MARLadona-Towards Cooperative Team Play Using Multi-Agent Reinforcement Learning


Date

2025

Publication Type

Conference Paper

ETH Bibliography

yes

Citations

Scopus:
Altmetric

Data

Abstract

Robot soccer, in its full complexity, poses an unsolved research challenge. Current solutions heavily rely on engineered heuristic strategies, which lack robustness and adaptability. Deep reinforcement learning has gained significant traction in various complex robotics tasks such as locomotion, manipulation, and competitive games (e.g., AlphaZero, OpenAI Five), making it a promising solution to the robot soccer prob lem. This paper introduces MARLadona. A decentralized multi agent reinforcement learning (MARL) training pipeline capable of producing agents with sophisticated team play behavior, bridging the shortcomings of heuristic methods. Furthermore, we created an open-source multi-agent soccer environment. Utilizing our MARL framework and a modified global entity encoder (GEE) as our core architecture, our approach achieves a 66.8 % win rate against HELIOS agent, which employs a state-of-the-art heuristic strategy. In addition, we provided an in-depth analysis of the policy behavior and interpreted the agent’s intention using the critic network.

Publication status

published

Editor

Book title

2025 IEEE International Conference on Robotics and Automation (ICRA)

Journal / series

Volume

Pages / Article No.

15014 - 15020

Publisher

IEEE

Event

42nd IEEE International Conference on Robotics and Automation (ICRA 2025)

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

Organisational unit

02620 - Inst. f. Robotik u. Intelligente Systeme / Inst. Robotics and Intelligent Systems

Notes

Conference lecture held on May 22, 2025

Funding

Related publications and datasets