From Generation to Simulation: How Far Are World Models from Being True Simulators?
2026-08-24 • Artificial Intelligence
Artificial IntelligenceComputer Vision and Pattern Recognition
AI summaryⓘ
The authors studied how well new AI models called world models can replace traditional simulators used in games and physics. They looked at eight important features of simulators and reviewed 200 papers from recent years to see which features world models can do well and where they fall short. They found that world models can handle interaction and control in some cases, but struggle with reliably following physical laws, giving detailed feedback about the system state, and maintaining stable long-term behavior. The authors suggest six areas where future research should focus to improve these AI-based simulators.
diffusion modelsworld modelsphysics enginecontrollabilitystate feedbacksimulationlatent dynamicsvideo generationreinforcement learninglong-horizon stability
Authors
Tong Wang, Huan Deng, Mucheng Yang, Yang He, Xiaohui Kuang, Gang Zhao
Abstract
With the rapid progress of diffusion models and large-scale video generation, generative world models are increasingly expected to replace traditional simulators, including physics engines, game engines, and reinforcement-learning environments. Yet the remaining distance from generation to simulation lacks a systematic assessment. We present a capability-based study using an external yardstick: eight capabilities of a traditional simulator, namely asset construction, physics engine, interaction, controllability, stability, state feedback, diversity, and evaluation metrics. We trace three main technical routes--latent dynamics, video generation, and joint-embedding prediction--and map exactly 200 representative works published from 2018 to June 2026 onto these capabilities. Our analysis shows that world models have achieved functional substitution in interaction and controllability for specific scenarios, but remain short of traditional simulators in formal guarantees of physical laws, structured state feedback, and reproducible long-horizon evolution. State feedback is the most neglected cross-route shortcoming: only 6 of 163 implementation papers expose a runtime interface for querying entity states or physical parameters. We identify six research directions: formalized physics, a unified action interface, first-class state feedback, long-horizon stability, downstream-utility evaluation, and cross-route hybridization. Project page: https://github.com/AtongWang/world-model-simulators