RoboClaw

RoboClaw is an open-source embodied intelligence assistant.

✨ What We Want To Build

RoboClaw is being built around four planes:

• General Embodied Entry Plane: the user-facing path from natural language to a runnable embodied session
• Embodiment Onboarding Pattern Plane: the reusable pattern for bringing new open-source embodiments into RoboClaw without hard-coding one robot into the generic stack
• Cross-Embodiment Skill Base Plane: the shared semantic layer where multiple embodiments can expose compatible skills and procedures
• Research Assistant Plane: evaluation, failure analysis, recovery, and research workflows built on top of the embodied execution stack

Right now, we are concentrated on the first plane.

Overall Data Flow

RoboClaw Agent
  ↓
Workspace / Catalog / Setup Resolution
  ↓
Procedure (connect / calibrate / move / debug / reset step graph)
  ↓
Runtime Adapter (active execution layer)
  ↓
Control Surface (interface profile + runtime server)
  ↓
Embodiment Runtime (hardware/sim-specific runtime)
  ↓
Real/Sim Embodiment

Current embodiment coverage is tracked like this:

Category	Representative	Simulation	Real
Arm	SO101	🟡	🟡
Dexterous Hand	Inspire	🔴	🔴
Humanoid	G1	🔴	🔴
Wheeled Robot	WBase	🔴	🔴

📦 Installation

For Users

• AI-assisted setup: ask your coding assistant:

Help me install RoboClaw from https://github.com/MINT-SJTU/RoboClaw

• Non-Docker Installation
• Docker Installation

For Developers

• Docker Workflow

📢 Community Co-Creation

RoboClaw is being built in the open. We want major direction-setting choices, such as embodiment support, simulator priorities, and roadmap focus, to be discussed with the community.

You can contribute through:

• Issues: bug reports, feature requests, and implementation suggestions
• Pull Requests: code and documentation improvements

The most useful contribution areas right now are:

• embodied AI architecture
• capability abstraction and semantic skill interfaces
• ROS2 and execution-layer integration
• simulator support and real robot adaptation
• evaluation, validation, and developer experience

If you want to contribute more actively, contact us at bozhaonanjing [[@]] gmail [[DOT]] com.

🗺️ To-do List

• Set up the open-source repository, publish the initial README, and add GitHub-native proposal entry points
• Document the first embodied stack and its boundaries
• Define unified embodied capabilities and semantic interfaces
• Run the first real setup end to end with workspace-generated assets
• Support the first real robot platform
• Design safe-stop and recovery mechanisms
• Improve first-run reliability for embodied setup and execution
• Expand from the first supported embodiment to more open-source robots
• ...

Coming soon.

🙏 Acknowledgments

RoboClaw references and inherits part of its initial thinking from nanobot. We appreciate its lightweight practice along the OpenClaw line, which helped us build the first prototype faster and continue evolving toward embodied intelligence.

Community Channels

• Discord: Join the server
• WeChat official post: Coming Soon
• GitHub Issues: Create an issue
• Email: business@evomind-tech.com

Affiliations

Citation

@misc{roboclaw2026,
  title        = {RoboClaw: An Open-Source Embodied Intelligence Assistant},
  author       = {RoboClaw Contributors},
  year         = {2026},
  howpublished = {\url{https://github.com/MINT-SJTU/RoboClaw}}
}