Active Perception in Adversarial Scenarios using Maximum Entropy Deep Reinforcement Learning