EMPIRIC: an agent that extends physics engines with code for missing mechanisms
tomssilver · x · 2026-10-05
EMPIRIC is an agent that learns a residual world model: a physics engine extended with code for mechanisms it can't natively model, like glue curing, water heating, or fan wind. It experiments, writes what it learned into its simulator as code, and plans in that model. It solves all 25 runs across five simulated domains where coding-agent baselines solve 14–16, and on a real robot it learns wind and domino masses from just two gusts, building a bridge one block longer than seen in training.
More from Embodied
- A SOTA real2sim recipe for robotics: agents build sims with real physics and affordance — simonguozirui · 2026-10-06
- Ghost AI launches Core, a $3,499 personal AI computer running models fully on-device — Jackyhuang · 2026-10-06
- Smart Machines Are Transforming China's Farms — yogthos · 2026-10-06
- Reverse engineering Neuralink's new decoder architecture from a sparse blog post — melnykowycz · 2026-10-06
- Home robot reviews will be wild: 'Amazing gait, put my keys in the fridge' — VraserX · 2026-10-06
- Snap Staffers on Why $2,195 Specs AR Glasses Landed Chunky: The Meta Boogeyman — richardjnieva · 2026-10-05