Risk Assessment Of Multi-Vehicle Interaction And Motion Planning In Typical Traffic Scenes

With the development of science and technology and the improvement of urban road system construction,the number of autonomous vehicles is increasing in urban and other traffic environments.However,the driving styles of vehicles are varied and vehicle movement is highly random in urban traffic environment,there are still many difficulties for intelligent vehicles to realize autonomous driving in complex environments such as cities.In response to this problem,this project proposes a motion planning method that combines trajectory prediction and scene confidence prediction risk field to plan a more anthropomorphic trajectory for autonomous vehicles,enabling them to achieve safe,efficient,and comfortable interactions with surrounding vehicles.First,analyze the driving behavior characteristics based on the natural driving data of the vehicle.Taking the Ubiquitous Traffic Eyes CKQ4,NGSIM US-101,and highD data sets as the research object,using the differential approximation method and RTS smoothing algorithm for data preprocessing to obtain various microscopic driving parameters of the vehicle;combined with vehicle driving safety,driving efficiency,and driving comfort characteristic factors,use the K-means algorithm for clustering to analyze the driving behavior characteristics of conservative and aggressive vehicles,and assign different weights to each characteristic factor of the subsequent dynamic programming cost function.Secondly,aiming at the problem that it is difficult to accurately predict the trajectory of interactive vehicles in complex traffic environments such as cities,a trajectory prediction model SIA-GAN based on a spatial-temporal attention mechanism is proposed.On the basis of Social GAN,a time attention mechanism is added to mine the time dependence on its own historical trajectory;the two-dimensional normal distribution function is improved by considering the vehicle state parameters such as driving speed,acceleration,heading angle and shape size,which modeled the vehicle interactive influence force field,according to the value of the interactive influence,’focuses’ on the surrounding vehicle information that has a greater influence on the predicted vehicle through the spatial attention mechanism.Comparing the prediction results on three public data sets,the results show that compared with the existing advanced trajectory prediction algorithms,the proposed SIA-GAN model not only increases the convergence speed during training,but also has the best effect on various evaluation indicators.And it can effectively improve the accuracy and interpretability of vehicle trajectory prediction.Then,aiming at the problem that the existing motion planning algorithms seldom consider the future motion of the interactive vehicle,a motion planning method combining the predicted trajectory of the interactive vehicle and the scene confidence prediction risk field is proposed.On the basis of the vehicle interaction force field,consider the trajectory prediction results and prediction errors to establish the vehicle confidence prediction risk field,and combine the road risk field to obtain the scene space-time confidence prediction risk field,so as to quantify the interaction risk between vehicles;calculate the reachable position domain of the vehicle at each planning time point and sample the trajectory points at equal intervals in the horizontal and vertical directions,according to the kinematic constraints and pre-collision detection,the discrete feasible trajectory cluster of the vehicle is obtained,and the multi-index dynamic programming cost function is designed to screen the discrete optimal planning trajectory.Using multi-segment quintic polynomial curve fitting on the x-t,y-t plane to obtain the continuous optimal planning trajectory satisfying the vehicle dynamics constraints.Based on the model predictive control and PID control algorithm,the horizontal and vertical decoupling control tracking of the planned trajectory is realized.Finally,PreScan-CarSim-Matlab&Simulink-Python joint simulation test platform is built to verify the feasibility and effectiveness of the vehicle motion planning method proposed in this topic.Excerpted from the natural driving data set,the real driving trajectory of the vehicle was designed to design three simulation test scenarios.The test results prove that the motion planning method proposed in this topic can enable vehicles with different driving grids to interact with surrounding vehicles in a safe,efficient and comfortable manner.