⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

Agent World Model: Infinity Synthetic Environments for Agentic Reinforcement Learning

arXiv 26.2 2026 48.1 method

TLDR

Proposes Agent World Model, a synthetic environment generation pipeline for scaling multi-turn tool-use agent reinforcement learning with reliable state transitions.

Reasoning

Strengths: Novel approach to generating diverse synthetic environments for agent training, enabling out-of-distribution generalization. Weaknesses: The term 'world model' is misleading as it does not involve learned dynamics prediction; the contribution is an environment generator, not a predictive model.

Read-first score

Read-first score 48.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 25.

Recency 6%
100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Reproducibility 18%
81

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=True; dataset=False; markers=code,github

Methodology quality 18%
80

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=benchmark,evaluation,experiment

Topical relevance 29%
35.7

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Citation impact 18%
19

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.1898907

Citation velocity 12%
0

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FrontierMethodology anchorReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 716.

Keyword Scores

interactive world model
7
world simulator
6
generative world model
5
world model
4
model-based reinforcement learning world model
2
world dynamics prediction
1
video world model
0

Deep Analysis

Innovations

  • Fully synthetic environment generation pipeline (Agent World Model) that scales to 1,000 diverse everyday scenarios
  • Code-driven environments backed by databases for reliable and consistent state transitions, unlike LLM-simulated environments
  • Efficient agent interaction compared to collecting trajectories from realistic environments
  • Reliable reward functions enabled by fully executable environments and accessible database states

Methodology

The paper proposes Agent World Model (AWM), a pipeline that generates fully synthetic, code-driven environments backed by databases. These environments support multi-turn tool-use interactions and are used to perform large-scale reinforcement learning for agents. Training is conducted exclusively in synthetic environments, and evaluation is done on three benchmarks to test out-of-distribution generalization.

Key Results

Training exclusively in synthetic environments, rather than benchmark-specific ones, yields strong out-of-distribution generalization on three benchmarks.

Limitations

  • Synthetic environments may not fully capture the complexity and stochasticity of real-world scenarios
  • Generalization to unseen real-world environments beyond the 1,000 synthetic ones remains to be further validated

Resources

PaperCode

Tags