Reverse Computing Offloading for Enhanced Computing Capacity in Cooperative Vehicle Infrastructure System

Cooperative Vehicle-Infrastructure System (CVIS) and Vehicular Edge Computing (VEC) have brought more possibilities for the intelligent transportation system. In CVIS, the amount of data generated by sensors need to be uploaded to VEC server for information fusion. However, the limited computation and storage resources of VEC server can not satisfy the computing requirements of ever-increasing sensing data, which are generated by vehicles and infrastructure. In this work, we propose a reverse computing offloading framework to reduce the burden of VEC servers in a multi-vehicle mobile edge network, by making full use of the computing resources of vehicles. We aim to maximize the system service carrying capacity while satisfying the average queuing delay and power consumption constraints. Considering the random task arrivals, time-varying channel and task queue status of vehicles and VEC server, the system service carrying capacity maximization problem is formulated as a constrained Markov decision process to obtain the offloading decisions. Next, we transform the CMDP problem into an equivalent min-max non-constrained MDP problem through the Lagrangian approach. In addition, a Q-Learning (QL)-based offloading policy learning algorithm is proposed to realize the joint optimization of reverse offloading decision, communication and computing resource allocation of the VEC server and vehicles. Simulation results show that the proposed scheme outperforms the other baseline schemes.