Research On Intelligent Vehicle Path Planning Based On Target Trajectory Prediction

With the development of 5G technology and Internet of Vehicles technology,intelligent vehicles have entered the era of rapid development.The application of intelligent vehicles can reduce the probability of traffic accidents and has a wide range of application value in many fields.Path planning is one of the most important problems in intelligent vehicle technology.At present,path planning is mainly realized by the combination of global path planning and local path planning.Global path planning generates global path points as constraints of local path planning.Local path planning generates collision avoidance trajectories through real-time sensing data obtained by sensors under the constraints of global path planning.The traditional path planning algorithm considers the instantaneous position of dynamic traffic targets when generating the planned path,which has the problems of high planning frequency and untimely planning.In the real scene,the movement of the traffic target has continuity in time and space.Therefore,this paper uses the neural network method to predict the trajectory,obtains the future trajectory of the target,and introduces it into the search space of path planning.To reduce the frequency of re route planning and improve the timeliness of planning,so as to ensure the safety of intelligent vehicles.The following three aspects of research have been carried out(1)A global path planning method of improved A* algorithm based on lidar scene modeling is studied.Collect the 3D point cloud of the scene through the laser detection and ranging(Li DAR)and inertial navigation system(INS)to generate the environmental point cloud map after calibration;After point cloud filtering,down sampling and ground point cloud removal,the grid method is used to generate a grid map that can be used for the global path planning of intelligent vehicles;Based on the constructed grid map,A* algorithm based on expanding search direction and priority strategy is proposed to reduce the number of turning points;Secondly,aiming at the problem that the path planned by A* algorithm is close to the obstacle,an improved A* algorithm heuristic function based on artificial potential field method is proposed.Simulation and real scene experiments show that this method has good security and rationality(2)The dynamic traffic target trajectory prediction based on LSTM Long Short Term Memory neural network is completed.In view of the risk that the traditional local path planning regards dynamic obstacles as instantaneous and static,this thesis establishes a dynamic traffic target trajectory prediction model based on LSTM encoder decoder by analyzing Artificial Neural Networks(ANN)and combining the problem description of trajectory prediction sequence to sequence;The historical input features in a fixed time are encoded into context vectors by encoding LSTM layer,and the prediction sequence is output recursively by decoding LSTM cycle to provide input for local path planning;In NGSIM data set,the root mean square error of the predicted target within 3s is2.258m;In ETH and UCY data sets,the average displacement error and final displacement error of the predicted target in 3.2s are 0.706 m and 1.302 m respectively.(3)A fifth-order polynomial local path planning method based on target trajectory prediction is studied.On the basis of global path planning,combined with dynamic target trajectory prediction,local path planning based on improved fifth-order polynomial local path planning in Frenet coordinate is adopted: firstly,the smooth trajectory of global path points after cubic B-spline fitting is used as the reference trajectory;Then,through the analysis of the fifth-order polynomial local path planning algorithm,the transverse motion planning model and longitudinal motion planning model are established,and the candidate trajectory set of local path planning is generated;The candidate trajectory screening method based on target trajectory prediction proposed in this thesis is used to extract the trajectory set without collision in the same time and space;The optimal trajectory planning method based on global constraints proposed in this thesis is used to generate the optimal trajectory without collision in a certain time in the future.The effectiveness of the proposed algorithm is verified by simulation experiment.