felsager/community_dataset_v3_ee_smolvla

This dataset was created using LeRobot.

Revisions

Revision	Contents
v3.0 (default)	Cleaned and relabelled; each frame's task text is one of several paraphrased wordings of its label
cleaned-no-paraphrases	The same episodes, frames and labels, with exactly one task text (the label) per episode
pre-label-fix-2026-10-08	The original conversion, before the label review

You are reading the v3.0 revision. Load another one with --dataset.revision=<name>.

Label review (2026-10-16)

• Task labels: wrong or vague task texts were corrected per dataset and per task (and per episode range where a dataset mixes tasks); punctuation and capitalisation are normalised. Labels that turned out to be the same task were merged. 713 labels in total.
• Paraphrases (v3.0 only): every frame carries one of 3 to 7 wordings of its label (pick up / grasp / grab / take, a / the, in / into, ...), picked by a hash of the frame index, so one episode is seen under several wordings. The first frame of every episode carries the label itself, so the label leads each episode's task list: LeRobot's train/eval split (last episodes of each first task) then groups episodes by label, as in cleaned-no-paraphrases, and holds out the same episodes. meta/tasks.parquet has a canonical column with each wording's label.
• Removed: 94 of the 1,086 source datasets (the reason is under exclusion in meta/merge_sources.json) and 283 episodes of the kept datasets (outliers and review exclusions).
• Idle starts trimmed: 2,092 episodes lose their still frames before the first motion (1 s is kept); 138,132 frames in total.
• Result: 36,057 episodes, 18,812,021 frames. Per source dataset, meta/merge_sources.json lists the labels, dropped episodes and trimmed frames under label_fix.
• For a few hours on 2026-10-16 the default was an intermediate version with one wording per episode (commit c11d0984); it is superseded by v3.0.

Dataset Structure

meta/info.json:

{
    "codebase_version": "v3.0",
    "robot_type": "so100_so101_ee",
    "fps": 30,
    "total_frames": 18812021,
    "total_episodes": 36057,
    "total_tasks": 2882,
    "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",
    "splits": {
        "train": "0:36057"
    },
    "features": {
        "observation.images.top": {
            "dtype": "video",
            "shape": [
                480,
                640,
                3
            ],
            "names": [
                "height",
                "width",
                "channels"
            ],
            "info": {
                "video.height": 480,
                "video.width": 640,
                "video.fps": 30,
                "video.channels": 3,
                "video.pix_fmt": "yuv420p",
                "video.codec": null,
                "video.is_depth_map": false,
                "has_audio": false
            }
        },
        "observation.images.wrist": {
            "dtype": "video",
            "shape": [
                480,
                640,
                3
            ],
            "names": [
                "height",
                "width",
                "channels"
            ],
            "info": {
                "video.height": 480,
                "video.width": 640,
                "video.fps": 30,
                "video.channels": 3,
                "video.pix_fmt": "yuv420p",
                "video.codec": null,
                "video.is_depth_map": false,
                "has_audio": false
            }
        },
        "observation.images.side": {
            "dtype": "video",
            "shape": [
                480,
                640,
                3
            ],
            "names": [
                "height",
                "width",
                "channels"
            ],
            "info": {
                "video.height": 480,
                "video.width": 640,
                "video.fps": 30,
                "video.channels": 3,
                "video.pix_fmt": "yuv420p",
                "video.codec": null,
                "video.is_depth_map": false,
                "has_audio": false
            }
        },
        "observation.state": {
            "dtype": "float32",
            "shape": [
                8
            ],
            "names": [
                "ee.x",
                "ee.y",
                "ee.z",
                "ee.qx",
                "ee.qy",
                "ee.qz",
                "ee.qw",
                "ee.gripper"
            ],
            "fps": 30
        },
        "action": {
            "dtype": "float32",
            "shape": [
                8
            ],
            "names": [
                "ee.x",
                "ee.y",
                "ee.z",
                "ee.qx",
                "ee.qy",
                "ee.qz",
                "ee.qw",
                "ee.gripper"
            ],
            "fps": 30
        },
        "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
        },
        "observation.images.top_padding_mask": {
            "dtype": "bool",
            "shape": [
                1
            ],
            "names": null
        },
        "observation.images.wrist_padding_mask": {
            "dtype": "bool",
            "shape": [
                1
            ],
            "names": null
        },
        "observation.images.side_padding_mask": {
            "dtype": "bool",
            "shape": [
                1
            ],
            "names": null
        },
        "source_dataset_index": {
            "dtype": "int64",
            "shape": [
                1
            ],
            "names": null
        }
    }
}

Citation

BibTeX:

[More Information Needed]