Research On Edge-based Collaborative Computing For Autonomous Driving

Nowadays,autonomous driving is one of the most popular technologies and is in a stage of rapid development.In the process of working,autonomous vehicles need to locate,perceive,predict and plan.The enormous amount of calculation results high energy consumption,which poses a challenge of endurance and heat dissipation.If other computing platforms can collaboratively compute with autonomous vehicles to reduce the amount of calculation on vehicles,it will save energy consumption of vehicles and effectively solve the problem of endurance and heat dissipation.Autonomous driving applications are strict with execution time – most results have to be returned in hundreds of or even tens of milliseconds.Due to the large network delay,it is difficult for the central cloud to collaboratively compute with autonomous vehicles.Autonomous driving has the need of edge-based collaborative computing.With the rise of mobile edge computing(MEC)technology and 5G technology,computing resources and storage resources can be deployed to MEC servers at the edge of the network.The network transmission rate will increase greatly.MEC server can collaboratively compute with autonomous vehicles.Computing offloading is an implementation of collaboratively computing.MEC servers can achieve edge-based collaborative computing by providing computing offloading service.However,existing computing offloading mechanism cannot simultaneously meet the multidimensional service requirements of autonomous driving in safety,privacy and efficiency.In order to solve the above problems,we studied how to realize edge-based collaborative computing for autonomous driving through MEC computing offloading.The main work of this paper is as follows:1.This paper proposes a Docker-based MEC computing offloading mechanism and designs the offloading service middleware.The service middleware has following characteristics:(1)Docker images is managed by image manager.Autonomous vehicles only need to transmit application type information and calculation data to complete offloading.(2)Resource manager is used to restrict container resources.It flexibly limits the CPU usage of the applications by setting CPU proportions.(3)Container manager is used to create and destroy containers and to speed up the offloading by using pre-launch container policy.Ultimately,the MEC servers are able to provide isolated operating environment for offloading applications while maintaining offloading efficiency.Therefore,MEC servers can effectively coordinate with the autonomous vehicles while meeting the safety and privacy requirements of autonomous driving.2.Taking energy consumption as optimization objective and taking response time as restriction condition,this paper discusses the offloading decision-making problem and offloading scheduling problem for resource-constrained MEC servers.For the offloading decision-making problem,this paper gives the offloading decision-making process considering the factors of network transmission speed,computation and remaining resources.For the offloading scheduling problem,this paper discusses single-application scheduling and multi-application scheduling respectively.In this paper,multi-application scheduling is formalized as multiple multidimensional knapsack problem.Since the problem is NP-complete,this paper proposes a greedy algorithm to solve it in order to guarantee the scheduling efficiency.The experimental results show that the offloading service middleware proposed in this paper has high feasibility and efficiency while guaranteeing the isolation of autonomous driving applications.It can realize computing offloading of autonomous driving applications at millisecond level.This paper verifies the practicability of scheduling strategy based on greedy algorithm through comparative experiments.