Efficient Generation of Diverse Cooperative Agents with World Models

arXiv 25.6 2025 47.9 method, application

TLDR

XPM-WM uses world models to generate simulated trajectories, improving sample efficiency and scalability for diverse cooperative agent generation in Zero-Shot Coordination.

Reasoning

The paper addresses a key bottleneck in ZSC training by replacing costly trajectory sampling with simulated trajectories from a learned world model, showing comparable performance with higher efficiency. However, the abstract lacks real-world validation and does not discuss limitations of the world model's fidelity or generalization.

Read-first score

Read-first score 47.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 39.

Recency 8%

86.7

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

Topical relevance 42%

55.7

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

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=none

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: 477.

Keyword Scores

world model

model-based reinforcement learning world model

world dynamics prediction

world simulator

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Using a learned world model to generate simulated trajectories for Cross-play Minimization (XPM) in Zero-Shot Coordination (ZSC) training
Removing the need to sample multiple types of trajectories from the real environment during XPM
Achieving significantly higher sample efficiency and scalability to larger numbers of partner agents

Methodology

The proposed XPM-WM framework trains a world model to simulate environment dynamics, then uses it to generate simulated trajectories for cross-play minimization. This eliminates the requirement for sampling multiple real trajectories, reducing computational cost and sample inefficiency.

Key Results

XPM-WM matches the performance of previous XPM methods in terms of self-play population training reward and training partners for ZSC agents, while being significantly more sample efficient and scalable to a larger number of partners.

Resources

Paper