Path Planning for Autonomous Vehicles in Unknown Dynamic Environment Based on Deep Reinforcement Learning

Autonomous vehicles can reduce labor power during cargo transportation, and then improve transportation efficiency, for example, the automated guided vehicle (AGV) in warehouse operation efficiency. To overcome limitations of traditional path planning algorithms unknown environments, such as reliance on high-precision maps, lack generalization ability, obstacle avoidance capability, this study focuses investigating Deep Q-Network its derivative algorithm to enhance network structures. A new named APF-D3QNPER is proposed, which combines action output method artificial potential field (APF) with Dueling Double Q Network algorithm, experience sample rewards are considered playback portion Reinforcement Learning (DRL) enhances convergence ability DRL algorithm. long short-term memory (LSTM) added state feature extraction part adaptability environments spatiotemporal sensitivity environment. The compared mainstream deep reinforcement learning using a robot operating system Gazebo simulation platform by conducting experiments. results demonstrate that exhibits excellent abilities environment, speed, loss value, time, length all less than other diverse scenarios.