Research On Multi-level And Multi-objective Decision-making And Planning For Self-driving Car

The core technology of self-driving car includes environmental perception,decision planning and control execution,in which decision-making and planning complement each other and play a key role in connecting the preceding and following.However,the research on global planning,local planning and intelligent decision-making is relatively independent,and it does not comprehensively consider and analyze the driver preference requirements from macro and micro levels,the shortest time and the lowest fuel consumption,as well as the safety and comfort of the vehicle,and there is still a lack of overall and logical research technology.Therefore,it is necessary to further comb out the internal relations between the three aspects of global planning,local planning and intelligent decision making,and establish a multi-level and multi-objective decision-making technology for automatic driving vehicle planning.This paper takes macro and micro decision planning as the main line,and conducts the following research:(1)In view of the problem that the traditional path planning idea cannot plan the global optimal path in the dynamic road network,the global optimal path planning idea based on dynamic travel time was proposed.The dynamic travel time database of Chongqing University City was established by using traffic simulation software VISSIM.The advantages and disadvantages of Dijkstra and ant colony algorithm in path planning were compared.After analyzing the shortcomings of static path planning and rolling path planning,a global optimal path planning method based on dynamic travel time was proposed.The simulation results show that the proposed idea is effective.(2)I In view of the rare research on global path planning that considers the lowest fuel consumption at present,an economic global optimal path planning method based on dynamic fuel consumption was proposed.First,the test data is preprocessed to obtain the experimental group and the control group data.The former was used for cluster analysis,and the latter is used for classification analysis;the cluster analysis results are used for BP neural network training,and the type recognition of considering working conditions is obtained.Based on the neural network fuel consumption estimation model,and the classification result was used as the test group,compared with the fuel consumption estimation model and steady-state fuel consumption estimation model that does not consider the recognition of the working condition type,the results show the estimation accuracy of the fuel consumption estimation model considering the recognition of the working condition Highest;Finally,a dynamic fuel consumption database was established based on the characteristics of fuel consumption cycle similarity,and the economic global optimal path is planned,which meets the driver’s preference demand for economic travel.(3)Aiming at the problems of less safety lane changing model and lack of reasonable screening of the lane changing path,a decision-making planning model based on safe lane changing region is established.Two typical changing scenes were analyzed,and the critical safe lane change angle model is established respectively.The physical properties of curvature and constraint of starting and ending points of different path changing curves such as polynomial curve and B-spline curve were compared.B-spline curve method is selected as local path planning method,and then,the optimal road change diameter gauge based on safe lane changing area is proposed based on the concept of safe lane change angle On this basis,the decision framework of lane change based on safe lane changing domain is established.Finally,the scene simulation of automatic driving was realized by Simulink and Pre Scan.(4)In view of the fact that the future trajectory of the surrounding vehicles has less influence on the decision-making and planning of the vehicle lane change,a new lane change decision and planning model was established considering the prediction of the vehicle trajectory.The NGSIM data set was processed,and the data set of lane change track with left turn,straight line and right turn was obtained;the LSTM neural network was trained by this data set to obtain the vehicle driving intention prediction model,and then a trajectory prediction model considering driving intention recognition was established;the scene of the changing track was set to expand the predicted track of the obstacle vehicle,and the trajectory prediction distribution was reflected The ST chart was shot as the constraint condition,and the optimal speed sequence was searched in ST chart by using dynamic programming idea.Finally,the pure tracking algorithm was used as the trajectory tracking control algorithm to verify the path and speed of the plan.