Research On Task Offloading Theory And Method Of Vehicle Collaborative Computing For Autonomous Vehicle Perception Data

In the computing-intensive and time-sensitive connected automated vehicles network,the computing task offloading service based on mobile edge computing can provide sufficient computing resources for vehicle users with perceptual tasks.However,the existing work does not take into account the low delay requirements of original perceptual data sharing,as well as the constraints of wireless communication link capacity and computing efficiency,which seriously threatens the security of cooperative autopilot in CAVs network.This paper studies the theory and method of task offloading of vehicle road collaborative computing for autonomous vehicle perception data,and proposes a centralized vehicle road collaborative computing task offloading method based on reinforcement learning to reduce the execution delay of task offloading.In addition,a task offloading and resource allocation scheme based on federated reinforcement learning is designed to reduce the communication transmission overhead for the autonomous driving scenario of the Internet of vehicles with different vehicle communication computing and computing abilities.The main contributions of this paper are as follows:(1)aiming at the problem that the low delay requirements of vehicle task offloading can not be met in the computing-intensive and delay-sensitive autonomous driving scenarios,a vehicle-road-base station cooperative architecture is designed.At the same time,considering the random traffic and communication uncertainty in the vehicle environment,a deep reinforcement learning scheme based on adaptive exploration is proposed.The test results show that the proposed scheme can not only improve the sensing performance of the receiver,but also improve the system throughput by 88%on the basis of reducing the task execution delay.(2)Aiming at the high overhead caused by the centralized reinforcement learning algorithm in the autonomous driving scenario of Internet of vehicles with limited bandwidth resources,a task offloading and resource allocation scheme of federated reinforcement learning is designed to reduce the communication transmission overhead.Aiming at the model weighting problem in the federated aggregation process in the heterogeneous vehicle networking scenario with different vehicle communication and computing abilities,a dual weight parameter weighting scheme based on the difference of heterogeneous device communication and computing abilities is designed to solve the problem of uneven learning state in heterogeneous device learning,which can improve the accuracy of the training model by 18.4%,and improve the efficiency of model training by 69.2%.