Training contact-rich humanoid manipulation policies purely inside a world model
chris_j_paxton · x · 2026-09-04
- RL has delivered gains in whole-body humanoid control but struggled with manipulation, where capturing contact dynamics in simulation is hard.
- Training inside learned world models is a promising fix, yet powerful world models are too computationally expensive to train in directly.
- A new paper from @Jsphamigo and @Rk4342R decomposes the problem: a large global world model generates forward trajectories, while a lightweight low-dimensional latent-space model approximates local contact dynamics — making contact-rich humanoid manipulation training tractable purely within a world model.
Related event: Robots learn contact-rich manipulation in world models without simulators(2 posts)→
More from Embodied
- Tuesday Labs' home robot Voxel charms with Pixar-collaborated personality design — cneuralnetwork · 2026-10-20
- Tuesday Labs opens reservations for Voxel Series-1 home robot, first 999 units — cneuralnetwork · 2026-10-20
- Wrist-worn dictaphone is the best ambient compute form factor, says dev — curious_vii · 2026-10-20
- a16z leads Eka Robotics Series A, betting on Vision-Force-Action models for dexterous manipulation — a16z · 2026-10-20
- Autonomous Lamp: a $999 open-source desk robot giving personal AI agents eyes, ears, and a body — dee_hw · 2026-10-20
- Paid $60 for two humanoid robots to clean for an hour: here's how it went — 2C_ornot2C · 2026-10-20