Edge Computing-Assisted Joint Offloading and Migration in Parking Lots Scenarios

Nowadays intelligent vehicles are equipped with increasing resources for computing intensive driving applications. When these vehicles are parked, their unused computing resources can form a large pool of edge computing resources and be provided to support other parked vehicles or mobile devices with needs of offloading computing tasks. While the parked vehicles based edge computing assistance presents a novel computing paradigm, there is a challenging problem of potential computing service disruption as the parked intelligent vehicles (PIVs) leave the parking lots. To address this issue, we propose a joint offloading and service migration framework for vehicular edge computing in parking lots. In the framework, unfinished tasks can be migrated to another PIV or remained until completion to avoid task failures, and the processing results will be transmitted back through multi-hop relays between vehicles. Considering computing resource requirements of tasks and vehicle movement characteristics, we formulate the problem as a Markov Decision Process (MDP) to minimize task completion delays, with offloading stability constraints established for improving service completion sustainability. Due to dynamic multi-user scenarios and heterogeneous vehicular computing capabilities, the problem is difficult to model and solve. We thus propose a joint offloading and migration decision algorithm (JOMD3) based on dueling double deep Q network. Numerical results demonstrate that the proposed framework and algorithm can achieve lower average delay and faster convergence in comparison to baseline algorithms.