SmallWorlds: Assessing Dynamics Understanding of World Models in Isolated Environments

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Xinyi, Xia, Zaishuo, Lu, Weyl, Hao, Chenjie, Chen, Yubei
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917112489967616
author Li, Xinyi
Xia, Zaishuo
Lu, Weyl
Hao, Chenjie
Chen, Yubei
author_facet Li, Xinyi
Xia, Zaishuo
Lu, Weyl
Hao, Chenjie
Chen, Yubei
contents Current world models lack a unified and controlled setting for systematic evaluation, making it difficult to assess whether they truly capture the underlying rules that govern environment dynamics. In this work, we address this open challenge by introducing the SmallWorld Benchmark, a testbed designed to assess world model capability under isolated and precisely controlled dynamics without relying on handcrafted reward signals. Using this benchmark, we conduct comprehensive experiments in the fully observable state space on representative architectures including Recurrent State Space Model, Transformer, Diffusion model, and Neural ODE, examining their behavior across six distinct domains. The experimental results reveal how effectively these models capture environment structure and how their predictions deteriorate over extended rollouts, highlighting both the strengths and limitations of current modeling paradigms and offering insights into future improvement directions in representation learning and dynamics modeling.
format Preprint
id arxiv_https___arxiv_org_abs_2511_23465
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SmallWorlds: Assessing Dynamics Understanding of World Models in Isolated Environments
Li, Xinyi
Xia, Zaishuo
Lu, Weyl
Hao, Chenjie
Chen, Yubei
Machine Learning
Current world models lack a unified and controlled setting for systematic evaluation, making it difficult to assess whether they truly capture the underlying rules that govern environment dynamics. In this work, we address this open challenge by introducing the SmallWorld Benchmark, a testbed designed to assess world model capability under isolated and precisely controlled dynamics without relying on handcrafted reward signals. Using this benchmark, we conduct comprehensive experiments in the fully observable state space on representative architectures including Recurrent State Space Model, Transformer, Diffusion model, and Neural ODE, examining their behavior across six distinct domains. The experimental results reveal how effectively these models capture environment structure and how their predictions deteriorate over extended rollouts, highlighting both the strengths and limitations of current modeling paradigms and offering insights into future improvement directions in representation learning and dynamics modeling.
title SmallWorlds: Assessing Dynamics Understanding of World Models in Isolated Environments
topic Machine Learning
url https://arxiv.org/abs/2511.23465