MIT's FLEX lets robots generalize common sense from a single natural-language command
MIT News AI · rss · 2026-11-09
MIT and NYU researchers will present FLEX at CoRL: a three-step feature-editing pipeline for robot motion planning. Add registers objects from a user's command; Generalize uses GPT-4o to infer broader preferences (laptop → avoid all electronics); Mask prunes absent objects via vision — all in 5 seconds. In coffee, candle and bowling-ball simulations, one or two sentences sufficed for robots to follow both literal and implied preferences, enabling household robots to adapt to individual homes.
More from Embodied
- Musk: Cybertruck orders outpace production; line to shut in ~8 months for Cybercab — elonmusk · 2026-11-09
- Google DeepMind to give sneak peek of robotics work at CoRL booth — egrefen · 2026-11-09
- Tesla Robotaxi expands to San Antonio starting tomorrow at 5am — mitchdeg · 2026-11-09
- Open-source robot manipulation stack Specialist: hardware, data, training, deployment all released — TinfoilTricorn · 2026-11-09
- RAI Institute's two-wheeled UMV robot does backflips with just 5 degrees of freedom — Scobleizer · 2026-11-09
- Team hacks robot arms to 85 points, wins MakerMods Physical AI Hackathon — IsaiahBallah · 2026-11-09