Subgoal-Driven Navigation in Dynamic Environments Using Attention-Based Deep Reinforcement Learning