Transformer World Model for Sample Efficient Multi-Agent Reinforcement Learning

arXiv 25.6 2025 58.9 method

TLDR

A transformer-based world model for multi-agent RL that achieves state-of-the-art sample efficiency via decentralized imagination and teammate prediction.

Reasoning

The paper presents a novel architecture with strong empirical results across multiple benchmarks and ablation studies, but the abstract lacks discussion of limitations or comparisons to other world model variants.

Read-first score

Read-first score 58.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 49.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Topical relevance 42%

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

Methodology quality 25%

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

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 364.

Keyword Scores

world model

world dynamics prediction

model-based reinforcement learning world model

interactive world model

world simulator

generative world model

video world model

Deep Analysis

Innovations

Multi-Agent Transformer World Model (MATWM) combining decentralized imagination with semi-centralized critic
Teammate prediction module for modeling and anticipating others' behavior under partial observability
Prioritized replay mechanism to address non-stationarity by training on recent experiences

Methodology

MATWM is a transformer-based world model for multi-agent reinforcement learning that uses a decentralized imagination framework, a semi-centralized critic, and a teammate prediction module. It incorporates a prioritized replay mechanism to adapt to evolving policies. The model is evaluated on vector- and image-based environments including StarCraft Multi-Agent Challenge, PettingZoo, and MeltingPot.

Key Results

MATWM achieves state-of-the-art performance, outperforming both model-free and prior world model approaches, and demonstrates strong sample efficiency with near-optimal performance in as few as 50K environment interactions.

Resources

Paper