Joint perception data caching and computation offloading in MEC-enabled vehicular networks

Abstract

The accelerated use of intelligent vehicles and the advancement of self-driving technologies have posed significant problems to the provision of real-time vehicular services, such as enormous amounts of computation, long transmission delay, and integration of sensor data. In this context, we propose to solve these problems to guarantee the Quality of Services (QoS) using computation offloading and perception data caching methods, where perception data means combinatorial sensor data, including sensor values and corresponding locations in an area. At first, we present a cooperative vehicular network framework with edge computing and sensor devices. Taking into account vehicle mobility, distributed resources, task properties, and user preferences, we jointly formulate a computation offloading strategy and a perception data caching strategy to minimize the average execution latency, which can be considered a Mixed-Integer Non-Linear Programming (MINLP) problem. We first propose a Genetic Algorithm (GA)-based scheme to solve our formulated problem. Then we propose a heuristic named Correlation-Monte Carlo Fast Search (CMCFS) method to obtain an effective strategy with low complexity. Simulation results reveal that both GA and CMCFS achieve less latency than other baseline schemes.