ENPIRE turns coding agents into autonomous robotics researchers, hitting 99% success on dexterous tasks
chris_j_paxton · x · 2026-10-09
A NVIDIA/CMU/UC Berkeley team unveiled ENPIRE (accepted to CoRL 2026), a harness framework that lets coding agents autonomously improve robot policies in the real world.
- Core idea: a repeatable physical feedback loop — reset scene, execute policy, verify outcome, refine — as the missing abstraction for automating robotics research.
- Four modules: Environment (auto reset & verification), Policy Improvement, Rollout (parallel real robots), Evolution (agents analyze logs, read literature, improve training infra and algorithm code).
- Results: 99% success rate policies on dexterous tasks like organizing a pin box or fastening a zip tie with minimal human effort.
On the RoboPapers podcast the authors quipped they couldn't name a single thing a robotics PhD does that a coding agent couldn't, given enough compute and a robot fleet.
Related event: ENPIRE Lets Coding Agents Autonomously Improve Robot Policies(2 posts)→
More from Embodied
- Odyssey-3 Tested: Frozen Backbone Trained on 20 Hours of Driving Data Transfers — testingcatalog · 2026-10-09
- IRVL Researcher to Present UHAS, HRT1 and VLA-Replica at Texas A&M Robotics Seminar — YuXiang_IRVL · 2026-10-09
- Google Gemma-Backed Hardware Hackathon at AGI House Gathers 120 Builders for On-Device AI — agihouse_org · 2026-10-09
- Odyssey launches Odyssey-3, claims SOTA world model on Physics-IQ benchmark — Scobleizer · 2026-10-09
- Figure valued at $39B: tracking Brett Adcock's promise of 100,000 robots in 4 years — annatonger · 2026-10-09
- Sentdex: a raw multimodal LLM drives a robot with zero training, VLA era may be over — Sentdex · 2026-10-09