seeds_finished": 5,
-  "command_vx": 1.05,
+  "command_vx": 1.0,
   "per_seed": [
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.47
+      "time_s": 2.2800000000000002
     },
     {
       "seed": 2,
       "status": "finish",
-      "time_s": 2.475
+      "time_s": 2.2600000000000002
     },
     {
       "seed": 3,
       "status": "finish",
-      "time_s": 2.47
+      "time_s": 2.27
     },
     {
       "seed": 4,
       "status": "finish",
-      "time_s": 2.515
+      "time_s": 2.32
     },
     {
       "seed": 5,
       "status": "finish",
-      "time_s": 2.42
+      "time_s": 2.355
     }
   ],
   "scan": [
     {
       "command_vx": 0.2,
-      "status": "finish",
-      "time_s": 4.075,
-      "progress_m": 2.0027654855974815
+      "status": "timeout",
+      "time_s": null,
+      "progress_m": 0.3598511611471979
     },
     {
       "command_vx": 0.25,
       "status": "timeout",
       "time_s": null,
-      "progress_m": 1.7524245463208155
+      "progress_m": 0.2442637948040293
     },
     {
       "command_vx": 0.3,
-      "status": "timeout",
-      "time_s": null,
-      "progress_m": 1.5185630165280777
+      "status": "finish",
+      "time_s": 2.805,
+      "progress_m": 2.01636582609771
     },
     {
       "command_vx": 0.35,
-      "status": "timeout",
-      "time_s": null,
-      "progress_m": 1.8141126072254157
+      "status": "finish",
+      "time_s": 2.725,
+      "progress_m": 2.011301354889784
     },
     {
       "command_vx": 0.4,
       "status": "finish",
-      "time_s": 3.25,
-      "progress_m": 2.0064218591194916
+      "time_s": 2.6550000000000002,
+      "progress_m": 2.006084639561046
     },
     {
       "command_vx": 0.45,
-      "status": "finish",
-      "time_s": 3.075,
-      "progress_m": 2.0029286949002083
+      "status": "fall",
+      "time_s": null,
+      "progress_m": 1.9509441717538014
     },
     {
       "command_vx": 0.5,
       "status": "finish",
-      "time_s": 2.93,
-      "progress_m": 2.0089074794267967
+      "time_s": 2.715,
+      "progress_m": 2.004769425865507
     },
     {
       "command_vx": 0.55,
       "status": "finish",
-      "time_s": 2.86,
-      "progress_m": 2.002858063919632
+      "time_s": 2.57,
+      "progress_m": 2.0072603277233942
     },
     {
       "command_vx": 0.6,
       "status": "finish",
-      "time_s": 2.765,
-      "progress_m": 2.0098707703945276
+      "time_s": 2.66,
+      "progress_m": 2.001558376847346
     },
     {
       "command_vx": 0.65,
       "status": "finish",
-      "time_s": 2.745,
-      "progress_m": 2.0113274636792853
+      "time_s": 2.48,
+      "progress_m": 2.0013880724861477
     },
     {
       "command_vx": 0.7,
       "status": "finish",
-      "time_s": 2.6750000000000003,
-      "progress_m": 2.0063363923510478
+      "time_s": 2.58,
+      "progress_m": 2.001275094950715
     },
     {
       "command_vx": 0.75,
       "status": "finish",
-      "time_s": 2.625,
-      "progress_m": 2.011192192776856
+      "time_s": 3.0500000000000003,
+      "progress_m": 2.009923419746193
     },
     {
       "command_vx": 0.8,
-      "status": "finish",
-      "time_s": 2.585,
-      "progress_m": 2.0108214850506148
+      "status": "timeout",
+      "time_s": null,
+      "progress_m": 1.8689499932304465
     },
     {
       "command_vx": 0.85,
       "status": "finish",
-      "time_s": 2.5300000000000002,
-      "progress_m": 2.01367437327696
+      "time_s": 2.4250000000000003,
+      "progress_m": 2.0141464980148225
     },
     {
       "command_vx": 0.9,
       "status": "finish",
-      "time_s": 2.495,
-      "progress_m": 2.0047354802880872
+      "time_s": 2.375,
+      "progress_m": 2.005986816866649
     },
     {
       "command_vx": 0.95,
       "status": "finish",
-      "time_s": 2.54,
-      "progress_m": 2.001232471052818
+      "time_s": 2.32,
+      "progress_m": 2.0036310636279273
     },
     {
       "command_vx": 1.0,
       "status": "finish",
-      "time_s": 2.545,
-      "progress_m": 2.007364170363112
+      "time_s": 2.2800000000000002,
+      "progress_m": 2.0017134097251543
     },
     {
       "command_vx": 1.05,
       "status": "finish",
-      "time_s": 2.47,
-      "progress_m": 2.01175540659192
+      "time_s": 2.285,
+      "progress_m": 2.014012112845102
     },
     {
       "command_vx": 1.1,
       "status": "finish",
-      "time_s": 2.535,
-      "progress_m": 2.0042140677466276
+      "time_s": 2.285,
+      "progress_m": 2.0170626570287307
     },
     {
       "command_vx": 1.15,
-      "status": "fall",
-      "time_s": null,
-      "progress_m": 1.7969064304494193
+      "status": "finish",
+      "time_s": 2.29,
+      "progress_m": 2.011771361209214
     },
     {
       "command_vx": 1.2,
       "status": "finish",
-      "time_s": 2.59,
-      "progress_m": 2.0071723023048436
+      "time_s": 2.345,
+      "progress_m": 2.0116227429510314
     }
   ],
   "all_runs": [
     {
       "seed": 1,
-      "status": "finish",
-      "time_s": 4.075,
-      "progress_m": 2.0027654855974815,
+      "status": "timeout",
+      "time_s": null,
+      "progress_m": 0.3598511611471979,
       "command": [
         0.2,
         0.0,
@@ -184,7 +184,7 @@
       "seed": 1,
       "status": "timeout",
       "time_s": null,
-      "progress_m": 1.7524245463208155,
+      "progress_m": 0.2442637948040293,
       "command": [
         0.25,
         0.0,
@@ -203,9 +203,9 @@
     },
     {
       "seed": 1,
-      "status": "timeout",
-      "time_s": null,
-      "progress_m": 1.5185630165280777,
+      "status": "finish",
+      "time_s": 2.805,
+      "progress_m": 2.01636582609771,
       "command": [
         0.3,
         0.0,
@@ -224,9 +224,9 @@
     },
     {
       "seed": 1,
-      "status": "timeout",
-      "time_s": null,
-      "progress_m": 1.8141126072254157,
+      "status": "finish",
+      "time_s": 2.725,
+      "progress_m": 2.011301354889784,
       "command": [
         0.35,
         0.0,
@@ -246,8 +246,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 3.25,
-      "progress_m": 2.0064218591194916,
+      "time_s": 2.6550000000000002,
+      "progress_m": 2.006084639561046,
       "command": [
         0.4,
         0.0,
@@ -266,9 +266,9 @@
     },
     {
       "seed": 1,
-      "status": "finish",
-      "time_s": 3.075,
-      "progress_m": 2.0029286949002083,
+      "status": "fall",
+      "time_s": null,
+      "progress_m": 1.9509441717538014,
       "command": [
         0.45,
         0.0,
@@ -288,8 +288,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.93,
-      "progress_m": 2.0089074794267967,
+      "time_s": 2.715,
+      "progress_m": 2.004769425865507,
       "command": [
         0.5,
         0.0,
@@ -309,8 +309,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.86,
-      "progress_m": 2.002858063919632,
+      "time_s": 2.57,
+      "progress_m": 2.0072603277233942,
       "command": [
         0.55,
         0.0,
@@ -330,8 +330,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.765,
-      "progress_m": 2.0098707703945276,
+      "time_s": 2.66,
+      "progress_m": 2.001558376847346,
       "command": [
         0.6,
         0.0,
@@ -351,8 +351,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.745,
-      "progress_m": 2.0113274636792853,
+      "time_s": 2.48,
+      "progress_m": 2.0013880724861477,
       "command": [
         0.65,
         0.0,
@@ -372,8 +372,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.6750000000000003,
-      "progress_m": 2.0063363923510478,
+      "time_s": 2.58,
+      "progress_m": 2.001275094950715,
       "command": [
         0.7,
         0.0,
@@ -393,8 +393,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.625,
-      "progress_m": 2.011192192776856,
+      "time_s": 3.0500000000000003,
+      "progress_m": 2.009923419746193,
       "command": [
         0.75,
         0.0,
@@ -413,9 +413,9 @@
     },
     {
       "seed": 1,
-      "status": "finish",
-      "time_s": 2.585,
-      "progress_m": 2.0108214850506148,
+      "status": "timeout",
+      "time_s": null,
+      "progress_m": 1.8689499932304465,
       "command": [
         0.8,
         0.0,
@@ -435,8 +435,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.5300000000000002,
-      "progress_m": 2.01367437327696,
+      "time_s": 2.4250000000000003,
+      "progress_m": 2.0141464980148225,
       "command": [
         0.85,
         0.0,
@@ -456,8 +456,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.495,
-      "progress_m": 2.0047354802880872,
+      "time_s": 2.375,
+      "progress_m": 2.005986816866649,
       "command": [
         0.9,
         0.0,
@@ -477,8 +477,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.54,
-      "progress_m": 2.001232471052818,
+      "time_s": 2.32,
+      "progress_m": 2.0036310636279273,
       "command": [
         0.95,
         0.0,
@@ -498,8 +498,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.545,
-      "progress_m": 2.007364170363112,
+      "time_s": 2.2800000000000002,
+      "progress_m": 2.0017134097251543,
       "command": [
         1.0,
         0.0,
@@ -519,8 +519,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.47,
-      "progress_m": 2.01175540659192,
+      "time_s": 2.285,
+      "progress_m": 2.014012112845102,
       "command": [
         1.05,
         0.0,
@@ -540,8 +540,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.535,
-      "progress_m": 2.0042140677466276,
+      "time_s": 2.285,
+      "progress_m": 2.0170626570287307,
       "command": [
         1.1,
         0.0,
@@ -560,9 +560,9 @@
     },
     {
       "seed": 1,
-      "status": "fall",
-      "time_s": null,
-      "progress_m": 1.7969064304494193,
+      "status": "finish",
+      "time_s": 2.29,
+      "progress_m": 2.011771361209214,
       "command": [
         1.15,
         0.0,
@@ -582,8 +582,8 @@
     {
       "seed": 1,
       "status": "finish",
-      "time_s": 2.59,
-      "progress_m": 2.0071723023048436,
+      "time_s": 2.345,
+      "progress_m": 2.0116227429510314,
       "command": [
         1.2,
         0.0,
@@ -603,10 +603,10 @@
     {
       "seed": 2,
       "status": "finish",
-      "time_s": 2.475,
-      "progress_m": 2.005020497796699,
+      "time_s": 2.2600000000000002,
+      "progress_m": 2.0003241319515483,
       "command": [
-        1.05,
+        1.0,
         0.0,
         0.0,
         0.0,
@@ -624,10 +624,10 @@
     {
       "seed": 3,
       "status": "finish",
-      "time_s": 2.47,
-      "progress_m": 2.007769333433072,
+      "time_s": 2.27,
+      "progress_m": 2.009889857375639,
       "command": [
-        1.05,
+        1.0,
         0.0,
         0.0,
         0.0,
@@ -645,10 +645,10 @@
     {
       "seed": 4,
       "status": "finish",
-      "time_s": 2.515,
-      "progress_m": 2.0035117251385888,
+      "time_s": 2.32,
+      "progress_m": 2.002928781119554,
       "command": [
-        1.05,
+        1.0,
         0.0,
         0.0,
         0.0,
@@ -666,10 +666,10 @@
     {
       "seed": 5,
       "status": "finish",
-      "time_s": 2.42,
-      "progress_m": 2.002224396023953,
+      "time_s": 2.355,
+      "progress_m": 2.006399994771144,
       "command": [
-        1.05,
+        1.0,
         0.0,
         0.0,
         0.0,
@@ -687,10 +687,10 @@
     {
       "seed": 2,
       "status": "finish",
-      "time_s": 2.5100000000000002,
-      "progress_m": 2.0123830488222874,
+      "time_s": 2.31,
+      "progress_m": 2.0111897154805987,
       "command": [
-        0.9,
+        1.05,
         0.0,
         0.0,
         0.0,
@@ -708,10 +708,10 @@
     {
       "seed": 3,
       "status": "finish",
-      "time_s": 2.555,
-      "progress_m": 2.0057061956800677,
+      "time_s": 2.33,
+      "progress_m": 2.009699913423101,
       "command": [
-        0.9,
+        1.05,
         0.0,
         0.0,
         0.0,
@@ -728,11 +728,11 @@
     },
     {
       "seed": 4,
-      "status": "fall",
-      "time_s": null,
-      "progress_m": 1.7937194787506907,
+      "status": "finish",
+      "time_s": 2.29,
+      "progress_m": 2.0113452586170832,
       "command": [
-        0.9,
+        1.05,
         0.0,
         0.0,
         0.0,
@@ -750,10 +750,10 @@
     {
       "seed": 5,
       "status": "finish",
-      "time_s": 2.46,
-      "progress_m": 2.0027438123974073,
+      "time_s": 2.36,
+      "progress_m": 2.003006035254662,
       "command": [
-        0.9,
+        1.05,
         0.0,
         0.0,
         0.0,
@@ -771,10 +771,10 @@
     {
       "seed": 2,
       "status": "finish",
-      "time_s": 2.545,
-      "progress_m": 2.008959976323715,
+      "time_s": 2.335,
+      "progress_m": 2.0059317412726596,
       "command": [
-        0.85,
+        1.1,
         0.0,
         0.0,
         0.0,
@@ -792,10 +792,10 @@
     {
       "seed": 3,
       "status": "finish",
-      "time_s": 2.545,
-      "progress_m": 2.016361387173898,
+      "time_s": 2.315,
+      "progress_m": 2.0073925343009686,
       "command": [
-        0.85,
+        1.1,
         0.0,
         0.0,
         0.0,
@@ -813,10 +813,10 @@
     {
       "seed": 4,
       "status": "finish",
-      "time_s": 2.6350000000000002,
-      "progress_m": 2.0029987039745634,
+      "time_s": 2.355,
+      "progress_m": 2.0069032408578753,
       "command": [
-        0.85,
+        1.1,
         0.0,
         0.0,
         0.0,
@@ -833,11 +833,11 @@
     },
     {
       "seed": 5,
-      "status": "finish",
-      "time_s": 2.515,
-      "progress_m": 2.0068040852076017,
+      "status": "fall",
+      "time_s": null,
+      "progress_m": 1.9443689105908013,
       "command": [
-        0.85,
+        1.1,
         0.0,
         0.0,
         0.0,
@@ -854,7 +854,7 @@
     }
   ],
   "runner_version": 5,
-  "model_sha256": "437b8a2d4f802ed8af3d8979326d184dc311f7b875437a271c190f93e09a9ea2",
+  "model_sha256": "6838debe046da035539e20b991023eb288771cc38506c7ab0d148b523b413a49",
   "arena_sha": "e05128f36bc45c453509fe32fdd9614d3731025e",
   "scope": "Local unmodified official rules; official result determined at submission."
 }

manifest.json

@@ -13,7 +13,7 @@
   "name": "sprint2m-optimized",
   "kind": "perpetual",
   "entry_pose": "standing",
-  "description": "Own progress-reward5 branch checkpoint500, static leg gain0.60 applied to original official ONNX export; local2.470s5/5; simulation only.",
+  "description": "Own progress-reward5 branch checkpoint500, static leg gain0.54 applied to original official ONNX export; local2.297s5/5; simulation only.",
   "command": {
     "encoding": "constant",
     "idle": [

policy.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:437b8a2d4f802ed8af3d8979326d184dc311f7b875437a271c190f93e09a9ea2
+oid sha256:6838debe046da035539e20b991023eb288771cc38506c7ab0d148b523b413a49
 size 793986