Research On Driver Intention Recognition And Intelligent Lane Change Control System For L3 Autonomous Driving

Automobile intelligence which is one of the inevitable development directions in the future can reduce the burden of drivers and improve the competitiveness of automobile brands.At present,most of the traditional automobile enterprises research on L3 autonomous driving technologies.Before mass production,vehicles require extensive testing to ensure safety and reliability.Compared with real road testing,the virtual scene test can be carried out under various working conditions,including extreme conditions and others.In the initial development phase of control algorithms,the virtual test is safer,and it can save a lot of cost.Therefore,virtual scene test is gradually being recognized among traditional automobile enterprises and automatic pilot startups.In order to make the virtual scene test more valuable,how to create the virtual scenes which is more realistic has become an urgent problem.Decision planning and motion control are the important technology research modules of intelligent vehicles.In order to meet the requirements of comfort and safety,it is particularly important to establish a behavioral intention recognition model for drivers and make appropriate trajectory planning according to the vehicle's state and external environment.In order to track the desired trajectory perfectly,it need to develop better trajectory tracking control algorithms.In this paper,the following aspects are researched:(1)For the common landing scenarios of L3 self-driving,this paper proposed a technical route of creating the virtual scenes that are close to the real road in VTD.The urban road and three-kilometer ring expressway of China international driverless vehicle competition were built.In addition,a technical route of creating virtual road scene in ArcGIS and CarSim is also proposed.Shazhong road in Shapingba district,Chongqing was built.(2)A gaussian mixture hidden markov model is proposed for driver intention recognition,and the selected data from NGSIM database was used for training and testing.Aiming at the problem that the different numbers of gaussian models in the hybrid model had a great impact on the accuracy of driver intention recognition,particle swarm optimization algorithm was used to optimize the number of gaussian models.And the important parameter values in the mixed model with the highest accuracy of driver intention recognition were obtained through training(3)An intelligent lane change control system based on model predictive control was designed.The polynomial was used to generate the alternative sets of the lateral and longitudinal trajectories,and the optimal trajectory was obtained by optimizing the objective function under the constraints.The trajectory tracking control algorithm based on model predictive control was designed,and the expected control variable inputs were obtained through the vehicle inverse dynamics model.In addition,the co-simulation model of CarSim and Simulink of intelligent lane change control system was built,and the stability of control system was verified by the simulation testing under different initial speed conditions.In addition,the influence of the prediction horizons on the controller performance was explored,and the effectiveness of the system was investigated when the road adhesion coefficients were under extreme condition.