DRL Based on Priority Evaluation for Collaborative Perception Task Offloading in Internet of Vehicles

With the explosive growth of the Internet of Vehicles (IoVs), autonomous driving has received numerous attention from academia and industry. However, the limited perceivable range of a single connected and automated vehicle (CAV) leads to missing real-time road condition information, which affects driving safety. Meanwhile, due to the complexity of applications and dependencies between tasks, perception tasks require high computing capability and low latency. However, a CAV has limited local computing capability, it is necessary to offload perception tasks to edge for processing. Therefore, to improve perception coverage and reduce perception task processing la-tency, we propose a Priority Evaluation and Deep Reinforcement Learning Task Offloading (PDRTO) algorithm. Firstly, a dual layer IoVs architecture for task offloading is constructed to improve the perception coverage of CAVs and achieve safe driving through collaborative perception among CAVs. At the same time, considering task dependency and CAV's mobility, a multi-CAV collaborative perception task graph is proposed to model the optimization problem of minimizing the processing delay of perception subtasks. Then, the optimization problem is transformed into a multi-objective mixed integer optimization problem, and solved by the proposed PDRTO algorithm to minimize the processing delay of perception tasks. Finally, the simulation results show that the proposed algorithm outperforms other compared algorithms in improving CAV perception cover-age and reducing perception task latency.