Verne Robotics shows open-weight world action model V0.5 beating π0.5 on RoboTwin 2.0 with ~2hrs data
chris_j_paxton · x · 2026-11-17
A tour of Verne Robotics in SF shows their new open-weight world action model V0.5 in action: full-body tasks learned from 2 hours of data, beating π0.5 on RoboTwin 2.0, with coordinated control of two robots. Chris Paxton comments that training your own models remains a big edge for systems integrators, for now.
More from Embodied
- Roboticist argues better action representations may beat scaling synthetic data — KyleMorgenstein · 2026-11-17
- Policies use only a sliver of image information — extracting it matters for safety — KyleMorgenstein · 2026-11-17
- Enactra shows agentic robot simulation with controllable worlds and per-frame annotations — Lianhuiq · 2026-11-17
- South Korea's robot density hits 1,066 per 10k workers, 7x the world average — randal_olson · 2026-11-17
- Indie dev shows new PCB batch of robotics BLDC controller passing unit tests — wavefnx · 2026-11-17
- FACE: contact-factorized imitation learning lets one policy handle unseen friction and geometry — KyleMorgenstein · 2026-11-17