Research on a Dynamic Path Planning Algorithm for Robots Based on Deep Reinforcement Learning

Abstract

With the advancement of robotic technology, robot path planning has garnered significant attention, particularly in fields such as automation, autonomous driving, and smart manufacturing. Path planning involves determining the optimal path for a robot to travel from a specific position to its destination while avoiding obstacles. Traditional path planning algorithms can effectively realize robot path planning in a static environment where obstacles do not move. In practice, some environments can change rapidly, and unexpected obstacles may arise. Traditional path planning algorithms often struggle to optimize paths in dynamic and complex real-world applications adaptively. Hence, the performance of traditional path planning algorithms is limited. This paper presents a robot path planning algorithm in a complex dynamic environment using deep reinforcement learning (DRL). Using Proximal Policy Optimization (PPO), reinforcement learning enables the robot to learn path planning methods by simulating its interactions with its environment. As the robot runs autonomously, continuously improving its experience, our proposed dynamic path planning can adaptively optimize paths according to the movement of obstacles. The experimental results confirm that the proposed DRL-based approach enables the robot to reach the destination faster by avoiding moving obstacles and achieving a better performance than conventional path planning approaches when the surrounding environment fluctuates unexpectedly. Therefore, the experiments suggest that the proposed algorithm enhances robots' ability to adapt to complex dynamic environments across a broad range of applications, including, but not limited to, autonomous vehicles, mobile robots, and automated industrial systems.