Research On Autonomous Driving Risk Avoidance Algorithm Considering Environmental Vehicle Driving Uncertainty

Investigations by relevant agencies show that most traffic accidents are caused by drivers taking improper driving operations.Therefore,replacing manual operation with autonomous driving technology can reduce the occurrence of traffic accidents.Autonomous vehicle technology has made great progress in recent years,but due to cost and other reasons,it will take a long time to achieve the popularization of autonomous vehicles,and autonomous vehicles will drive together with ordinary cars on the road for a long time.Scenarios,high-speed road vehicle driving faster and intention has the very big uncertainty,in order to ensure the safety of self-driving cars to efficiently driven in high speed road scene,self-driving cars must give full consideration to the environment when making decisions vehicle intent uncertainty,predict its future trajectory,a period of time may evade the risks in advance.In this paper,the driving risk avoidance algorithm of autonomous vehicles considering the uncertainty of environmental vehicle trajectory is studied in high-speed road scenarios.The work mainly focuses on the following aspects: target vehicle trajectory prediction considering target vehicle driving intention and interaction characteristics with the environment,autonomous vehicle trajectory planning considering environmental vehicle trajectory prediction and autonomous vehicle behavior decision based on multi-attribute index cost evaluation.Firstly,the graph convolutional neural network(GCN)and Long and short term memory network(LSTM)were used to establish the vehicle trajectory prediction model,and the target vehicle trajectory prediction was carried out.The interaction characteristics that affect each other between vehicles were extracted using GCN network,and the interaction weight function was designed according to the strength of the interaction.LSTM network was used to extract the historical characteristics of vehicle driving,and the predicted trajectory was expressed as multi-modal probability distribution by predicting the driving intention of the target vehicle.Vehicle tracks under different driving intents are extracted from NGSIM,a public traffic flow data set,to train the trajectory prediction model.The training process and prediction results of the proposed model are compared with those of the same type of trajectory prediction model.The results show that the proposed model can extract vehicle driving characteristics better and predict vehicle trajectory more accurately.Secondly,the autonomous vehicle driving risk avoidance decision algorithm is designed.Using the numerical optimization method,the drivable path cluster was generated according to the current driving state and road structure of the autonomous vehicle.The s-T graph was generated by projecting the probability distribution of environmental vehicle trajectory onto the path cluster.The speed curves representing different driving behaviors were obtained on the S-T graph by using the dynamic programming method,and the optimal curves were selected.The quadratic programming method was used to smooth the speed curves.The collision risk of the trajectory was evaluated by the intensity of the velocity change on the trajectory,and multi-index cost functions were designed to evaluate the trajectory and select the optimal trajectory based on the velocity change,acceleration change,curvature and trajectory offset,so that the vehicle could drive safely and efficiently with the optimal decision.Finally,the vehicle kinematics model and model Predictive control(MPC)motion controller are established,and the effectiveness of the controller is verified by Simulink and Carsim co-simulation.On this basis,traffic scenarios are designed to test the risk avoidance algorithm designed in this paper.The results show that,The algorithm designed in this paper can control the vehicle safely and efficiently at the decision level,and has good robustness and generality.