Aligning Agentic World Models via Knowledgeable Experience Learning

TL;DR

This paper introduces WorldMind, a framework that builds a symbolic knowledge base from environmental feedback to improve physical reasoning in large language models, enhancing their ability to generate feasible plans without extensive retraining.

Contribution

WorldMind is a novel approach that constructs a symbolic world knowledge repository from environmental feedback, enabling better physical reasoning and transferability in language models.

Findings

Outperforms baseline models on EB-ALFRED and EB-Habitat tasks.

Demonstrates strong transferability across models and environments.

Improves physical feasibility of generated plans.

Abstract

Current Large Language Models (LLMs) exhibit a critical modal disconnect: they possess vast semantic knowledge but lack the procedural grounding to respect the immutable laws of the physical world. Consequently, while these agents implicitly function as world models, their simulations often suffer from physical hallucinations-generating plans that are logically sound but physically unexecutable. Existing alignment strategies predominantly rely on resource-intensive training or fine-tuning, which attempt to compress dynamic environmental rules into static model parameters. However, such parametric encapsulation is inherently rigid, struggling to adapt to the open-ended variability of physical dynamics without continuous, costly retraining. To bridge this gap, we introduce WorldMind, a framework that autonomously constructs a symbolic World Knowledge Repository by synthesizing environmental feedback. Specifically, it unifies Process Experience to enforce physical feasibility via prediction errors and Goal Experience to guide task optimality through successful trajectories. Experiments on EB-ALFRED and EB-Habitat demonstrate that WorldMind achieves superior performance compared to baselines with remarkable cross-model and cross-environment transferability.