Interaction-Breaking Adversarial Learning Framework for Robust Multi-Agent Reinforcement Learning

Sunwoo Lee, Mingu Kang, Yonghyeon Jo, Seungyul Han

International Conference on Machine Learning (ICML), 2026

Paper Code BibTeX

Abstract

Cooperation is central to multi-agent reinforcement learning (MARL), yet learned coordination can be fragile when external perturbations disrupt inter-agent interactions. Prior robust MARL methods have primarily considered value-oriented attacks, leaving a gap in robustness when interaction structures themselves are corrupted. We propose Interaction-Breaking Adversarial Learning (IBAL), an information-theoretic framework that constructs attacks by perturbing agents’ observations and actions to impede coordination, and trains agents to perform reliably under such disruptions. Experiments show that IBAL improves robustness over existing robust MARL baselines across diverse attack settings and yields stronger performance under agent-missing scenarios.

BibTeX

@inproceedings{lee2026ibal,
  title={Interaction-Breaking Adversarial Learning Framework for Robust Multi-Agent Reinforcement Learning},
  author={Lee, Sunwoo and Kang, Mingu and Jo, Yonghyeon and Han, Seungyul},
  booktitle={International Conference on Machine Learning},
  year={2026}
}