Driving · 2026-09-30
EMPIRIC: Experiment-Driven Learning of Residual World Models for Robot Planning
Yichao Liang, Amber Li, Dat Nguyen, Emily Bunnapradist, Michelangelo Naim, Sreela Kodali, Matteo Merler, Bowen Li, Kiran Gopinathan, Yiyun Liu, Nikhil Pimpalkhare, Joshua B. Tenenbaum, Adrian Weller, Zenna Tavares, Tom Silver, Kevin Ellis
Auto-summarized: this summary was generated by a language model from the paper’s text and has not been reviewed by an editor. Check the paper before relying on it. Benchmark numbers appear only after a human has verified them.
TL;DR
EMPIRIC learns residual world models for robot planning by extending a base physics simulator with Python programs for missing physical mechanisms (forces, constraints, recurrent hidden state). It uses Bayesian inference to estimate program parameters and hidden state from noisy observations, chooses informative experiments, revises the programs when predictions fail, and plans under uncertainty. It reports solving more tasks with fewer environment interactions than three baselines across five simulated domains, and demonstrates on a physical robot.
Why it matters
The authors address learning unknown physical mechanisms (e.g., glue curing, water heating, wind) on top of known rigid-body physics, including hidden state and uncertainty needed for safe planning and experiment selection. EMPIRIC’s residual executable models are interpretable and reusable, and the evaluation uses a continual experiment-driven protocol measuring both success and environment-step efficiency.
Method
- Learn residual world models as Python programs that subclass a base rigid-body physics simulator, keeping engine dynamics and adding missing mechanisms with parameters and recurrent hidden state.
- Use Bayesian inference with a mean-field approximation to estimate a belief over program parameters and the current recurrent model state from noisy replayed observation-action history.
- Plan with the learned simulator: simulate skill sequences from belief draws to maximize success probability; also select experiments via mutual information over predicate readings; monitor predicates and revise programs after mismatches.
Abstract (from arXiv)
A robot should be able to learn through experiments how unfamiliar objects behave and interact, then plan with that knowledge. It need not start from scratch: physics engines supply knowledge of motion and contact, but can omit entire mechanisms, such as glue curing, water heating, or wind. We present EMPIRIC, an agent that learns a residual world model: a physics engine extended with code for the missing mechanisms. The learned programs can introduce new forces, constraints, and hidden state, and Bayesian inference estimates their parameters and states from noisy observations. The resulting model lets the agent predict the outcomes of actions, choose informative experiments, and revise its hypotheses when predictions fail. Across five simulated domains, EMPIRIC learns interpretable, reusable models, and solves more tasks with fewer environment interactions than all three baselines. On a physical robot, it learns wind forces and domino masses to solve a manipulation task. Website and code: https://yichao-liang.github.io/empiric
Related papers
- Copper-Policy: Focus on the Representation for Robust Robot Manipulation
- SkillWeaver: Agentic Exploration over Neural Interaction Skills for Scalable Robot Data Generation
- One from Infinity: Actualizing Futures from Pretrained World Models into Robot Actions
- RoboHarn-Evo: Evolving Hierarchical Physical Knowledge for Self-Improving Robotic Manipulation
- WorldLine: Action-Driven Visual Simulation for Robotic Manipulation