nullPointerExcept1on/realworld_go2_task0_clean_smolvla

This dataset was created using LeRobot.

Dataset Structure

meta/info.json:

{
    "codebase_version": "v3.0",
    "fps": 10,
    "features": {
        "observation.images.front": {
            "dtype": "video",
            "shape": [
                480,
                640,
                3
            ],
            "names": [
                "height",
                "width",
                "channels"
            ],
            "info": {
                "is_depth_map": false,
                "video.height": 480,
                "video.width": 640,
                "video.codec": "av1",
                "video.pix_fmt": "yuv420p",
                "video.fps": 10,
                "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": {}
            }
        },
        "observation.state": {
            "dtype": "float32",
            "shape": [
                1
            ],
            "names": [
                "move"
            ]
        },
        "action": {
            "dtype": "float32",
            "shape": [
                3
            ],
            "names": [
                "vx",
                "vy",
                "vyaw"
            ]
        },
        "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": 50,
    "total_frames": 9325,
    "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": "unitree_go2",
    "splits": {
        "train": "0:50"
    }
}

Citation

BibTeX:

[More Information Needed]

Notes

Cleaned action labels of nullPointerExcept1on/realworld_go2_task0_smolvla. task0 was driven with the Go2's own remote, so the action is the measured body velocity, and that signal carries a strong gait-phase ripple plus frequent single-sample dropouts to exactly 0 (22% of frames read 0, mean frame-to-frame |dvx| 0.53 m/s, outliers down to -3.17 m/s) — noise a teleoperated command does not have, which a policy otherwise learns to reproduce as a stutter. Here the action is re-derived from observation.state with a median(7) + mean(3) filter (0.7 s) and clipped, then the standing-still tail of each episode is trimmed: vx==0 frames 22% -> 9%, mean |dvx| 0.53 -> 0.05, vx range [-3.17, 1.83] -> [0.00, 1.46]. Real starts and stops are preserved. observation.state=[move] is fixed to 1.0. Train with --rename_map='{"observation.images.front": "observation.images.camera1"}'. 50 episodes, 9,325 frames (2026-09-14).