Agentic Robot Creation through Generation, Verification, and Revision

This research investigates whether an AI agent can transform high-level user intent into a functioning robot and teach it to behave. The study addresses embodiment and behavior creation through computational metrics for evaluating mechanical designs and the Roboforge framework, which translates task specifications into executable policies. The framework incorporates motion generation, policy learning, and execution-grounded revision, with results showing improved task performance and behavior quality across multiple robots.