Self-Evolving World Models for LLM Agent Planning

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Xuan Zhang, Wenxuan Zhang, See-Kiong Ng, Yang Deng
• Venue: arXiv: cs.AI
• Published: 2026-06-29
• Collected: 2026-06-30
• URL: http://arxiv.org/abs/2606.30639v1
• PDF: https://arxiv.org/pdf/2606.30639v1
• DOI: UNKNOWN

Plain-Language Brief

The paper introduces a new framework called WorldEvolver for long-horizon agents. This framework improves the accuracy of predictions about action consequences before execution, which is crucial for agents to make informed decisions. It uses three modules: Episodic Memory, Semantic Memory, and Selective Foresight. The framework was tested on ALFWorld and ScienceWorld and showed better results than other world model baselines.

What the Document Says

• The paper presents WorldEvolver, a self-evolving world model framework that revises its deployment-time context while keeping the downstream agent and all model parameters frozen.
• WorldEvolver includes three modules: Episodic Memory, Semantic Memory, and Selective Foresight.
• Episodic Memory uses real action transitions through retrieval-based simulation.
• Semantic Memory extracts persistent heuristic rules from prediction-observation mismatches.
• Selective Foresight filters low-confidence predictions before integrating them into agent reasoning context.
• The framework was evaluated on ALFWorld and ScienceWorld, measuring world model prediction accuracy on Word2World and downstream agent success rate on AgentBoard.
• Extensive experiments show that WorldEvolver achieves the highest prediction accuracy across three backbones and leads other world model baselines on downstream agent success rate.

Key Concepts Mentioned

• World models
• Long-horizon LLM agents
• Foresight
• Predictions of action consequences
• Episodic Memory
• Semantic Memory
• Selective Foresight

Claims Made by the Document

• WorldEvolver enhances both predictive fidelity and planning performance.
• WorldEvolver leads other world model baselines on downstream agent success rate.

Evidence Used

• Results from ALFWorld and ScienceWorld experiments showing higher prediction accuracy and downstream agent success rate for WorldEvolver.
• Comparison with other world model baselines in the same experiments.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

World models offer a principled way to equip long-horizon LLM agents with foresight: predictions of action consequences before execution. However, unreliable foresight can be ignored, misused, or even degrade downstream decision-making. In this paper, we introduce WorldEvolver, a self-evolving world model framework that revises its deployment-time context while keeping the downstream agent and all model parameters frozen. WorldEvolver integrates three modules: (i) Episodic Memory, which exploits real action transitions through retrieval-based simulation; (ii) Semantic Memory, which extracts persistent heuristic rules from prediction-observation mismatches; and (iii) Selective Foresight, which filters low-confidence predictions before integrating them into agent reasoning context. We evaluate WorldEvolver on ALFWorld and ScienceWorld, measuring world model prediction accuracy on Word2World and downstream agent success rate on AgentBoard. Extensive experiments show that WorldEvolver achieves the highest prediction accuracy across three backbones and leads other world model baselines on downstream agent success rate, demonstrating that test-time memory revision enhances both predictive fidelity and planning performance.