Dex-One2Many: Real2Sim2Real Turns One Human Video Into Robots Generalizing Across Configurations
furongh · x · 2026-10-10
Dex-One2Many uses a Real2Sim2Real pipeline with neuro-symbolic representations and a new technique called 'controlled diversification': a vision-language model converts one human demonstration video into scene-graph sequences capturing task structure, while the symbolic constraints specify what must hold and a neural policy learns how to achieve it. By generating diverse simulated training experiences that preserve task constraints, the resulting robot generalizes beyond the demo to new object positions, goal poses, and grasps.
Related event: Dex-One2Many: One Human Video Trains Multi-Morphology Robot Policies(2 posts)→
More from Embodied
- Former SpaceX engineers raise $100M for self-driving electric freight trains — chrisgrayson · 2026-10-10
- Kleiner Perkins' Ilya Fushman: the moment for robotics is finally now — Sethwinterroth · 2026-10-10
- Coevolved robot communication transfers poorly to 3D: 1 success in 30 seeds — uv-mex · 2026-10-10
- Transferring co-evolved robot communication from 2D to 3D physics simulation — uv-mex · 2026-10-10
- SGS tweak to RL resets lets sim-trained robots mesh gears at 94% zero-shot — abhishekunique7 · 2026-10-10
- Blogger suggests adding CUA data to robotic model pre-training mix — eigenron · 2026-10-10