Study On Interactive-aware Prediction Of Driving Maneuvers And Trajectory Under Ice And Snow Conditions

Accurate prediction of the behavior and trajectory of surrounding vehicles is a necessary prerequisite for intelligent systems such as autonomous vehicles to make a safe and high-quality decision making and motion planning.Due to the hostile road conditions in winter,traffic congestion and safety problems are increasingly serious in the cold areas of north China.The road conditions are complicated in the snow and ice covered environment,and the probability of random changes in traffic vehicle behavior increases.Therefore,it is of great practical and theoretical significance to study the trajectory prediction of surrounding vehicles behavior under the snow and ice covered roads.For this reason,this paper studies how to make interactive prediction of driving maneuvers and trajectory,realize prediction of driving maneuvers and trajectory under ice and snow conditions and collision risk assessment of intelligent vehicles.The specific research content includes the following aspects:（1）An interactive multi-modal behavior trajectory algorithm based on attention mechanism is proposed.Aiming at the bidirectional interaction between vehicles in the prediction scene,GRU was used to encode the historical information of vehicles,and then the interaction information was extracted by the cavity convolution pooling operation according to the relative positions of vehicles.Then,the historical spacialsocial vectors at different times are coded.Finally,hierarchical prediction is used to output the behavioral mobility of the predicted vehicle first,and then the multi-modal prediction distribution of the predicted vehicle’s future trajectory is decoded by combining the attention mechanism.Using NGSIM data set of public freeway scenarios,the results show that the model can accurately predict vehicle behavior and trajectory distribution.（2）The prediction method of vehicle behavior trajectory in terms of snow and ice roads is established.Aiming at the characteristics of low visibility and cautious driving in snow and ice covered environment,lane change repulsion factor was introduced and embedded into the vehicle kinematics model based on interactive prediction.This method can predict the complete motion state of the circumferential vehicle,so as to achieve more accurate position prediction and ensure the feasibility of the motion trajectory.Driving samples with snow and ice covered roads in the CADC natural driving dataset were used to verify the effectiveness of the algorithm in real driving scenarios.（3）A vehicle collision risk assessment method based on interactive trajectory prediction is designed.On the basis of vehicle trajectory prediction distribution probability,the collision probability of target vehicle in the driving environment is calculated by a geometric collision model.The remaining time of critical collision probability is introduced to solve the problem that the prediction time of risk assessment algorithm based on kinematic model is short and the potential danger cannot be predicted.Finally,a typical dangerous driving condition was built to verify the collision risk assessment method,and the results show that the method can effectively predict vehicle driving risk.