ENPIRE: Agentic Robot Policy Self-Improvement in the Real World
arXiv·medium signal
A closed-loop framework that lets coding agents autonomously improve real-world robot policies via four modules: environment management, policy refinement, parallel rollout evaluation, and agent-driven evolution. Reports 99% success on manipulation tasks (pin-box organization, zip-tie fastening, tool use), with gains accelerating when run across multiple robots in parallel. Extends the 'agent that improves itself' pattern from code into physical robotics.