Budget-Efficient Attacks and Robustness Training for Cooperative MARL

Cooperative multi-agent reinforcement learning (CMARL) policies are vulnerable to action hijacking even when only a few timesteps are compromised. Recent adversarial attacks and adversarial training methods have been explored, but under an explicit attack budget, existing attacks often fail to accurately expose critical coordination weaknesses and incur substantial training cost. We propose Budgeted Hierarchical Efficient Attack (BHEA), a budgeted hierarchical adversarial attack that separates decisions on when and which agents to hijack from action replacement, enabling more precise vulnerability discovery under limited attack opportunities. We further show that training cooperative policies against BHEA substantially improves robustness to limited-step action hijacking while reducing training overhead. Experiments on the StarCraft Multi-Agent Challenge (SMAC) demonstrate stronger attacks under the same attack budget and improved robustness. Code is available at https://anonymous.4open.science/r/BHEA-068D.

This paper studies how teams of AI agents can fail when an attacker is able to interfere with only a small number of their actions. In many real-world settings, multiple AI agents need to work together, and even brief disruptions can cause poor coordination. We introduce a new method, called BHEA, that helps identify the most important moments and agents to target when testing such systems. This makes it easier to find hidden weaknesses under a limited attack budget. We also show that training AI teams against these more focused attacks makes them more reliable, while requiring less training cost. Experiments in a challenging multi-agent game environment show that our method finds stronger weaknesses and helps produce more robust cooperative behavior.