Position: The Physics-Physical Reasoning Interplay is Key for Future Embodied World Models

Terry Jingchen Zhang, Kun Xiang, Yinya Huang, Jixi He, Zirong Liu, Yueling Tang, Ruizhe Zhou, Chengyu Yu, Xiaodan Liang

Keywords: World Model, Embodied Intelligence, Physics Reasoning, Physical Reasoning

TL;DR: We posit the interplay of physics reasoning (to understand the laws of physics) and physical reasoning (to apply these laws for inference in real-world physical behavior) is key for developing future embodied AI agents powered by world models.

Abstract:

World modeling represents a critical frontier towards autonomous AI agents capable of capturing environmental dynamics for intelligent decision-making. While current progress scales on massive data corpora, we argue that future world models require combining physics reasoning (understanding the fundamental laws of nature) with physical reasoning (applying these laws to predict observable behaviors and outcomes of intervention).
In this paper, we outline the promise and reality as to how frontier models perform in both physics and physical reasoning, then propose a new pathway for future world models and embodied intelligence to internalize the laws of physics to internalize the mechanism of surrounding physical world just like how humans learn to interact with the world in a universally generalizable manner, thereby establishing the foundation for autonomous, efficient and reliable embodied intelligence.

Submission Type: Position/Review Paper (4-9 Pages)

NeurIPS Resubmit Attestation: This submission is not a resubmission of a NeurIPS 2025 submission.

Submission Number: 59