Reinforcement World Model Learning for LLM-based Agents

arXiv:2602.05842 · cs.CL · Submitted 2026-02-05 · Read on arXiv

cs.CL

Submitted: 2026-02-05

Updated: 2026-11-01

Comments: neurips 2026 camera ready

Code: https://github.com/modelscope/evalscope

License: http://creativecommons.org/licenses/by/4.0/

The gist: Large language models (LLMs) have achieved strong performance in language-centric tasks.

Terminology

Abstract

Large language models (LLMs) have achieved strong performance in language-centric tasks. However, in agentic settings, LLMs often struggle to anticipate action consequences and adapt to environment dynamics, highlighting the need for world-modeling capabilities in LLM-based agents. We propose Reinforcement World Model Learning (RWML), a self-supervised method that learns action-conditioned world models for LLM-based agents on textual states using sim-to-real gap rewards. Our method aligns simulated next states produced by the model with realized next states observed from the environment, encouraging consistency between internal world simulations and actual environment dynamics in a pre-trained embedding space. Unlike next-state token prediction, which prioritizes token-level fidelity (i.e., reproducing exact wording) over semantic equivalence and can lead to model collapse, our method provides a more robust training signal and is empirically less susceptible to reward hacking than LLM-as-a-judge. We evaluate our method on ALFWorld and τ squared Bench and observe significant gains over the base model, despite using no expert data, task-success rewards, or strong-LLM annotations - only an off-the-shelf embedding model as a distance metric. When combined with task-success rewards, our method outperforms direct task-success reward RL by 6.9 and 5.7 points on ALFWorld and τ squared Bench respectively; it further outperforms expert-data training on ALFWorld and is comparable to it on τ squared Bench.

Sources

Related papers