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

2026-05-18

AI Summary

Collaborative strategies in multi-agent reinforcement learning (MARL) are highly vulnerable when the interaction structure is perturbed, yet existing robustness methods primarily address value-targeted attacks and struggle to handle scenarios involving disrupted interactions. This work addresses this gap by formulating such interaction-based attacks through an information-theoretic lens and proposing a novel adversarial learning framework for robust training. The approach simulates interaction disruptions by perturbing agents’ observations and actions, then optimizes collaborative policies under these adversarial conditions. By explicitly modeling structural fragility in agent interactions, the method establishes the first robustness solution tailored to interaction-structure damage in MARL. Empirical results demonstrate substantial performance gains over current baselines across diverse attack settings, with consistent resilience even under extreme conditions such as partial agent dropout.

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. In this paper, we propose an interaction-breaking adversarial learning (IBAL) framework that takes an information-theoretic view to construct attacks that impede coordination by perturbing agents' observations and actions, and trains agents to perform reliably under such disruptions. Empirically, our approach improves robustness over existing robust MARL baselines across diverse attack settings and yields stronger performance even under agent-missing scenarios.

Problem

• multi-agent reinforcement learning
• robustness
• interaction disruption
• adversarial attacks
• coordination

Innovation

• interaction-breaking
• adversarial learning
• multi-agent reinforcement learning
• robustness
• information-theoretic