Datasets:
blog: HQ case videos (1)
Browse files
blog/roboolympics-fun-catch-and-shoot-i00-standard-realtime--3eudcdx/meta.json
ADDED
|
@@ -0,0 +1,162 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"slug": "roboolympics-fun-catch-and-shoot-i00-standard-realtime--3eudcdx",
|
| 3 |
+
"status": "rendered",
|
| 4 |
+
"task": "roboolympics-fun-catch-and-shoot-i00-standard-realtime",
|
| 5 |
+
"bench": "roboolympics-fun",
|
| 6 |
+
"mode": "standard-realtime",
|
| 7 |
+
"run": "codex-gpt6_astra-high",
|
| 8 |
+
"outcome": "failed",
|
| 9 |
+
"card": "GPT-6 Astra, standard-realtime, failed; catches but shots fall short; 119 min run",
|
| 10 |
+
"source_trial": "codex/gpt6_astra/high/codex-gpt6_astra-high-roboolympics-fun-catch-and-shoot-i00-standard-realtime/roboolympics-catch-and-shoot-i00__3euDcdx",
|
| 11 |
+
"what": "re-rendered: kinematic playback of this trial's own episode 5 (its recorded control steps from the sidecar's journal, replayed from the task's frozen instance with the task's own tooling, physics unchanged) in Blender Cycles",
|
| 12 |
+
"renderer": "Blender 4.2.23 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 MuJoCo replay poses (MuJoCo's own compiled meshes, primitives and colours); daylight: a sun lamp (2 degree disc) and a Nishita sky (no sun disc), the venue's own ground colour on a large ground plane with a faint world-space variation; AgX view transform, Medium High Contrast look; the suite masters' settings (outdoor look, sun 50 degrees from the camera); a fixed side-on camera (azimuth -100 degrees, elevation 12 degrees) framing the robot, the ball's whole flight and the basket; a slow eased 10-degree orbit; 2560x1440, 30 fps, no text or overlays",
|
| 13 |
+
"rendering": "path-tracing",
|
| 14 |
+
"width": 2560,
|
| 15 |
+
"height": 1440,
|
| 16 |
+
"fps": 30,
|
| 17 |
+
"frames_rendered": 210,
|
| 18 |
+
"render_seconds": 597.1,
|
| 19 |
+
"seconds_per_frame": 2.84,
|
| 20 |
+
"time_range": {
|
| 21 |
+
"episode": "5 of 11 (journal rows 2503-3128; the trial's standard-realtime run had 11 episodes, 10 resets)",
|
| 22 |
+
"from_step": 0,
|
| 23 |
+
"to_step": 625,
|
| 24 |
+
"sim_t_from_s": 0.0,
|
| 25 |
+
"sim_t_to_s": 6.25,
|
| 26 |
+
"why": "the pass, the catch and a shot, whole: the clearest of the six episodes (5-10) in which it caught the pass and shot; the ball peaks at 3.13 m but passes 3.1 m from the ring at its closest and lands short (episode verdict: 'the ball touched the floor', stages catch + release of 3). The graded last episode (11) caught the pass and then fell before shooting."
|
| 27 |
+
},
|
| 28 |
+
"speed": 0.9,
|
| 29 |
+
"holds": {
|
| 30 |
+
"start_s": 0.5,
|
| 31 |
+
"end_s": 1.5,
|
| 32 |
+
"note": "first/last frame held, no extra simulation"
|
| 33 |
+
},
|
| 34 |
+
"replay_check": {
|
| 35 |
+
"reproduced": true,
|
| 36 |
+
"how": "the trial's journal (artifacts/data/trajectory_live.f64, every executed control step) split at the recorded reset/end events; each episode replayed from the frozen instance with the task's own tooling (rcb_rfo.Episode) and its end state hash compared with the sidecar's recorded `end` event; for the rendered episode a second plain rcb_rfo.replay() must agree",
|
| 37 |
+
"rendered_episode": {
|
| 38 |
+
"episode": 5,
|
| 39 |
+
"recorded_state_sha": "3a7db7b11cc4dec7bce2c9190d59ab1cd4d85807595188c328804767df968f96",
|
| 40 |
+
"replayed_state_sha": "3a7db7b11cc4dec7bce2c9190d59ab1cd4d85807595188c328804767df968f96",
|
| 41 |
+
"sha_match": true,
|
| 42 |
+
"replayed_verdict": {
|
| 43 |
+
"success": false,
|
| 44 |
+
"failed": "the ball touched the floor",
|
| 45 |
+
"stages": [
|
| 46 |
+
"catch",
|
| 47 |
+
"release"
|
| 48 |
+
],
|
| 49 |
+
"stages_total": 3,
|
| 50 |
+
"progress": 0.6667,
|
| 51 |
+
"metrics": {
|
| 52 |
+
"ball_height_m": 0.119
|
| 53 |
+
},
|
| 54 |
+
"time_s": 6.25
|
| 55 |
+
},
|
| 56 |
+
"plain_replay_agrees": true
|
| 57 |
+
},
|
| 58 |
+
"graded_episode": {
|
| 59 |
+
"episode": 11,
|
| 60 |
+
"recorded_state_sha (final.json)": "6f9fd8831c7481b27264823ec6411e48a861acfb86e691778b22c1d54233a0fd",
|
| 61 |
+
"replayed_state_sha": "6f9fd8831c7481b27264823ec6411e48a861acfb86e691778b22c1d54233a0fd",
|
| 62 |
+
"sha_match": true,
|
| 63 |
+
"replayed_verdict": {
|
| 64 |
+
"success": false,
|
| 65 |
+
"failed": "fell",
|
| 66 |
+
"stages": [
|
| 67 |
+
"catch"
|
| 68 |
+
],
|
| 69 |
+
"progress": 0.3333
|
| 70 |
+
},
|
| 71 |
+
"recorded_verdict": "failed (fell; stages: catch)"
|
| 72 |
+
},
|
| 73 |
+
"all_episodes_sha_match": true,
|
| 74 |
+
"episodes_checked": 11
|
| 75 |
+
},
|
| 76 |
+
"camera": {
|
| 77 |
+
"mode": "fixed",
|
| 78 |
+
"az_deg": -100.0,
|
| 79 |
+
"az_rule": "fixed",
|
| 80 |
+
"el_deg": 12.0,
|
| 81 |
+
"fovy_deg": 34.0,
|
| 82 |
+
"fill": 0.85,
|
| 83 |
+
"distance_m": [
|
| 84 |
+
9.239,
|
| 85 |
+
9.239
|
| 86 |
+
],
|
| 87 |
+
"robot_root": "pelvis",
|
| 88 |
+
"travel_m": 0.148,
|
| 89 |
+
"focus": {},
|
| 90 |
+
"objects": {
|
| 91 |
+
"ball": 5.19
|
| 92 |
+
}
|
| 93 |
+
},
|
| 94 |
+
"visual_changes": [
|
| 95 |
+
"daylight sky and a plain ground plane in the venue's ground colour instead of the venue's painted sky, hills, trees, grandstands and perimeter boards (drawn-only scenery), as in the suite's masters"
|
| 96 |
+
],
|
| 97 |
+
"export": {
|
| 98 |
+
"kind": "rfo",
|
| 99 |
+
"n_actions": 625,
|
| 100 |
+
"control_steps_replayed": 625,
|
| 101 |
+
"control_hz": 100.0,
|
| 102 |
+
"every": 3,
|
| 103 |
+
"frames": 210,
|
| 104 |
+
"mujoco": "3.3.7",
|
| 105 |
+
"n_geoms": 278
|
| 106 |
+
},
|
| 107 |
+
"render": {
|
| 108 |
+
"render_s": 597.1,
|
| 109 |
+
"frames": 210,
|
| 110 |
+
"device": "GPU OptiX",
|
| 111 |
+
"look": "outdoor",
|
| 112 |
+
"min_spp": 64,
|
| 113 |
+
"max_spp": 128,
|
| 114 |
+
"w": 2560,
|
| 115 |
+
"h": 1440,
|
| 116 |
+
"blender": "4.2.23 LTS",
|
| 117 |
+
"orbit_deg": 10.0,
|
| 118 |
+
"gpus": 3,
|
| 119 |
+
"wall_s_max_slot": 202
|
| 120 |
+
},
|
| 121 |
+
"encode": "master: libx264 preset slow CRF 16, yuv420p, bt709, 30 fps, +faststart, no audio",
|
| 122 |
+
"speed_note": "uniform 0.9x (every 3rd control step at 100 Hz, 30 fps): practically real time",
|
| 123 |
+
"story_check": "shots fall short: confirmed (episodes 5-10 each catch and release; the ball's closest approach to the ring is 2.7-5.4 m and it lands short). The verdict episode (11) caught the pass and fell before shooting.",
|
| 124 |
+
"files": {
|
| 125 |
+
"video.mp4": {
|
| 126 |
+
"codec": "h264",
|
| 127 |
+
"profile": "High",
|
| 128 |
+
"pix_fmt": "yuv420p",
|
| 129 |
+
"width": 2560,
|
| 130 |
+
"height": 1440,
|
| 131 |
+
"fps": "30/1",
|
| 132 |
+
"frames": 270,
|
| 133 |
+
"duration_s": 9.0,
|
| 134 |
+
"bytes": 4405206,
|
| 135 |
+
"audio": false,
|
| 136 |
+
"faststart": true,
|
| 137 |
+
"sha256": "fdae3bcc6d9eb06eb21d3f78405c2b3d94806da301f33571ad87abf334d7a30a",
|
| 138 |
+
"note": "stream copy of the master (remuxed with +faststart)"
|
| 139 |
+
},
|
| 140 |
+
"web.mp4": {
|
| 141 |
+
"codec": "h264",
|
| 142 |
+
"profile": "High",
|
| 143 |
+
"pix_fmt": "yuv420p",
|
| 144 |
+
"width": 1920,
|
| 145 |
+
"height": 1080,
|
| 146 |
+
"fps": "30/1",
|
| 147 |
+
"frames": 270,
|
| 148 |
+
"duration_s": 9.0,
|
| 149 |
+
"bytes": 1276419,
|
| 150 |
+
"audio": false,
|
| 151 |
+
"faststart": true,
|
| 152 |
+
"crf": 20,
|
| 153 |
+
"sha256": "321b06abab3e72b32c56a870276836f9535c334cb984af49a042f3d66d554b55"
|
| 154 |
+
},
|
| 155 |
+
"poster.jpg": {
|
| 156 |
+
"bytes": 37797,
|
| 157 |
+
"size": "1280x720",
|
| 158 |
+
"from": "master frame at t=7.00 s",
|
| 159 |
+
"sha256": "979a6d406f7aac903306fe4ab686efc725f65a8d60e6963bec09a65df98a191e"
|
| 160 |
+
}
|
| 161 |
+
}
|
| 162 |
+
}
|
blog/roboolympics-fun-catch-and-shoot-i00-standard-realtime--3eudcdx/poster.jpg
ADDED
|
Git LFS Details
|
blog/roboolympics-fun-catch-and-shoot-i00-standard-realtime--3eudcdx/video.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:321b06abab3e72b32c56a870276836f9535c334cb984af49a042f3d66d554b55
|
| 3 |
+
size 1276419
|