Nautilus accepted to NeurIPS: one prompt, any policy, any benchmark, any robot
GeorgiaChal · x · 2026-09-25
Nautilus is an agentic harness for robot learning, aiming to take a request like "evaluate π0 on LIBERO" to a real-robot rollout in roughly 1.5 hours.
- Core thesis: better models alone won't scale robot learning — we need the right interfaces and structure so policies, benchmarks, and robots plug together.
- Form: one prompt adapts to any policy, any benchmark, any robot, with a Claude Code plugin.
- Accepted to NeurIPS 2026, led by Yufeng Jin.
More from Embodied
- Tesla Robotaxi fleet now spans 420 Model Ys and 69 Cybercabs — bendomingueztv · 2026-09-25
- NVIDIA Talks Cosmos 3: Simulation and Verifiable Rewards for Physical AI — liu_mingyu · 2026-09-25
- C5R builds an AI-run research facility in 12 weeks, launches SciUniverse benchmark — akilian · 2026-09-25
- YC-backed Cosmic Robotics has installed 11,000 solar panels and signed a major contract — ycombinator · 2026-09-25
- Meta debuts palm-sized dedicated 5G device for Muse, report says — _xjdr · 2026-09-25
- Father of surgical robotics Russell Taylor wins 2027 IEEE Robotics Award — mdredze · 2026-09-25