Research On Anthropomorphic Lane-changing Decision And Lane Change Trajectory Of Autonomous Vehicles

Compared with other driving behaviors,drivers need to pay attention to the front,rear and lateral traffic scenarios during lane-changing,and make lane-changing decisions based on multi-dimensional information input.For automatic driving vehicles,the on-board computer system also needs to observe the surrounding traffic environment during lane-changing,and make lane-changing decisions according to the traffic environment state and the self-driving motion state,to decide whether to take lane-changing operation and how to do lane-changing.At the present stage,the autonomous lane-changing ability of autonomous driving vehicles is relatively weak,especially the lane-changing decision-making and lane-changing execution mode need to be further improved.Auto-driving vehicles and traditional manual-driving vehicles will be mixed on the road in a certain period of time.Considering the requirements of safety and traffic flow stability,it is better to achieve the consistency of the two types of vehicle behavior.Obviously,it is unrealistic to transform the driver into a machine-like driving mode,and it is more realistic to extract the driver’s driving characteristics and input them into the computer to realize the humanoid driving of the automatic driving vehicle.Take lane-changing as an example,lane-changing is usually divided into normal lane-changing and emergency avoidance lane-changing.In this study,aiming at the normal lane-changing process of automobile,aiming at ensuring the safety and comfort of lane-changing of automobile,by learning the lane-changing decision-making and lane-changing trajectory characteristics of human drivers,the lane-changing decision-making and lane-changing trajectory model of automobile in non-emergency avoidance are established to realize the humane lane-changing of automobile.In order to meet the above requirements,this study builds an experimental platform on a small passenger car and conducts normal driving experiments in real road environment.Based on radar,visual sensors and other instruments,a large number of lane-changing data of 43 drivers were obtained.Based on these lane-changing data,aiming at realizing the decision-making and trajectory of human-like lane-changing of automatic driving vehicles,the modeling analysis is carried out.The main research contents and results are as follows:1.By analyzing all the lane-changing data,the real lane-changing scenarios are divided into 15 kinds.For each scenario,the factors affecting the driver’s lane-changing decision-making are determined.Based on the decision tree method,the driver’s lane-changing decision-making model in this scenario is established.The model is tested with 400 groups of randomly selected lane-changing data.The results show that the decision accuracy of the model is 63.2% and 65.7% respectively for implementing lane-changing and refusing lane-changing.On this basis,multi-dimensional driving style scale is introduced to classify drivers’ driving styles,and the accuracy of lane change decision-making is tested for three different types of drivers.The results show that the accuracy of lane change and refusal decision-making increases from 63.2% and 65.7% to 83.2% and 84.3% respectively.2.The difference of driver’s style will directly affect the lane-changing decision-making mode,which indicates that in the same traffic scenario,some drivers will choose to perform lane-changing,while some drivers will refuse lane-changing.For the decision-making process of autonomous lane-changing of unmanned vehicles,from the safety point of view,a relatively cautious lane-changing execution strategy can be adopted.At the same time,by using the corresponding lane change decision model under different driving style modes,the change strategy of the lane change decision-making strategy of the unmanned vehicle can be realized,thus catering to the difference in demand of the independent driver or passengers for the independent lane change mode of the unmanned vehicle.3.Through a large number of real lane-changing trajectories fitting research,the results show that the 7-degree polynomial is the best fitting result for lane-changing trajectory,and the fitting accuracy is above 0.99.Based on the 7-degree polynomial trajectory model,a new lane-changing efficiency parameter is proposed to characterize the characteristics of lane-changing process.The average lane-changing efficiency is 7.78 and 6.14 respectively in the free lane-changing stage before and after car-free lane-changing stage and in the lane-changing stage before and after other vehicles.On this basis,aiming at the above two types of lane-changing behavior,this paper establishes a database of lane-changing trajectories at different speeds by fitting a large number of lane-changing trajectories.In the phase of lane change,according to the difference of speed and traffic environment,the most suitable lane change trajectory is selected directly from the lane change trajectory database.As the expected lane change trajectory,combined with the corresponding control methods,it can ensure that the unmanned vehicle performs lane change according to the planned trajectory and realize the humane lane change execution process.