IMAP: 内在动机驱动的对抗策略（英文版）

英文标题：IMAP: Intrinsically Motivated Adversarial Policy中文摘要：提出了一种名为 Intrinsically Motivated Adversarial Policy（简称 IMAP） 的策略，用于在无需了解受害者策略的情况下，有效地进行黑盒攻击，IMAP 利用基于状态覆盖、策略覆盖、风险和策略差异的四个内在目标，鼓励探索和发现更强的攻击技能，并进行了实验验证其有效性。英文摘要：Reinforcement learning (RL) agents are known to be vulnerable to evasionattacks during deployment. In single-agent environments, attackers can injectimperceptible perturbations on the policy or value network's inputs or outputs;in multi-agent environments, attackers can control an ad