LLMPhy: Parameter-Identifiable Physical Reasoning Combining Large Language Models and Physics Engines

TL;DR

LLMPhy is a novel framework that combines large language models with physics engines to accurately identify physical parameters and construct digital scene twins for improved physical reasoning.

Contribution

It introduces a black-box optimization method that leverages LLMs for parameter estimation in physical scenes, integrating textual knowledge with physics simulation.

Findings

LLMPhy outperforms prior methods in parameter recovery accuracy.

Achieves state-of-the-art results on new physical reasoning benchmarks.

Demonstrates reliable convergence in zero-shot physical reasoning tasks.

Abstract

Most learning-based approaches to complex physical reasoning sidestep the crucial problem of parameter identification (e.g., mass, friction) that governs scene dynamics, despite its importance in real-world applications such as collision avoidance and robotic manipulation. In this paper, we present LLMPhy, a black-box optimization framework that integrates large language models (LLMs) with physics simulators for physical reasoning. The core insight of LLMPhy is to bridge the textbook physical knowledge embedded in LLMs with the world models implemented in modern physics engines, enabling the construction of digital twins of input scenes via latent parameter estimation. Specifically, LLMPhy decomposes digital twin construction into two subproblems: (i) a continuous problem of estimating physical parameters and (ii) a discrete problem of estimating scene layout. For each subproblem, LLMPhy iteratively prompts the LLM to generate computer programs encoding parameter estimates, executes them in the physics engine to reconstruct the scene, and uses the resulting reconstruction error as feedback to refine the LLM's predictions. As existing physical reasoning benchmarks rarely account for parameter identifiability, we introduce three new datasets designed to evaluate physical reasoning in zero-shot settings. Our results show that LLMPhy achieves state-of-the-art performance on our tasks, recovers physical parameters more accurately, and converges more reliably than prior black-box methods. See the LLMPhy project page for details: https://www.merl.com/research/highlights/LLMPhy