blog: HQ case videos (2)

blog/bigym2-dishwasher-close-i00-privileged--nfkkn2w/meta.json

@@ -0,0 +1,166 @@
+{
+ "slug": "bigym2-dishwasher-close-i00-privileged--nfkkn2w",
+ "status": "rendered",
+ "task": "bigym2-dishwasher-close-i00-privileged",
+ "bench": "bigym2",
+ "mode": "privileged",
+ "run": "antigravity_cli-gemini_4_argon-high",
+ "outcome": "success",
+ "card": "Gemini 4 Argon, privileged, solved",
+ "source_trial": "antigravity_cli/gemini_4_argon/high/antigravity_cli-gemini_4_argon-high-bigym2-dishwasher-close-i00-privileged/bigym2-dishwasher-close-i00-priv__nFkKN2w",
+ "what": "re-rendered: kinematic playback of this trial's own replay (its recorded actions replayed from the task's frozen instance with the task's own tooling, physics unchanged) in Blender Cycles",
+ "renderer": "Blender 4.2.23 LTS LTS Cycles path tracing (GPU OptiX, RTX 4090), adaptive sampling (threshold 0.01, 64-128 spp), OpenImageDenoise with albedo + normal passes; kinematic playback of the exact replay poses (the simulator's own compiled meshes, primitives and colours); daylight: a sun lamp (2 degree disc) and a Nishita sky (no sun disc), the scene's own floor colour on a large ground plane with a faint world-space variation; AgX view transform, Medium High Contrast look; camera director with a slow eased 10.0-degree orbit; 2560x1440, 30 fps, no text or overlays",
+ "rendering": "path-tracing",
+ "width": 2560,
+ "height": 1440,
+ "fps": 30,
+ "frames_rendered": 348,
+ "render_seconds": 1208.7,
+ "seconds_per_frame": 3.47,
+ "time_range": {
+  "episode": "the submitted trajectory (privileged: one replay from the frozen instance, as the verifier runs it)",
+  "from_step": 0,
+  "to_step": 693,
+  "sim_t_from_s": 0.0,
+  "sim_t_to_s": 13.86,
+  "why": "the whole graded replay (693 rows, 13.9 s at 50 Hz): with the left arm raised it walks beside the dishwasher, pushes the upper and lower racks in, squats, slides the flat hand under the open door and sweeps it shut while standing up (success)"
+ },
+ "speed": 1.2,
+ "holds": {
+  "start_s": 0.5,
+  "end_s": 1.5,
+  "note": "first/last frame held, no extra simulation"
+ },
+ "replay_check": {
+  "reproduced": true,
+  "how": "bigym2: every recorded row replayed from the frozen instance with the task's own tooling (physics unchanged), the verdict + final state hash taken where the episode ends; a second independent replay (the tooling's own replay()) must agree; then compared with the trial's records",
+  "recorded": {
+   "success": true,
+   "state_sha": "48a90df87b66a689ece31189e09db886b2857163",
+   "executed_rows": 693,
+   "episode_over": true,
+   "fell": false
+  },
+  "replayed": {
+   "success": true,
+   "state_sha": "48a90df87b66a689ece31189e09db886b2857163",
+   "executed_rows": 693,
+   "episode_over": true,
+   "outcome": "success",
+   "fell": false,
+   "horizon": 4650
+  },
+  "plain_replay": {
+   "success": true,
+   "state_sha": "48a90df87b66a689ece31189e09db886b2857163",
+   "executed_rows": 693,
+   "episode_over": true,
+   "outcome": "success",
+   "fell": false,
+   "instance_exact": true
+  },
+  "matches": {
+   "success": true,
+   "state_sha": true,
+   "executed_rows": true,
+   "episode_over": true,
+   "fell": true
+  },
+  "replays_agree": true,
+  "instance": {
+   "onboard_render": "onboard camera images blanked (mujoco Renderer.render -> zeros); physics and controller untouched",
+   "instance_check": {
+    "ok": true,
+    "exact": true,
+    "max_joint_dev": 0.0
+   },
+   "t0_state_sha": "712d920d41fa309109fcdee7c0247f6ff9e4d051",
+   "t0_state_sha_file": "712d920d41fa309109fcdee7c0247f6ff9e4d051"
+  }
+ },
+ "camera": {
+  "mode": "director",
+  "az_deg": 160.0,
+  "az_rule": "fixed",
+  "el_deg": 26.0,
+  "fovy_deg": 34.0,
+  "fill": 0.8,
+  "distance_m": [
+   3.075,
+   3.381
+  ],
+  "robot_root": "g1_29dof_with_dex1_1/pelvis",
+  "travel_m": 0.219,
+  "focus": {
+   "robot+extra": 348
+  },
+  "objects": {
+   "dishwasher/": 0.45
+  }
+ },
+ "visual_changes": [],
+ "export": {
+  "kind": "bigym2",
+  "n_actions": 693,
+  "control_steps_replayed": 693,
+  "control_hz": 50.0,
+  "every": 2,
+  "frames": 348,
+  "mujoco": "3.8.1",
+  "n_geoms": 106
+ },
+ "render": {
+  "render_s": 1208.7,
+  "frames": 348,
+  "device": "GPU OptiX",
+  "look": "outdoor",
+  "min_spp": 64,
+  "max_spp": 128,
+  "w": 2560,
+  "h": 1440,
+  "blender": "4.2.23 LTS",
+  "orbit_deg": 10.0,
+  "split": "two GPUs (RTX 4090), frames split in halves"
+ },
+ "encode": "master: libx264 preset slow CRF 16, yuv420p, bt709, 30 fps, +faststart, no audio",
+ "speed_note": "uniform 1.2x speed-up (every 2nd 50 Hz control step at 30 fps)",
+ "replay_check_note": "the verifier of this privileged task records the verdict but no final hash; recorded.state_sha is the hash the agent's own two final CLI replays of the submitted trajectory.npz printed (agent log)",
+ "files": {
+  "video.mp4": {
+   "codec": "h264",
+   "profile": "High",
+   "pix_fmt": "yuv420p",
+   "width": 2560,
+   "height": 1440,
+   "fps": "30/1",
+   "frames": 408,
+   "duration_s": 13.6,
+   "bytes": 8482420,
+   "audio": false,
+   "faststart": true,
+   "sha256": "0537ee8ce860ed1fe64034647abed9ad049b4bd0b37c322517e10281914d5b92",
+   "note": "stream copy of the master (remuxed with +faststart)"
+  },
+  "web.mp4": {
+   "codec": "h264",
+   "profile": "High",
+   "pix_fmt": "yuv420p",
+   "width": 1920,
+   "height": 1080,
+   "fps": "30/1",
+   "frames": 408,
+   "duration_s": 13.6,
+   "bytes": 2900603,
+   "audio": false,
+   "faststart": true,
+   "crf": 20,
+   "sha256": "c09b94ac482cef4cfb8180dd83d1056471b23f928a9e8b861b3e3d93c0572e47"
+  },
+  "poster.jpg": {
+   "bytes": 24430,
+   "size": "1280x720",
+   "from": "master frame at t=13.55 s",
+   "sha256": "fb2c6b2c251a101751bdd9258fb003200db445bb9bd4a96f8e7237250ed1125a"
+  }
+ }
+}

blog/bigym2-dishwasher-close-i00-privileged--nfkkn2w/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c09b94ac482cef4cfb8180dd83d1056471b23f928a9e8b861b3e3d93c0572e47
+size 2900603

blog/humanoidbench-manip-door-i00-standard--cfxzsqf/meta.json

@@ -0,0 +1,153 @@
+{
+ "slug": "humanoidbench-manip-door-i00-standard--cfxzsqf",
+ "status": "rendered",
+ "task": "humanoidbench-manip-door-i00-standard",
+ "bench": "humanoidbench",
+ "mode": "standard",
+ "run": "claude_code-opus_5_5-high",
+ "outcome": "failed",
+ "card": "Claude Opus 5.5, standard, failed",
+ "source_trial": "claude_code/opus_5_5/high/claude_code-opus_5_5-high-humanoidbench-manip-door-i00-standard/humanoidbench-manip-door-i00-sta__cfxZsQf",
+ "what": "re-rendered: kinematic playback of this trial's own replay (its recorded actions replayed from the task's frozen instance with the task's own tooling, physics unchanged) in Blender Cycles",
+ "renderer": "Blender 4.2.23 LTS LTS Cycles path tracing (GPU OptiX, RTX 4090), adaptive sampling (threshold 0.01, 64-128 spp), OpenImageDenoise with albedo + normal passes; kinematic playback of the exact replay poses (the simulator's own compiled meshes, primitives and colours); daylight: a sun lamp (2 degree disc) and a Nishita sky (no sun disc), the scene's own floor colour on a large ground plane with a faint world-space variation; AgX view transform, Medium High Contrast look; camera director with a slow eased 10.0-degree orbit; 2560x1440, 30 fps, no text or overlays",
+ "rendering": "path-tracing",
+ "width": 2560,
+ "height": 1440,
+ "fps": 30,
+ "frames_rendered": 251,
+ "render_seconds": 865.3,
+ "seconds_per_frame": 3.45,
+ "time_range": {
+  "episode": "46 of 46 (the graded, last episode, after reset 45; journal rows 15223-16222)",
+  "from_step": 0,
+  "to_step": 1000,
+  "sim_t_from_s": 0.0,
+  "sim_t_to_s": 20.0,
+  "why": "the graded 46th episode, whole: 1000 control steps (20 s) from the frozen start to the time limit (failed): the robot reaches the handle within ~3 s, then stays crouched at the door with its hand on the handle; the door never opens"
+ },
+ "speed": 2.4,
+ "holds": {
+  "start_s": 0.5,
+  "end_s": 1.5,
+  "note": "first/last frame held, no extra simulation"
+ },
+ "replay_check": {
+  "reproduced": true,
+  "how": "hb: every recorded row replayed from the frozen instance with the task's own tooling (physics unchanged), the verdict + final state hash taken where the episode ends; a second independent replay (the tooling's own replay()) must agree; then compared with the trial's records",
+  "recorded": {
+   "success": false,
+   "state_sha": "4ac2f9d5bcb4",
+   "steps": 1000
+  },
+  "replayed": {
+   "success": false,
+   "state_sha": "4ac2f9d5bcb4",
+   "steps": 1000,
+   "episode_return": 179.995,
+   "success_bar": 600.0,
+   "ended": true
+  },
+  "plain_replay": {
+   "success": false,
+   "state_sha": "4ac2f9d5bcb4",
+   "steps": 1000,
+   "episode_return": 179.995
+  },
+  "matches": {
+   "success": true,
+   "state_sha": true,
+   "steps": true
+  },
+  "replays_agree": true,
+  "instance": {
+   "start_state_sha": "298e68548640",
+   "instance_sha": "34a77691b373",
+   "instance_sha_file": "34a77691b373"
+  }
+ },
+ "camera": {
+  "mode": "director",
+  "az_deg": 140.0,
+  "az_rule": "fixed",
+  "el_deg": 20.0,
+  "fovy_deg": 34.0,
+  "fill": 0.78,
+  "distance_m": [
+   4.177,
+   4.194
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 0.06,
+  "focus": {
+   "robot+extra": 251
+  },
+  "objects": {}
+ },
+ "visual_changes": [
+  "the room walls (left_wall, right_wall, behind_wall; alpha 0.3 in MuJoCo) are not drawn",
+  "the door frame (wall_door_connect_a/b, ceiling_door_connect; alpha 0.3 grey in MuJoCo) drawn opaque in the door colour"
+ ],
+ "export": {
+  "kind": "hb",
+  "n_actions": 1000,
+  "control_steps_replayed": 1000,
+  "control_hz": 50.0,
+  "every": 4,
+  "frames": 251,
+  "mujoco": "3.3.7",
+  "n_geoms": 31
+ },
+ "render": {
+  "render_s": 865.3,
+  "frames": 251,
+  "device": "GPU OptiX",
+  "look": "outdoor",
+  "min_spp": 64,
+  "max_spp": 128,
+  "w": 2560,
+  "h": 1440,
+  "blender": "4.2.23 LTS",
+  "orbit_deg": 10.0,
+  "split": "two GPUs (RTX 4090), frames split in halves"
+ },
+ "encode": "master: libx264 preset slow CRF 16, yuv420p, bt709, 30 fps, +faststart, no audio",
+ "speed_note": "uniform 2.4x speed-up (every 4th 50 Hz control step at 30 fps)",
+ "files": {
+  "video.mp4": {
+   "codec": "h264",
+   "profile": "High",
+   "pix_fmt": "yuv420p",
+   "width": 2560,
+   "height": 1440,
+   "fps": "30/1",
+   "frames": 311,
+   "duration_s": 10.367,
+   "bytes": 3925913,
+   "audio": false,
+   "faststart": true,
+   "sha256": "52eb93452c2e98c23d32ffc50d9058d5f1a75070e0177c45609214823c111e28",
+   "note": "stream copy of the master (remuxed with +faststart)"
+  },
+  "web.mp4": {
+   "codec": "h264",
+   "profile": "High",
+   "pix_fmt": "yuv420p",
+   "width": 1920,
+   "height": 1080,
+   "fps": "30/1",
+   "frames": 311,
+   "duration_s": 10.367,
+   "bytes": 1160652,
+   "audio": false,
+   "faststart": true,
+   "crf": 20,
+   "sha256": "f80478d9b9fd973de307fc0c9754fe54a7821384a5c6ba677073f235f2e8423b"
+  },
+  "poster.jpg": {
+   "bytes": 28199,
+   "size": "1280x720",
+   "from": "master frame at t=10.32 s",
+   "sha256": "bc9e8525a6d6a46e9da117494fda5a9988a0261a83cdff5ce155605072380988"
+  }
+ }
+}

blog/humanoidbench-manip-door-i00-standard--cfxzsqf/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f80478d9b9fd973de307fc0c9754fe54a7821384a5c6ba677073f235f2e8423b
+size 1160652