RATs: Multi-Agent Robot Lifelong Skill Learning Without Gradients or RL

rsasaki0109 · x · 2026-07-30

RATs is a multi-agent Code-as-Policy system for lifelong robot skill learning. During free-form play, LLM agents invent tasks, write code-as-policy, and distill successful executions into a reusable skill library. At evaluation, skills are reused as planner context, with no gradients or RL; all learning through structured natural-language feedback and code reuse.

Original post →

More from Embodied

Embodied channel →