A Vehicular Edge Computing-Based Architecture and Task Scheduling Scheme for Cooperative Perception in Autonomous Driving

Abstract

Cooperative perception is an important domain of autonomous driving that helps to improve road safety and traffic efficiency. Nevertheless, the large amount of sensed data and complicated algorithms make storage and computation for autonomous vehicles (AVs) challenging. Furthermore, not every AV needs to individually process all sensed data from other AVs because the environmental information is the same in a small region. Inspired by vehicular edge computing (VEC), where AVs are interconnected with the help of roadside units (RSUs) for better storage and computation capabilities, we propose a VEC-based architecture for cooperative perception and design a key task scheduling algorithm for the above challenges. Specifically, a time slot-based VEC architecture with the help of an RSU is designed, and the task scheduling problem in the proposed architecture is formulated as a multitask multitarget scheduling problem with assignment restrictions. A two-stage heuristic scheme (TSHS) is designed for the problem. Finally, extensive simulations indicate that the proposed architecture with the TSHS can enable cooperative perception, with a fast running speed and advanced performance, that is superior to that of the benchmarks, especially when most AVs face limitations in terms of storage and computation.