智源 FlagOpen 开源 CoRobot 1.0 具身数据框架

FlagOpen; CoRobot

Repository navigation

NotificationsYou must be signed in to change notification settings; Fork6 (6); Star55 (55)You must be signed in to star a repository

CoRobot embodied data framework

Repository files navigation

README

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

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.

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.

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.

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.

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

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.

CoRobot embodied data framework