ENPIRE: NVIDIA/CMU/Berkeley harness lets coding agents self-improve real robot policies to 99% success
chris_j_paxton · x · 2026-10-07
- NVIDIA, CMU and UC Berkeley (incl. Linxi "Jim" Fan, Yuke Zhu, Guanya Shi) unveil ENPIRE, a harness that lets coding agents autonomously improve robot policies in the real world.
- Key idea: turn real-world robot learning into a repeatable feedback loop — reset the scene, execute a policy, verify the outcome, refine — minimizing human supervision.
- Four modules: Environment (auto reset/verification), Policy Improvement, Rollout (parallel physical robots), and Evolution (agents analyze logs, consult literature, improve training infra and algorithm code).
- Powered by ENPIRE, frontier coding agents autonomously developed a policy achieving 99% success on challenging dexterous manipulation tasks. A RoboPapers podcast episode is coming soon.
Related event: NVIDIA, CMU and Berkeley Unveil ENPIRE for Self-Improving Robot Policies(2 posts)→
More from coding & agent
- CAVEAT testbed exposes how merchants can steer your shopping AI agent — ZacharyHuang12 · 2026-10-07
- Defending skills: a markdown file that helps your agent deserves praise, says eptwts — eptwts · 2026-10-07
- GEA treats agent groups as the evolution unit, hitting 71% SWE-bench Verified with zero human help — xwang_lk · 2026-10-07
- Claude Code v2.1.292 ships plugin marketplace install, agent effort param, security fixes — ashwin-ant · 2026-10-07
- LangChain Releases 90-Second Explainer of Its Deep Agents Framework — LangChain · 2026-10-07
- 1.5B Investor Posts Over 15 Years Wrapped Into a Free MCP Server, Rumor.io — rumor_io · 2026-10-07