Traditional robust methods in multi-agent reinforcement learning (MARL) often struggle against coordinated adversarial attacks in cooperative scenarios. To address this limitation, we propose the Wolfpack Adversarial Attack framework, inspired by wolf hunting strategies, which targets an initial agent and its assisting agents to disrupt cooperation. Additionally, we introduce the Wolfpack-Adversarial Learning for MARL (WALL) framework, which trains robust MARL policies to defend against the proposed Wolfpack attack by fostering system-wide collaboration. Experimental results underscore the devastating impact of the Wolfpack attack and the significant robustness improvements achieved by WALL. Our code is available at https://github.com/sunwoolee0504/WALL.

AI systems where many agents work together are used in real-world settings, but unexpected problems like sensor errors or lost signals can break their coordination. Some training methods add small disturbances to help agents prepare, but these attacks usually target only one agent, making it easy for others to adapt by helping. Inspired by how wolves hunt in packs, we introduce a “Wolfpack” attack that targets one agent and the teammates who try to help, making it harder for the team to stay organized. We also propose a method to select critical moment to apply the attack during training. In addition, we introduce WALL, a training approach that helps agents build flexible teamwork and recover quickly. Our approach enables more adaptive cooperation and resilience in real-world situations.