Evolution Of Dominant Driving Strategies For Large-scale Autonomous Vehicles

With the development of economy and the expansion of urban scale,the scale and load of traffic are increasing.How to increase the road efficiency and safety becomes prominent questions.In order to improve the overall performance of transportation,multiple optimization objectives,such as traffic accidents,traffic jams and road capacity,need to be considered comprehensively.Optimization considering only one indicator may cause degradation of other indicators.On the other hand,autonomous vehicles and intelligent transportation technologies provides possibilities to solve the comprehensive urban traffic problems,through interactions between vehicles,where cooperation and competition are everywhere could be exploited to optimize the overall traffic performance.Tt is necessary to study the cooperative action between vehicles in order to discover the dominated driving strategy of autonomous vehiclesAiming at the multi-objective driving strategy optimization problem of large-scale autonomous vehicles,a driving strategy optimization method based on dynamic evolution is proposed in this paper.This method combines multi-objective optimization and replicator dynamic model to explore the learning-imitation-selection process of autonomous vehicles,and to solve the population state with optimal overall utility by means of dynamic evolution.In the process of dynamic evolution,autonomous vehicles observe the behavior of other agents through the game and choose a better driving strategy by comparing their utilities.Finally,the population state gradually converges to the evolutionary equilibrium point,at which time the dominant strategy emerges and the overall traffic situation is improved.Through simulation experiments,this work gets the dynamic evolution results under different traffic conditions.This work imports over 40,000 virtual autonomous vehicles and the real city topologies into the SUMO traffic simulation platform.Simulation experiments show that under different traffic conditions,the traffic system will converge to different stable states after dynamic evolution.Under the condition of low-density traffic flow,the aggressive strategy and rational strategy coexist in a certain proportion,while the conservative strategy disappears.However,under the condition of high-density traffic flow,the rational strategy occupies the whole system,while both aggressive and conservative strategies die out.After the dynamic evolution,the utility of the system is improved in different degrees,in which the overall average speed is increased by 8% ～ 16%,and the accident rate is reduced by 10% ～ 78%.Finally,according to the characteristics of the replicator dynamic model,we prove that the population state obtained by the dynamic evolution experiment is the Nash equilibrium,and the stable solution of the multi-objective optimization problem can be obtained by using the driving strategy dynamic evolution algorithm.This work could help improve the driving strategy design,from the perspective of traffic management.