CoRobot 1.0

English | 中文

CoRobot 1.0 is an open embodied-data framework for end-to-end data collection, conversion, processing, retrieval, preview, download, and training. It follows three core principles—Collaboration, Coherence, and Collective—to improve standardization and reuse efficiency for multi-robot datasets through unified data infrastructure.

Repository Modules

Module	Role	Highlights	Status
RoboCOIN	Data & model asset management	Multi-robot bimanual dataset: 16 robot bodies, 200k+ trajectories, 10+ scenes, 1,000+ tasks, 50+ skills, 500+ objects.	Open-sourced
RoboCOIN-DataManager	Data governance	Visual search: keyword search, visualization, helps targeted downloads.	Open-sourced
DataTrain	Model training	Unified training tools: quick integration/training for embodied models like OpenPI, RDT, DP.	Coming soon
DataCollect	Data collection	A multi-robot data collection system built on LeRobot, composed of RoboDriver, RoboDriver-Server, and RoboXstudio.	Open-sourced
DataConvert	Data conversion	Conversion toolkit: bi-directional transforms among RLDS, HDF5, JSONL, and LeRobotDataset.	Coming soon
DataForge	Data processing	Processing toolkit: defect filtering (static frames, frame drops, dimensional mismatch, missing fields, jitter) and enrichment labels (scene, sub-task, motion description).	Coming soon

RoboCOIN

RoboCOIN is a large-scale, multi-robot, bimanual dataset for manipulation, covering 15 robot platforms, 421 tasks across 16 scenes, and 180K demonstration trajectories.

• Resources:

  • Homepage: https://flagopen.github.io/RoboCOIN/
  • Hugging Face: https://huggingface.co/RoboCOIN
  • ModelScope: https://modelscope.cn/organization/RoboCOIN
  • The datasets are continuously updated; licenses and versions follow each dataset page.

• Dataset overview:

  

  

RoboCOIN-DataManager

RoboCOIN-DataManager is the visualization and management platform for RoboCOIN, enabling online search, preview, filtering, and download.

• Online: https://flagopen.github.io/RoboCOIN-DataManager/

• Key features:

  • Multi-dimensional filtering: filter by scene, robot, end-effector, action, object, etc.
  • Instant preview: auto-playing videos and detailed info with hover overlays
  • Intelligent search: keyword search and Filter Finder for quick targeting
  • Cart management: batch select/add/remove datasets with persisted selections
  • One-click download: generate ModelScope/HuggingFace download commands; export Python scripts
  • Selection import/export: save and restore filters in JSON for team collaboration

• UI preview:

  

  

  

DataCollect

DataCollect is a multi-robot data collection system built on the LeRobot framework. It consists of RoboDriver, RoboDriver-Server, and RoboXstudio: RoboDriver provides a development toolkit for robot drivers with a multi-protocol data-ingestion framework; RoboXstudio is an all-in-one embodied data platform for task dispatch, data management, and annotation; RoboDriver-Server bridges RoboDriver and RoboXstudio to forward and sync data.

• Core features:
  • Multi-protocol streams (ROS1/ROS2, Dora, sockets)
  • Unified collection framework built on LeRobot
  • Multi-robot, multi-sensor fusion
  • Full pipeline for live capture, validation, and submission
  • Integrations with task system, annotation system, and training pipeline on the platform side

It bridges on-device capabilities with platform AI services.

• System diagram:

• Data collection UI:

DataConvert

DataConvert is a dedicated toolkit for embodied data format conversion, supporting bi-directional transforms among RLDS, HDF5, LeRobotDataset, JSONL, and LeRobot formats to meet diverse training needs.

• Dataset ingest: author dataset yaml files (tasks, device models, scenes, actions, etc.), which are validated to produce info metadata for ingestion.
• Format conversion: write conversion configs, align pose files and check frame counts, then run the official conversion to produce standard LeRobot datasets (including parquet, mp4, and meta).
• Tech overview:

More updates coming soon.

DataForge

DataForge is a toolkit for embodied data filtering and hierarchical labeling, refining raw data into high-quality CoRobot datasets. It handles dropped frames, freezes, artifacts, noise, grayscale frames, joint info gaps, and timestamp misalignment while adding structured labels.

• Data pre-processing: remove abnormal or incomplete episodes; filter static frames, frame drops, and similar issues.
• Post-process & validation: fix trajectory jitter, dimensional mismatch, and missing fields; verify consistency/reproducibility via Sim Replay.
• Labeling & integration: add RoboCOIN-specific three-level labels (scene, sub-task, motion) and embed them into parquet and meta to produce final CoRobot-format data.
• Tech overview:
• Hierarchical annotation:

More updates coming soon.

Community

• Issues: Please report bugs, feature requests, or data protocol suggestions in GitHub Issues.
• Discussions: Coming soon for solution sharing and co-building.

Contributor

Thanks to all contributors and multi-robot partners! Join via PRs, Issues, or community discussions:

• FlagOpen / CoRobot Team
• Community volunteers

Citation

@misc{corobot2025,
  title        = {CoRobot 1.0: An Open Embodied Data Infrastructure for Multi-Robot Collaboration},
  author       = {CoRobot Team},
  year         = {2025},
  publisher    = {FlagOpen},
  howpublished = {\url{https://github.com/FlagOpen/CoRobot}}
}

License

The main repo and submodules follow the licenses in their respective directories. For commercial use or large-scale deployments, please contact the maintainers; a unified open-source license will be announced in future versions.