Reinforcement Learning for Mobile Robot Obstacle Avoidance Under Dynamic Environments

Collision avoidance under dynamic environments is a challenging problem for mobile robots. Navigating the robot safely to the target is extremely significant especially in the dynamic environments. In this paper, a new approach based on reinforcement learning is proposed to navigate the robot from the start location to the target location without collisions with static and dynamic obstacles. In the proposed method, we improve the original Q-learning algorithm in environment modeling, reward function, and the adapted policy to make the robot stay away from obstacles, reduce the probability of collisions, and reach the target as fast as possible. Finally, simulations of some test scenarios and the comparisons between the original Q-learning and improved Q-learning are respectively conducted to validate that the proposed approach has high efficiency and adaptability in solving dynamic obstacle avoidance problem.