Deep Reinforcement Learning-Based (DRLB) Optimization for Autonomous Driving Vehicle Path Planning

This comprehensive study presents a novel DRLB assisted path planning (PP) algorithm for the autonomous driving (AD) vehicles with the novel designed deep reinforcement learning (DRL). The proposed algorithm addresses the challenges faced by traditional path planning methods, such as large Q-table sizes, high computational complexity, and convergence problems. Unlike conventional approaches that rely on predefined environment maps or potential field methods, the proposed algorithm uses DRL to learn optimal paths in dynamic and complex environments. The algorithm uses a grid map representation for environmental modeling, where obstacles are enclosed in rectangular frames to ensure path accuracy and safety. Additionally, integrating an artificial potential field approach (APFA) helps guide the vehicle to the target while avoiding obstacles. The algorithm uses a novel Q-learning algorithm to approximate the value function between the vehicle state and action, which enables efficient path planning. The algorithm is visually evaluated through extensive simulation tests, demonstrating its effectiveness and superiority over traditional methods from the perspectives of the accuracy and stability.