Research On Intelligent Vehicle Autonomous Lane Changing System Considering Traffic Situation

Intelligent vehicle autonomous lane changing system in complex traffic road environment is one of the research hotspots in the automotive industry.Lane changing is a complex vehicle behavior,which will not only affect the safety,efficiency and comfort of the vehicle itself,but also greatly affect the performance of the whole transportation system as a part of the traffic flow.Therefore,it is of great significance to develop a set of intelligent vehicle autonomous lane changing system to meet multi-objective optimization.Based on the sub-project of National Key Research and Development Program: ‘Collaborative Perception and Target Tracking Technology in Complex Road Environment'(No.2017YFB0102601).Aiming at many common problems existing in the autonomous lane changing system of intelligent vehicles at present,this paper built a dynamic occupancy grid probability map to realize the semantic expression of lane-changing driving environment,and a prediction model of vehicle interaction behavior under dynamic traffic situation was established.On the basis of comprehensive consideration of safety,efficiency and comfort,a set of multi-objective optimization autonomous lane-changing decision-making and control strategy was constructed.The research of this paper was carried out from three aspects: ‘environment perception and scene modeling',‘intelligent decision-making and path planning',‘trajectory tracking and intelligent control'.The main contents of this paper include:1)Vehicle behavior trajectory prediction in lane changing process of intelligent vehicle.Based on the current classic prediction algorithm Long Short-Term Memory(LSTM)neural network model,the residual attention mechanism was introduced,and a bottom-up top-down residual attention module was fused before the convolution pool layer of the neural network.Therefore,an improved res-LSTM(residual attention LSTM)trajectory prediction algorithm was proposed by calculating the weight coefficient of the influence of surrounding vehicles on the future driving trajectory of the target vehicle,The accuracy of trajectory prediction was improved.The NGSIM(Next Generation Simulation)data set was used to train and analyze the algorithm,and the accuracy of vehicle lane changing trajectory prediction was verified.2)Construction of dynamic occupancy grid lane change probability map.Based on the prediction of vehicle trajectory,the environmental information of driving state,lane line,obstacle grid occupancy,obstacle vehicle behavior prediction and other influencing factors were mapped to a grid discretized semantic map,and a four-layer dynamic occupancy grid lane-changing probability map was built,which realized the effective expression on the environment semantic map.Through the construction of the dynamic probability map,the processing method of dynamic grid discretization can be used to provide accurate lane-changing driving probability information for autonomous lane-changing intelligent decision-making.3)Intelligent vehicle lane change decision based on lane change trigger and situation change judgment.This paper taken the expected accumulation degree of vehicle lane change as the lane change trigger condition,and proposed a calculation method of lane change safety area based on the combination of occupancy grid and ‘rectangular clustering collision cone model'.The vehicles were clustered and expanded,and the lane change area was calculated according to the collision constraints derived from the collision cone principle.Based on the above research,combined with dynamic occupancy probability map,the decision-making process of autonomous lane changing of intelligent vehicles has been formed.4)Lane changing path planning and tracking control.Considering lane change safety,efficiency and comfort,a ‘lane change cost factor' lane change trajectory evaluation method was proposed.The optimal lane change trajectory was determined through the secondary screening mechanism of ‘suboptimal trajectory+optimal trajectory'.The desired trajectory tracking control adopted the model predictive control algorithm of the state space model,and the model comprehensively considered the constraints such as lateral acceleration and constant swing angular velocity in the process of lane changing.Rewards and punishments were given to the state difference,control increment and control quantity in the objective function,and the relaxation factor was added to realize the effect of autonomous lane changing trajectory tracking of intelligent vehicles.5)Car Sim/Simulink joint simulation and verification of real vehicle test platform.The simulation analysis and real vehicle verification of the autonomous lane changing algorithm designed in this paper were carried out under typical working conditions respectively,and the results showed that the proposed Res-LSTM model has more advantages than the traditional LSTM algorithm in complex road environment.The predicted trajectory is more in line with the real driving situation.The dynamic occupancy grid lane-changing probability map can realize accurate lane-changing decision.The two-level lane-changing trajectory screening mechanism of‘lane-change cost factor' can complete the effective screening of the optimal trajectory,and the tracking control of the lane-change trajectory was carried out by the model predictive control algorithm of the state space model,and the tracking error was controlled within a reasonable range.