MIT's WorldParticle: one AI model simulates sand, cloth, fluids and proteins ~100x faster
winer666 · reddit · 2026-11-10
Researchers at MIT, UBC and partners built WorldParticle, a single transformer-based model that simulates cloth, elastic solids, fluids, granular materials, molecular dynamics and more — behaviors that each previously required dedicated solvers.
Method: rather than being programmed with governing equations, it treats everything as particles and learns dynamics from data, predicting motion step by step.
Results:
- Outperforms existing neural simulators on accuracy across all six categories
- Models protein motion 100x faster than numerical simulators
Published in ACM Transactions on Graphics, to be presented at SIGGRAPH Asia; covered by MIT News.
Caveats: trades some per-phenomenon accuracy vs. specialist solvers, and each category still needs its own parameter set — a single shared model remains future work.
More from Embodied
- Steve Waslander appointed Director of U of T Robotics Institute — VectorInst · 2026-11-10
- Boston Dynamics' Atlas robot learns to roller skate and pull off a 360 aerial spin — adariostrange · 2026-11-10
- Embodied Turing Machines: coding agents write reusable robot routines, 70.24% on 42 bimanual tasks — jiqizhixin · 2026-11-10
- CoRL 2026: EgoScale lands oral session, latest GR00T2 results at H2R workshop — chris_j_paxton · 2026-11-10
- Startup builds robotic hands with a single automated fiber-braiding process — TinfoilTricorn · 2026-11-10
- Blind Skateboarder: robot lands timed 3-minute manual with no vision or markers — carlosdponx · 2026-11-10