Datasets:
File size: 2,583 Bytes
8d8e9d9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 | {
"version": 1,
"scope": "Exact published retained frames; no changes to data or training splits.",
"spikes": {
"acceleration_floor_rad_s2": 40,
"jerk_floor_rad_s3": 800,
"robust_mad_multiplier": 12,
"minimum_flagged_frames": 3
},
"glitch": {
"median_window_seconds": 0.3,
"residual_rad": 0.3
},
"tracking": {
"direction_agreement_minimum": 0.7,
"command_range_rad": 0.05,
"minimum_scored_differences": 10,
"lag_search_seconds": 1,
"command_speed_gate_rad_s": 0.02,
"state_speed_gate_rad_s": 0.002,
"smoothing": "5-frame triangular [1,2,3,2,1]/9; shortest-arc unwrapped angles",
"note": "Score commanded motion even if state is frozen. Copied commands are untestable."
},
"reversed_lag": {
"minimum_correlation": 0.8,
"minimum_improvement_over_zero": 0.02,
"minimum_lead_frames": 2,
"exclude_search_boundary": true
},
"idle": {
"joint_speed_rad_s": 0.02,
"gripper_speed_per_s": 0.02,
"maximum_end_trim_fraction": 0.7,
"note": "Arm/gripper motion only; no trimming applied."
},
"workspace": {
"training_episodes_per_robot": 2048,
"samples_per_episode": 16,
"quantiles": [
0.001,
0.999
],
"padding_m": 0.1,
"minimum_outside_frames": 3,
"note": "Statistical review flag, not a robot safety envelope. Fitted on training split only."
},
"video": {
"stall_fraction": 0.2,
"black_luma_threshold": 32,
"black_pixel_percent": 98,
"note": "Every decoded full-resolution camera frame; exact adjacent YUV frame hashes for stalls; blackframe luma pixels below 32 in at least 98% of view."
},
"native_fk": {
"position_error_m": 0.0022,
"rotation_error_rad": 0.05,
"note": "Compare nominal FK to native recorded state poses only. RoboMIND native fields are not the published FK targets. No FK-fitted rotation correction."
},
"duplicates": "Match robot, fps, frame count, exact joint binary SHA256, EEF array content SHA256, and all decoded camera-frame sequence SHA256s. Choose canonical training member first. Metadata/instructions ignored; no removal.",
"limitations": [
"Not an exact reproduction of an unspecified S1/S2/S3 implementation.",
"Checks flag candidates for review; valid rare motion, stationary views and tasks can trigger them.",
"Audits retained processed frames; anomalies exclusively in source frames dropped during export are out of scope.",
"AgiBot has no independently verified FK model; physical calibration and image/pose reprojection are not certified."
]
}
|