Vehicular Edge Computing for Multi-Vehicle Perception

Autonomous vehicle systems require sensor data to make crucial driving and traffic management decisions. Reliable data as well as computational resources become critical. In this paper, we develop a Vehicular Edge Computing Scheduling Pipeline for connected and autonomous vehicles (CAVs) exploring scheduling optimization, pipeline design and vehicle to edge interactions. Through our pipeline, the data, generated by on-board sensors, is used towards various edge serviceable tasks. Due to the limited view of a vehicle, sensor data from one vehicle cannot be used to perceive road and traffic condition of a larger area. To address this problem, our pipeline facilitates data transfer and fusion for cooperative object detection of multiple vehicles. Through real-world experiments, we evaluate the performance and robustness of our pipeline on different device architectures and under different scenarios. We demonstrate that our pipeline achieves a real-time deadline capable edge to vehicle interaction via vehicle-edge data transfer and on-edge computation.