| { |
| "@context": { |
| "@language": "en", |
| "@vocab": "https://schema.org/", |
| "citeAs": "cr:citeAs", |
| "column": "cr:column", |
| "conformsTo": "dct:conformsTo", |
| "cr": "http://mlcommons.org/croissant/", |
| "rai": "http://mlcommons.org/croissant/RAI/", |
| "data": { |
| "@id": "cr:data", |
| "@type": "@json" |
| }, |
| "dataType": { |
| "@id": "cr:dataType", |
| "@type": "@vocab" |
| }, |
| "dct": "http://purl.org/dc/terms/", |
| "prov": "http://www.w3.org/ns/prov#", |
| "equivalentProperty": "cr:equivalentProperty", |
| "examples": { |
| "@id": "cr:examples", |
| "@type": "@json" |
| }, |
| "extract": "cr:extract", |
| "field": "cr:field", |
| "fileProperty": "cr:fileProperty", |
| "fileObject": "cr:fileObject", |
| "fileSet": "cr:fileSet", |
| "format": "cr:format", |
| "includes": "cr:includes", |
| "isLiveDataset": "cr:isLiveDataset", |
| "jsonPath": "cr:jsonPath", |
| "key": "cr:key", |
| "md5": "cr:md5", |
| "parentField": "cr:parentField", |
| "path": "cr:path", |
| "recordSet": "cr:recordSet", |
| "references": "cr:references", |
| "regex": "cr:regex", |
| "repeated": "cr:repeated", |
| "replace": "cr:replace", |
| "samplingRate": "cr:samplingRate", |
| "sc": "https://schema.org/", |
| "separator": "cr:separator", |
| "source": "cr:source", |
| "subField": "cr:subField", |
| "transform": "cr:transform", |
| "wd": "https://www.wikidata.org/wiki/" |
| }, |
| "@type": "sc:Dataset", |
| "name": "RoboPro", |
| "alternateName": "roboreal_all_80tasks", |
| "description": "RoboPro is an 80-task bimanual manipulation dataset collected on a dual-arm Aloha-Agilex platform. It contains 16,000 expert demonstration episodes (~3.73M frames at 50 Hz) recorded from three RGB cameras (top + two wrist) and synchronised 14-DoF joint commands. Each task includes a clean variant and ten cluttered-scene variants (d6-d15) with progressively added distractor objects. The dataset is released in LeRobot v2.1 format (Parquet for proprioception, MP4/H.264 for video) and is the training corpus used for the RoboPro behavior models.", |
| "conformsTo": "http://mlcommons.org/croissant/1.0", |
| "license": "https://creativecommons.org/licenses/by/4.0/", |
| "url": "https://huggingface.co/datasets/Hoshipu/RoboPro", |
| "version": "1.1.0", |
| "datePublished": "2026-05-01", |
| "creator": { |
| "@type": "Person", |
| "name": "Zhiyuan Li", |
| "url": "https://huggingface.co/Hoshipu" |
| }, |
| "publisher": { |
| "@type": "Organization", |
| "name": "Hoshipu" |
| }, |
| "keywords": [ |
| "robotics", |
| "imitation learning", |
| "bimanual manipulation", |
| "Aloha-Agilex", |
| "LeRobot", |
| "robot learning" |
| ], |
| "citeAs": "@dataset{roboPro2026, title={RoboPro: 80-Task Bimanual Manipulation Demonstrations on Aloha-Agilex}, author={Li, Zhiyuan}, year={2026}, url={https://huggingface.co/datasets/Hoshipu/RoboPro}}", |
| "rai:dataCollection": "Demonstrations were teleoperated by trained human operators on a dual-arm Aloha-Agilex platform across 80 distinct manipulation tasks (kitchen, office, and household scenarios). Each episode was logged at 50 Hz with three RGB cameras (one ceiling-mounted, two wrist-mounted) and the 14-DoF joint state/command stream. Episodes are organised per task into a clean condition (uncluttered scene) and ten cluttered conditions (d6…d15) with progressively added distractor objects.", |
| "rai:dataCollectionType": [ |
| "Human-operated teleoperation", |
| "Sensor recording (RGB, joint encoders)" |
| ], |
| "rai:dataCollectionTimeframe": "2025-08 – 2026-04", |
| "rai:dataAnnotationProtocol": "Each episode is paired with one of 1,622 free-form natural-language task descriptions (paraphrases of 80 base tasks) derived from the originating task definition; no per-frame human annotation is performed.", |
| "rai:dataPreprocessingProtocol": [ |
| "Videos are encoded as H.264 yuv420p MP4 at 480x640 with GOP=2 to enable fast random-access frame decoding for training.", |
| "Joint angles and gripper widths are stored as raw float32 in the original Aloha order; left-arm joints are flipped at training time per pi-zero/pi-0.5 convention.", |
| "Episode and frame indices are globally unique across the dataset." |
| ], |
| "rai:dataLimitations": [ |
| "All demonstrations were recorded on a single Aloha-Agilex hardware unit in a fixed laboratory environment; lighting, camera intrinsics, and table geometry do not vary across episodes.", |
| "Tasks are limited to short-horizon (<10 s) tabletop manipulation; no long-horizon, navigation, or contact-rich tasks (e.g., insertion, peg-in-hole) are included.", |
| "Object instances are drawn from a fixed inventory of household and office props; the dataset does not span an open-vocabulary distribution of real-world clutter.", |
| "One source HDF5 (study/empty_box/clean/episode80.hdf5) was unrecoverable at conversion time, so episode 15999 in the released LeRobot dataset is a byte-for-byte duplicate of an existing empty_box/clean episode (global episode_index 12000) to maintain a 16,000-episode count. Users training on the dataset should be aware that one demonstration is repeated." |
| ], |
| "rai:hasSyntheticData": false, |
| "rai:dataBiases": [ |
| "Demonstrations reflect the kinematic preferences and strategies of a small number of teleoperators and may not represent task-optimal trajectories.", |
| "Object placement, lighting, and viewpoint are fixed per task variant — models trained on this data may overfit to the specific scene composition.", |
| "The clean/cluttered split follows a deterministic protocol (one clean condition vs. ten cluttered d6–d15 conditions); evaluating generalisation requires held-out scenes outside this distribution." |
| ], |
| "rai:dataUseCases": [ |
| "Training behavior-cloning policies (e.g., π0/π0.5, ACT, Diffusion Policy) for bimanual manipulation.", |
| "Studying the effect of scene clutter on imitation-learning performance via the dataset's clean-only / mixed / cluttered-only ablation splits.", |
| "Benchmarking vision-language-action (VLA) models with multiple natural-language paraphrases per task." |
| ], |
| "rai:dataSocialImpact": "The dataset is intended to advance research in robot manipulation policies. It does not contain depictions of identifiable persons, sensitive scenes, or harmful content. Risks are limited to those of any imitation-learning corpus (e.g., transferring suboptimal demonstrator behaviors to a deployed policy).", |
| "rai:personalSensitiveInformation": "None. No human faces, voices, biometric data, or other personal information is captured. All scenes consist of inanimate objects on a laboratory table.", |
| "rai:dataReleaseMaintenancePlan": "The dataset is hosted on Hugging Face under Hoshipu/RoboPro and is versioned via Hugging Face Git LFS. Bug fixes and additional task variants will be released as new tagged versions; the current release is v1.0.0.", |
| "prov:wasGeneratedBy": { |
| "@type": "prov:Activity", |
| "name": "Aloha-Agilex teleoperation data collection", |
| "description": "Trained human operators teleoperated a dual-arm Aloha-Agilex platform to perform 80 distinct manipulation tasks. For each task, demonstrations were collected in a clean condition and ten cluttered conditions (d6-d15). Each episode was recorded at 50 Hz with three RGB cameras (top + two wrist) and synchronised 14-DoF joint state/command streams. Trajectories were converted to the LeRobot v2.1 format (Parquet for proprioception, H.264 MP4 for video) using the convert_roboreal_fast.py pipeline.", |
| "prov:startedAtTime": "2025-08-01", |
| "prov:endedAtTime": "2026-04-30", |
| "prov:wasAssociatedWith": { |
| "@type": "prov:Agent", |
| "name": "Hoshipu", |
| "url": "https://huggingface.co/Hoshipu" |
| } |
| }, |
| "isLiveDataset": false, |
| "distribution": [ |
| { |
| "@type": "cr:FileObject", |
| "@id": "huggingface-repository", |
| "name": "huggingface-repository", |
| "description": "RoboPro dataset repository on the Hugging Face Hub.", |
| "contentUrl": "https://huggingface.co/datasets/Hoshipu/RoboPro", |
| "encodingFormat": "git+https", |
| "sha256": "main" |
| }, |
| { |
| "@type": "cr:FileObject", |
| "@id": "info-json", |
| "name": "info-json", |
| "description": "LeRobot dataset metadata (features, FPS, episode count).", |
| "containedIn": { |
| "@id": "huggingface-repository" |
| }, |
| "contentUrl": "lerobot/roboreal_all_80tasks/meta/info.json", |
| "encodingFormat": "application/json", |
| "sha256": "main" |
| }, |
| { |
| "@type": "cr:FileObject", |
| "@id": "tasks-jsonl", |
| "name": "tasks-jsonl", |
| "description": "Per-task natural-language descriptions (one JSON object per line).", |
| "containedIn": { |
| "@id": "huggingface-repository" |
| }, |
| "contentUrl": "lerobot/roboreal_all_80tasks/meta/tasks.jsonl", |
| "encodingFormat": "application/jsonlines", |
| "sha256": "main" |
| }, |
| { |
| "@type": "cr:FileObject", |
| "@id": "episodes-jsonl", |
| "name": "episodes-jsonl", |
| "description": "Per-episode metadata (length, task_index, etc.).", |
| "containedIn": { |
| "@id": "huggingface-repository" |
| }, |
| "contentUrl": "lerobot/roboreal_all_80tasks/meta/episodes.jsonl", |
| "encodingFormat": "application/jsonlines", |
| "sha256": "main" |
| }, |
| { |
| "@type": "cr:FileSet", |
| "@id": "parquet-files", |
| "name": "parquet-files", |
| "description": "One Parquet file per episode containing 50 Hz proprioception (observation.state, action, timestamp, frame_index, episode_index, index, task_index).", |
| "containedIn": { |
| "@id": "huggingface-repository" |
| }, |
| "encodingFormat": "application/x-parquet", |
| "includes": "lerobot/roboreal_all_80tasks/data/chunk-*/episode_*.parquet" |
| }, |
| { |
| "@type": "cr:FileSet", |
| "@id": "video-files", |
| "name": "video-files", |
| "description": "One MP4 (H.264, yuv420p, 480x640, 50 fps, GOP=2) per episode and camera. Camera keys: cam_high (ceiling), cam_left_wrist, cam_right_wrist.", |
| "containedIn": { |
| "@id": "huggingface-repository" |
| }, |
| "encodingFormat": "video/mp4", |
| "includes": "lerobot/roboreal_all_80tasks/videos/chunk-*/observation.images.*/episode_*.mp4" |
| } |
| ], |
| "recordSet": [ |
| { |
| "@type": "cr:RecordSet", |
| "@id": "tasks", |
| "name": "tasks", |
| "description": "Natural-language task descriptions, one per task_index.", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "tasks/task_index", |
| "name": "task_index", |
| "description": "Integer task identifier referenced by every frame.", |
| "dataType": "sc:Integer", |
| "source": { |
| "fileObject": { |
| "@id": "tasks-jsonl" |
| }, |
| "extract": { |
| "column": "task_index" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "tasks/task", |
| "name": "task", |
| "description": "Free-form English instruction describing the manipulation goal.", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { |
| "@id": "tasks-jsonl" |
| }, |
| "extract": { |
| "column": "task" |
| } |
| } |
| } |
| ] |
| }, |
| { |
| "@type": "cr:RecordSet", |
| "@id": "frames", |
| "name": "frames", |
| "description": "Per-frame proprioception. One row per 50 Hz timestep across all 16,000 episodes (3,728,773 rows total).", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "frames/observation_state", |
| "name": "observation.state", |
| "description": "14-DoF joint state vector: [left_waist, left_shoulder, left_elbow, left_forearm_roll, left_wrist_angle, left_wrist_rotate, left_gripper, right_waist, right_shoulder, right_elbow, right_forearm_roll, right_wrist_angle, right_wrist_rotate, right_gripper].", |
| "dataType": "sc:Float", |
| "repeated": true, |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files" |
| }, |
| "extract": { |
| "column": "observation.state" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "frames/action", |
| "name": "action", |
| "description": "14-DoF target joint command vector (same channel order as observation.state).", |
| "dataType": "sc:Float", |
| "repeated": true, |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files" |
| }, |
| "extract": { |
| "column": "action" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "frames/timestamp", |
| "name": "timestamp", |
| "description": "Time within the episode in seconds.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files" |
| }, |
| "extract": { |
| "column": "timestamp" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "frames/frame_index", |
| "name": "frame_index", |
| "description": "Frame index within the episode (0-based).", |
| "dataType": "sc:Integer", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files" |
| }, |
| "extract": { |
| "column": "frame_index" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "frames/episode_index", |
| "name": "episode_index", |
| "description": "Global 0-based episode index (0..15999).", |
| "dataType": "sc:Integer", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files" |
| }, |
| "extract": { |
| "column": "episode_index" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "frames/index", |
| "name": "index", |
| "description": "Global 0-based row index across the entire dataset.", |
| "dataType": "sc:Integer", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files" |
| }, |
| "extract": { |
| "column": "index" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "frames/task_index", |
| "name": "task_index", |
| "description": "References the natural-language task description in the `tasks` record set.", |
| "dataType": "sc:Integer", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files" |
| }, |
| "extract": { |
| "column": "task_index" |
| } |
| }, |
| "references": { |
| "field": { |
| "@id": "tasks/task_index" |
| } |
| } |
| } |
| ] |
| }, |
| { |
| "@type": "cr:RecordSet", |
| "@id": "videos", |
| "name": "videos", |
| "description": "MP4 video files; one per (episode, camera) pair. The video filename embeds episode_index and camera key.", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "videos/filename", |
| "name": "filename", |
| "description": "Path of the MP4 file relative to the repository root.", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "video-files" |
| }, |
| "extract": { |
| "fileProperty": "fullpath" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "videos/content", |
| "name": "content", |
| "description": "Raw H.264 MP4 bytes for one (episode, camera) pair, 480x640 @ 50 fps, GOP=2.", |
| "dataType": "sc:VideoObject", |
| "source": { |
| "fileSet": { |
| "@id": "video-files" |
| }, |
| "extract": { |
| "fileProperty": "content" |
| } |
| } |
| } |
| ] |
| } |
| ] |
| } |
|
|
