跳转至

文章背景与核心概要

本文介绍了名为 CoMAP 的全新框架，旨在通过闭环交互实现文本世界模型与智能体策略的同步（共同演化）发展，从而克服传统大语言模型（LLM）智能体在动态环境中适应性不足的缺陷。传统语言智能体通常依赖固定的世界模型或外部验证器，而 CoMAP 引入了未来感知反思机制与自蒸馏技术，使世界模型能够随智能体策略的演化而不断动态更新。

该框架在具身任务规划、网页导航和工具使用等多个基准测试中表现优异，使用 Qwen3-4B 模型取得了高达 +16.75% 的相对性能提升。实验分析表明，这种共同演化循环不仅显著提高了世界模型随时间推移的预测准确率，还赋能智能体实现了更高效的长视距决策。

CoMAP: Co-Evolving World Models and Agent Policies for LLM Agents

Summary

CoMAP is a novel framework designed to improve Large Language Model (LLM) agents by enabling the simultaneous (co-evolutionary) development of textual world models and agent policies through closed-loop interaction.

Traditional language agents rely on fixed world models or external verifiers/rewards, limiting their adaptability in dynamic environments. In contrast, CoMAP features: * Future-Aware Reflection: The agent estimates the reliability of feedback from the world model and refines its actions accordingly at each decision step. * Self-Distillation: On-policy trajectories are continuously used to update the world model, allowing it to adapt to the agent's evolving interaction distribution. * Superior Performance: Outperforms competitive baselines across various tasks—such as embodied task planning, Web navigation, and tool-use benchmarks—achieving a +16.75% relative improvement using Qwen3-4B.

Paper Metadata & Overview

论文元数据与概述

  • arXiv ID: arXiv:2606.02372 [cs.AI]
  • Authors: Youwei Liu, Jian Wang, Hanlin Wang, Wenjie Li
  • Primary Subject: Artificial Intelligence (cs.AI)
  • Secondary Subject: Computation and Language (cs.CL)
  • Conference Status: Accepted by EMNLP 2026 Main Conference
  • Submitted: June 1, 2026; Last Revised: September 3, 2026
  • Code Repository: GitHub - loyiv/CoMAP

Abstract

摘要

Equipping language agents with world models enables them to anticipate environment dynamics and evaluate candidate actions before execution. However, existing textual world models are typically fixed after training, preventing them from adapting to the on-policy state-action distributions induced by an evolving agent. Meanwhile, agent-improvement methods often rely on external rewards or verifiers, limiting their applicability in realistic interactive environments.

In this paper, we propose CoMAP, a novel framework that co-evolves textual world models and agent policies through closed-loop interaction. At each decision step, the world model predicts future state feedback for candidate actions, and the agent performs future-aware reflection by estimating the reliability of this feedback and refining its action accordingly. The resulting on-policy trajectories are then used to update the world model via self-distillation, allowing it to better match the agent's evolving interaction distribution.

Across embodied task planning, Web navigation, and tool-use benchmarks, CoMAP consistently outperforms competitive baselines, e.g., +16.75% relative improvement with Qwen3-4B. Further analyses show that the co-evolutionary loop improves the world model's prediction accuracy over time and leads to more effective long-horizon decision-making.

Access & Resources

访问与资源

Note: Preserved image asset hook from the original source layout: 注：保留了原始源布局中的图像资源钩子： license icon