Research On Intelligent Obstacle Avoidance Decision Of Unmanned In-wheel Driven Vehicle

With the continuous increase in the number of cars in the world,the resulting traffic accidents and energy shortages have become increasingly serious,posing a huge threat to people’s lives and living environment.The rise of unmanned wheel-drive vehicle provides new ideas for solving this problem.Driver improper operation is the main cause of traffic accidents,and unmanned wheel-drive vehicles do not require driver operation,which can effectively avoid traffic accidents.And the transmission efficiency and energy utilization rate are improved.Therefore,it will become the development trend of future automobiles.Obstacle avoidance decision-making technology,as the core technology of unmanned driving,has always been a hot spot for researchers.Unmanned vehicles make decisions based on impending circumstances to avoid obstacles while driving.The resulting decisions directly affect the safety of unmanned vehicles,the efficiency of traffic and the comfort of passengers,and play a vital role in smooth road traffic safety and people’s driving experience.Due to the complex and changeable traffic environment,higher requirements are put forward for obstacle avoidance decision-making technology.Because most of the traffic accidents occur in the scene of vehicle lane changing and passing through the intersection without signal lights,the obstacle avoidance decision in the above two scenarios is studied in this paper and the specific research content includes the following aspects:1.Construction of virtual driving experiment platform.The virtual driving experiment platform was composed of prescan,Car Sim,MATLAB / Simulink and Logitech G29 driving simulator.The traffic scene was established in prescan,the sensors used in the simulation experiment were selected,and the vehicle model driven by front wheel hub motor was established in Car Sim.The vehicle model in prescan was replaced by this model.Relevant models and algorithms can be established in MATLAB / Simulink,and there were interconnected channels between Prescan,Carsim and MATLAB/Simulink.The Logitech G29 driving simulator was connected to the computer through the USB interface.In Prescan,there was an interface specifically connected with Logitech G29,so that the data between Prescan,Carsim,MATLAB/Simulink and Logitech G29 driving simulator can be interacted from time to time,laying the foundation for the following data collection,model training and verification.2.Research on the prediction of vehicle driving intention.Driving intention prediction was divided into lane-changing intention and driving intention prediction through unsignalized intersections.In terms of lane-changing intention prediction,the influence of the three factors of the motion state of the predicted vehicle,the positional relationship between the predicted vehicle and the lane,and the motion state of the vehicles around the predicted vehicle on the prediction of the vehicle lane change intention were considered by the system,and the lane change intention prediction features were screened.Then,the feed-forward neural network was used to establish the vehicle lane changing intention prediction model,and compared and analyzed it with the model of commonly used predictive features at current stage and the Support Vector Machine model.Finally,the versatility of the lane-changing intention prediction of the preceding vehicle and the preceding vehicle in the adjacent lane on the left was analyzed.In terms of driving intention prediction at intersections with no signal lights,firstly,the prediction features of driving intentions when the vehicle passes through the intersection were screened,and then the feedforward neural network was used to establish the vehicle driving intention prediction model.Finally,the virtual driving experiment platform was used to verify the validity of the built model.3.Research on vehicle trajectory and traffic situation prediction.First,on the basis of predicting the intention of the vehicle to change lanes,the trajectory prediction of the vehicle was decoupled into the prediction of the X-axis and Y-axis trajectory in the ground coordinate system.In the X-axis direction,K-Means was used to cluster the X-axis acceleration,and on this basis,the feedforward neural network was used to sequentially establish the X-axis acceleration type recognition model and the X-axis trajectory prediction model.In the Y-axis direction,the lateral trajectories were clustered based on the clustering of the X-axis acceleration,and the Y-axis trajectory type recognition model and the Y-axis trajectory prediction model were established in turn using the feedforward neural network.Then,the predicted trajectory was coupled to predict the complete trajectory of the vehicle.In this way,the trajectory and traffic situation of vehicles changing lanes and passing through intersections without signal lights were predicted.Finally,the virtual driving experiment platform was used to verify the decoupled trajectory prediction and traffic situation prediction by simulation experiments.4.Research on intelligent obstacle avoidance decision of unmanned vehicle.First,based on the depth deterministic policy gradient model,the decision models for automatic lane changing and intersections were established respectively: the predicted traffic situation of surrounding vehicles and the space state of unmanned vehicles were used as input,and longitudinal speed and front wheel angle were used as output.The multi-objective reward function of safety,comfort and timeliness is designed,and the exploration strategy was set according to the actual driving operation,and then the model was trained to obtain the planned trajectory with the largest total benefit.Finally,the virtual driving experiment platform was used to verify the decision-making model of automatic lane changing and obstacle avoidance through intersections without traffic lights.5.Research on trajectory tracking control.The trajectory tracking controller was composed of a speed tracking control module and a lateral trajectory tracking control module.The speed tracking control module adopted a PID controller,and the lateral trajectory tracking control module was composed of a steering wheel control sub-module based on fuzzy control and a yaw motion control sub-module based on PID.The steering wheel control sub-module based on fuzzy control takes the lateral error and the change rate of the error as the input,and the steering wheel angle as the output.Based on the PID yaw motion control sub-module,the torque of the left and right wheel hub motors was distributed according to the yaw moment to achieve a more accurate tracking of the lateral trajectory.Then,the trajectory tracking effects of the ordinary-driven unmanned vehicle,the wheel-driven unmanned vehicle and the reference model were compared and verified.Finally,simulation experiments were performed to verify the tracking control of the planned trajectory when the hub-driven unmanned vehicle automatically changes lanes and passes through an intersection without traffic lights.