Verne Robotics bets on on-device inference: robots can't rely on WiFi to run policies
danfei_xu · x · 2026-09-15
Responding to SemiAnalysis' deep dive on robot compute featuring Verne Robotics, founder Danfei Xu explains why the company invested heavily in on-device inference: in warehouses and factories WiFi is unreliable, and a network hiccup breaks high-frequency policy execution. That requires vertical integration from models to inference to controllers. She frames robot learning infra as a science in its own right, and argues scaling models is necessary for robot intelligence. Notably truncated in the quote, but the two directions she's most excited about start with connecting System 1-style fast reactions to deeper reasoning.
More from Embodied
- Breath of the Wild PCVR mod running on Apple Vision Pro with an RTX 5090 — bradneuberg · 2026-09-15
- Apple foldable hands-on: no visible crease, excellent contrast even in bright light — BenBajarin · 2026-09-15
- 30-minute hackathon build: real-time tabletop 3D reconstruction bridge for robot sim — chrismatthieu · 2026-09-15
- The Embodiment Gap in Robotics and the Race to Close It — enlightened_none · 2026-09-15
- Robot Foundation Models Like Astra May Be Key to Truly Safe Robots — chris_j_paxton · 2026-09-15
- BAAI's World Action Models use video generation to let robots imagine before acting — stepjamUK · 2026-09-15