Dynamic Obstacle Avoidance with Bounded Rationality Adversarial Reinforcement Learning