Build your robotics experiment#
RoboAgentLab connects a coding agent to robot environments, policies, and tools. Start with one simulation episode, inspect its interaction trace, then change the model, interface, or environment.
Install the extension and use /embodied-setup to configure LIBERO.
Use action units, native tools, or generated Python programs.
Run task–seed grids, inspect outcomes, and replay sessions.
Expose a robot or algorithm through a service, extension, and skill.
How the pieces fit together#
pi provides the agent loop, model access, and sessions. Robot extensions register observations and callable capabilities. Python services run simulators, robot adapters, perception, and policies in their own environments. Skills explain how the agent should use them. A shared primitive declaration also exposes supported operations to generated Python programs.
The platform code is available in RoboAgentLab/RoboAgentLab. The v1.0 source release is a fixed snapshot; these guides follow the maintained main branch.
Get started