Imitating Opponent to Win: Adversarial Policy Imitation Learning in Two-player Competitive Games

Recent research on vulnerabilities of deep reinforcement learning (RL) has shown that adversarial policies can influence a target RL agent (victim agent) to perform poorly. In existing studies, adversarial policies are directly trained based on experiences of interacting with the victim agent. A key shortcoming of this approach is that knowledge derived from historical interactions may not be properly generalized to unexplored policy regions of the victim agent, making the trained adversarial policy significantly less effective. In this work, we design a new effective adversarial policy learning algorithm that overcomes this shortcoming. The core idea of our new algorithm is to create a new imitator-the imitator will learn to imitate the victim agent's policy while the adversarial policy will be trained based on both interactions with the victim agent and feedback from the imitator to forecast victim's intention. By doing so, we can leverage the capability of imitation learning in well capturing underlying characteristics of the victim policy. Our victim imitation learning model differs from prior models as the environment's dynamics are driven by adversary's policy and will keep changing during the adversarial policy training. We provide a provable bound to guarantee a desired imitating policy when the adversary's policy becomes stable. We further strengthen our adversarial policy learning by incorporating the opposite of the adversary's value function to the imitation objective, leading the imitator not only to learn the victim policy but also to be adversarial to the adversary. Finally, our extensive experiments using four competitive MuJoCo game environments show that our proposed algorithm outperforms state-of-the-art algorithms. 1