AgenticROS: Connects ROS with OpenClaw, Claude (code, desktop, dispatch), and Google Gemini
Open-source bridge lets AI agents like Claude and Gemini directly control robots via four deployment modes.
Developer Chris Matthieu has open-sourced AgenticROS, a project that bridges the gap between the Robot Operating System 2 (ROS2) and leading AI agent frameworks. It enables robots to be controlled via natural language by connecting them to agents like Anthropic's Claude (in its Code, Desktop, and Dispatch variants), Google's Gemini, and the OpenClaw/NemoClaw projects. The system's core innovation is its four flexible deployment modes, allowing the AI to run directly on the robot's computer (Mode A), on a separate machine on a local network via WebSocket (Mode B), from a cloud server using WebRTC for NAT traversal (Mode C), or integrated with the Zenoh middleware for low-latency communication (Mode D).
Beyond basic connectivity, AgenticROS introduces a modular 'Skills' architecture. Developers can install optional skill packages—like the reference 'Follow Me' skill for person-following—that register specific tools (e.g., `follow_robot`, `check_battery`) with the AI agent. This allows users to issue high-level commands such as 'Move forward 1 meter,' 'What do you see?,' or 'Navigate to the kitchen,' which the agent translates into ROS2 topics and actions. The project provides a curated list for discovering and sharing these skills, turning generic AI models into capable robotics controllers.
- Connects ROS2 to multiple AI agents including Claude, Gemini, and OpenClaw via four deployment modes (Local, Network, Cloud, Zenoh).
- Features a plugin-based 'Skills' architecture for adding capabilities like person-following, navigation, and emergency stops.
- Allows natural language control of robots with commands like 'Navigate to the kitchen' or 'Follow me'.
Why It Matters
Dramatically lowers the barrier for using advanced LLMs to control and program robots, accelerating development and remote operation.