azorematter/dice_white_pnp_und_500_b02

This dataset was created using LeRobot.

Dataset Description

FANUC CRX-5iA dice pick-and-place demonstrations recorded on the cell by the scripted servo (the white-block collector): 170 episodes, 110184 frames at 30 fps, 4 cameras (gripper, cam0, cam1, cam2) at 640x480. Task: pick the dice up and place it on the empty white block. Every episode was saved only after the collector verified the dice seated on its block.

Frame geometry

Frames are UNDISTORTED (mode keep): remapped with the factory lens calibration (OpenCV rational model, all 14 terms, K kept) before encoding. Per-camera K, K_new and coefficients: meta/undistort.json. A policy trained here must be fed undistorted frames at inference (fanuc_control.undistort, mode keep).

State and action

observation.state and action are 13 wide: J1, J2, J3, J4, J5, J6, X, Y, Z, W, P, R, grip. J1..J6 are joint degrees, X/Y/Z the tool tip in mm (world frame, robot base at the origin), W/P/R the FANUC tool orientation in degrees, grip 1.0 = OPEN and 0.0 = closed. action[t] is observation.state[t+1] (the last frame repeats), i.e. absolute targets, not deltas. Poses come from the controller's 125 Hz status stream sampled at each frame's camera capture stamp - nothing is interpolated; episodes with a pose hole under a frame, a close more than 3 mm above the descent bottom, or a stale camera were discarded at collection time.

Collection metrics

Collection metrics

170 episodes, 110184 frames at 30 fps, 3673 s of demonstration. Generated 2026-09-06T20:03:59Z by fanuc_control.data_utils.metrics v1. Definitions and thresholds: docs/metrics.md.

Pose channel	value	note
identical consecutive poses	0.0%	old corpus 18% (cache frozen during moves)
exact-linear ramp frames	23.6%	old corpus 79% (interpolated)
stationary frames (joint speed < 0.5 deg/s)	39.4%	RMI stop-and-go corpus 39%
stationary run length p50 / p90 / max (frames)	10 / 50 / 89	a run >= 30 is a whole ACT chunk
frames inside stationary runs >= 30	21.1%	hover fixed point; was 21%
'stay put' 30-row action chunks	15.5%	label ambiguity at rest; was 15%
joint speed p50 / p90 / max (deg/s)	4.0 / 49.7 / 79.6

Grasp geometry	value	note
episodes with a close / release	170 / 170
close above descent bottom p50 / p90 / max (mm)	0.05 / 0.09 / 0.13	old corpus 35.1 mm in 99/99; limit 3
frames at depth before the close p50 / p90 / max	12 / 13 / 14	the settle before grip
z at close p50 / p90 / max (mm)	-126.5 / -126.5 / -126.5
z at release p50 / p90 / max (mm)	-125.0 / -125.0 / -125.0

Placement (x, y at the close)	value
closes measured	170
x range (mm)	[176.19, 483.15]
y range (mm)	[329.15, 535.49]
inside the randomiser region	95% (outside: [[415.7, 330.8], [414.6, 331.6], [415.6, 330.2], [415.3, 330.6], [414.9, 329.7], [414.9, 330.2], [415.3, 330.1], [414.1, 329.1], [411.8, 329.5]])

Sync and yield (collection_stats.jsonl)	value	note
yield (kept / attempts)	170 / 172 = 99%
discards	{'unknown': 1, 'camera 3 handed out a stale frame: 1942 ms behind camera 0 at frame 628 (limit 100 ms)': 1}
repaired frame ticks (total, episodes)	3, 1	duplicated frames over a sampler hole
pose cadence p50 / p90 / max (ms)	8.00 / 8.00 / 8.00	8.00 nominal
pose worst gap p50 / p90 / max (ms)	8.0 / 8.0 / 32.0	recorder refuses > 40 under a frame
pose sequence gaps	12
pose receive jitter p50 / p90 / max (ms)	13.1 / 19.4 / 29.1	absorbed by the controller clock
close above depth p50 / p90 / max (mm)	0.05 / 0.09 / 0.13	recorder refuses > 3
depth lead before close p50 / p90 / max (ms)	400 / 433 / 466
inter-camera skew p50 / p90 / max (ms)	49 / 50 / 51	recorder refuses > 100
wrist camera vs pose lag median /	median	/ max

Camera	files	frames	rows	match	brightness median	clipped
observation.images.gripper	4	110184	110184	True	46	0.08%
observation.images.cam0	7	110184	110184	True	127	0.94%
observation.images.cam1	7	110184	110184	True	9	0.55%
observation.images.cam2	7	110184	110184	True	8	1.43%

Flags: stationary frames 39.4% > 30.0%; frames in stationary runs >= 30 21.1% > 10.0%; stay-put 30-row chunks 15.5% > 5.0%; placements outside the randomiser region: 9; pose stream sequence gaps: 12; observation.images.cam1: median brightness 9; observation.images.cam2: median brightness 8

Dataset Structure

meta/info.json:

{
    "codebase_version": "v3.0",
    "fps": 30,
    "features": {
        "observation.state": {
            "dtype": "float32",
            "shape": [
                13
            ],
            "names": [
                "J1",
                "J2",
                "J3",
                "J4",
                "J5",
                "J6",
                "x",
                "y",
                "z",
                "W",
                "P",
                "R",
                "gripper"
            ]
        },
        "action": {
            "dtype": "float32",
            "shape": [
                13
            ],
            "names": [
                "J1",
                "J2",
                "J3",
                "J4",
                "J5",
                "J6",
                "x",
                "y",
                "z",
                "W",
                "P",
                "R",
                "gripper"
            ]
        },
        "observation.images.gripper": {
            "dtype": "video",
            "shape": [
                480,
                640,
                3
            ],
            "names": [
                "height",
                "width",
                "channels"
            ],
            "info": {
                "video.height": 480,
                "video.width": 640,
                "video.codec": "av1",
                "video.pix_fmt": "yuv420p",
                "video.fps": 30,
                "video.channels": 3,
                "has_audio": false,
                "video.g": 2,
                "video.crf": 30,
                "video.preset": 12,
                "video.fast_decode": 0,
                "video.video_backend": "pyav",
                "video.extra_options": {},
                "is_depth_map": false
            }
        },
        "observation.images.cam0": {
            "dtype": "video",
            "shape": [
                480,
                640,
                3
            ],
            "names": [
                "height",
                "width",
                "channels"
            ],
            "info": {
                "video.height": 480,
                "video.width": 640,
                "video.codec": "av1",
                "video.pix_fmt": "yuv420p",
                "video.fps": 30,
                "video.channels": 3,
                "has_audio": false,
                "video.g": 2,
                "video.crf": 30,
                "video.preset": 12,
                "video.fast_decode": 0,
                "video.video_backend": "pyav",
                "video.extra_options": {},
                "is_depth_map": false
            }
        },
        "observation.images.cam1": {
            "dtype": "video",
            "shape": [
                480,
                640,
                3
            ],
            "names": [
                "height",
                "width",
                "channels"
            ],
            "info": {
                "video.height": 480,
                "video.width": 640,
                "video.codec": "av1",
                "video.pix_fmt": "yuv420p",
                "video.fps": 30,
                "video.channels": 3,
                "has_audio": false,
                "video.g": 2,
                "video.crf": 30,
                "video.preset": 12,
                "video.fast_decode": 0,
                "video.video_backend": "pyav",
                "video.extra_options": {},
                "is_depth_map": false
            }
        },
        "observation.images.cam2": {
            "dtype": "video",
            "shape": [
                480,
                640,
                3
            ],
            "names": [
                "height",
                "width",
                "channels"
            ],
            "info": {
                "video.height": 480,
                "video.width": 640,
                "video.codec": "av1",
                "video.pix_fmt": "yuv420p",
                "video.fps": 30,
                "video.channels": 3,
                "has_audio": false,
                "video.g": 2,
                "video.crf": 30,
                "video.preset": 12,
                "video.fast_decode": 0,
                "video.video_backend": "pyav",
                "video.extra_options": {},
                "is_depth_map": false
            }
        },
        "timestamp": {
            "dtype": "float32",
            "shape": [
                1
            ],
            "names": null
        },
        "frame_index": {
            "dtype": "int64",
            "shape": [
                1
            ],
            "names": null
        },
        "episode_index": {
            "dtype": "int64",
            "shape": [
                1
            ],
            "names": null
        },
        "index": {
            "dtype": "int64",
            "shape": [
                1
            ],
            "names": null
        },
        "task_index": {
            "dtype": "int64",
            "shape": [
                1
            ],
            "names": null
        }
    },
    "total_episodes": 170,
    "total_frames": 110184,
    "total_tasks": 1,
    "chunks_size": 1000,
    "data_files_size_in_mb": 100,
    "video_files_size_in_mb": 200,
    "data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
    "video_path": "videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4",
    "robot_type": "fanuc_crx5ia",
    "splits": {
        "train": "0:170"
    }
}