Smooth Trajectory Collision Avoidance through Deep Reinforcement Learning