Agent2World: Learning to Generate Symbolic World Models via Adaptive Multi-Agent Feedback

TL;DR

Agent2World presents a tool-augmented, multi-agent framework that generates symbolic world models from natural language by orchestrating a Deep Researcher, a Model Developer, and a Testing Team in a three-stage loop. The knowledge-synthesis, implementation, and evaluation-driven refinement stages are coupled with an agent-in-the-loop formalization and verifier-guided rejection sampling, producing multi-turn trajectories for training. Across Text2World, CWMB, and ByteSized32, the approach yields state-of-the-art inference performance and substantial gains from supervised fine-tuning, including a reported average relative gain of 30.95% in world-model quality after training. The work demonstrates that interactive, behavior-aware feedback can transform LLMs into more reliable generators of executable world models, with clear implications for scalable, automated environment synthesis and planning. A dedicated dataset of 1526 high-quality verified trajectories is introduced to support supervised fine-tuning, and extensive ablations underscore the complementary roles of knowledge synthesis, automated testing, and repair in achieving robust, executable world models.

Abstract

Symbolic world models (e.g., PDDL domains or executable simulators) are central to model-based planning, but training LLMs to generate such world models is limited by the lack of large-scale verifiable supervision. Current approaches rely primarily on static validation methods that fail to catch behavior-level errors arising from interactive execution. In this paper, we propose Agent2World, a tool-augmented multi-agent framework that achieves strong inference-time world-model generation and also serves as a data engine for supervised fine-tuning, by grounding generation in multi-agent feedback. Agent2World follows a three-stage pipeline: (i) A Deep Researcher agent performs knowledge synthesis by web searching to address specification gaps; (ii) A Model Developer agent implements executable world models; And (iii) a specialized Testing Team conducts adaptive unit testing and simulation-based validation. Agent2World demonstrates superior inference-time performance across three benchmarks spanning both Planning Domain Definition Language (PDDL) and executable code representations, achieving consistent state-of-the-art results. Beyond inference, Testing Team serves as an interactive environment for the Model Developer, providing behavior-aware adaptive feedback that yields multi-turn training trajectories. The model fine-tuned on these trajectories substantially improves world-model generation, yielding an average relative gain of 30.95% over the same model before training. Project page: https://agent2world.github.io.

Figures (8)

• Figure 1: Comparison of Agent2World and previous world-model generation paradigms.
• Figure 2: Overall Pipeline of Agent2World.
• Figure 3: Ablation Study on CWMB.
• Figure 4: Pair-wise Win–Tie–Loss analysis.
• Figure 5: Primary performance evolution across successive rounds of testing team feedback (§ \ref{['sec:edr']}.)