TOOA-DDPG: A Joint Optimization Method of Task Offloading and Obstacle Avoidance Strategy for Autonomous Driving Vehicles Based on DDPG

Autonomous driving systems use online path planning with global information to determine optimal routes. However, during movement between adjacent nodes, vehicle dynamics and kinematic constraints must be considered for effective obstacle avoidance. This paper proposes an online optimization algorithm for task offloading and obstacle avoidance, named TOOA-DDPG, based on Deep Deterministic Policy Gradient network (DDPG). This algorithm jointly optimizes task offloading and obstacle avoidance, enhancing both avoidance efficiency and overall travel performance. The optimization problem involves a complex action space with continuous and discrete variables, further complicated by dynamic traffic and fluctuating wireless conditions. TOOADDPG efficiently handles high-dimensional continuous action spaces by directly outputting deterministic actions, avoiding exhaustive search. Simulations confirm its accuracy and effectiveness.