Home
Under Review

MAGMA-GEN

Teaching robots to execute complex, multi-step tasks from natural language goals and scene observations without expert data.

Collaborators Siléane, Gepetto Team (LAAS-CNRS)

robotics language manipulation

🏗️ Content on this page is a work in progress and will be updated as projects and publications are finalized.

We study how to bridge language, perception, and control to solve long-horizon tasks like “prepare a workspace for soldering” or “pack a lunch.” The system plans with compositional skills, adapts to failures, and executes in real robotic environments.

Share this page