AgenticROS lets AI agents control ROS2 robots via skills marketplace
Open-source framework connects Claude, Gemini, and more to robot bodies.
AgenticROS is a new open-source framework that connects large language models and AI agents directly to ROS2-powered robots. Created by developer Chris_Matthieu, it supports multiple agent backends including OpenClaw, Anthropic's Claude (code, desktop, and dispatch modes), Google Gemini, Codex, and Hermes. The system has been demonstrated on a physical robot equipped with RealSense depth cameras, wheels, legs, and articulated arms, giving AI models full control over sensors and actuators.
The project also introduces a skills marketplace at skills.agenticros.com where users can browse, share, and deploy pre-built AI behaviors for common robotics tasks—from object grasping to autonomous navigation. By abstracting ROS2 complexity behind agent interfaces, AgenticROS enables developers to command robots using natural language while retaining low-level hardware control. The codebase is released under Apache 2 license, inviting community contributions and forks. This approach parallels NASA JPL's ROSA but focuses on broad multi-agent integration rather than single-assistant interaction.
- AgenticROS bridges OpenClaw, Claude, Gemini, Codex, and Hermes to ROS2 for physical robot control
- Robot hardware includes RealSense cameras, wheels, legs, and arms for full sensorimotor AI control
- Skills marketplace at skills.agenticros.com offers pre-built AI behaviors under Apache 2 license
Why It Matters
Democratizes AI-powered robotics by letting non-ROS experts command physical robots via natural language.