Research On Behavior Decision-Making Method For Smart Driving

Smart driving integrates environment perception,behavior planning and decision-making,and motion control.By using machines to assist or replace human drivers in driving vehicles,it not only reduces the human consumption in transportation issues,but also improves the safety and efficiency of road traffic.In smart driving,driving behavior decision is the key to achieve legal and reasonable driving for smart vehicles in the traffic environment.This thesis focuses on the driving behavior decision of smart vehicles.Aiming at the traffic scenarios that occur during the development of smart driving,the driving behavior decision method of a single smart vehicle in dynamic traffic is first studied,and then the cooperative driving behavior decision method of multiple smart vehicles is studied.The main contents of this thesis include:For a single smart vehicle in the hybrid driving scenario or the fully autonomous driving scenario,reinforcement learning(RL)theory is applied to study its behavior decision,in order to achieve safe,comfortable,and efficient driving.Firstly,an RL model is established.Then,a method of single vehicle’s behavior decision is proposed based on the actor-critic mechanism.Finally,real traffic data is introduced to build a hybrid driving scenario,and the performance of the proposed method is evaluated and analyzed.The simulation results show that in the proposed method,the value of collision cost and unstable cost directly affect the driving safety and comfort of the smart vehicle;there is an optimal number of nearby vehicles whose driving status is takes into account when the smart vehicle makes its driving behavior decisions;the maximum speed limit of smart vehicle affects the comfort and efficiency while guaranteeing the driving safety.For multiple smart vehicles in the fully autonomous driving scenario,a multi-vehicle cooperative behavior decision is proposed based on the policy gradient theory in RL.Firstly,vehicle clusters are established for vehicles on the road in order to simplify the decision complexity and the driving priorities of the vehicles in a cluster are assigned.Then,an RL model for all vehicles in each cluster is established,and the optimal policy for cooperative driving in each vehicle cluster is found.Finally,the performance of the proposed method is evaluated and analyzed.Simulation results show that the performance of cooperative driving is related to the driving priority,and a reasonable weighting of driving priority can make the method achieve the best cooperative driving.