Research On 3D Path Planning And Path Tracking Control For Autonomous Vehicle

With the advantages of freeing human drivers and improving the safety of traffic,autonomous vehicle(AV)is conducive to solving problems such as traffic safety and congestion,and it is regarded as the main means of transportation for future travel.In recent years,many research institutes have invested in AV technology,and new theories and technologies have been generated to promote the rapid development of the AV industry.Autonomous driving technology contains four aspects: global path planning,complex environment awareness,local path planning,and motion control.In this paper,we focus on the global path planning and motion control.In the area of path planning,most of the current planning methods focus on a single goal of distance(or travel time,and paths with short distances for mountainous cities may include too much upslope.For both fuel and electric vehicles,upslope energy consumption is much greater than that of flat roads,which does not meet the needs of users.Then,it is necessary to find the paths that are moderately far away to avoid upslope but do not increase the distance significantly.To this end,this paper investigates a heuristic-based and a Deep Q Network(DQN)based three-dimensional(3D)multi-objective path planning method,which realizes planning with good trade-off between distance and energy consumption.In terms of motion control,this paper studies adaptive path tracking methods in special scenarios.Path tracking control is of importance to ensure the safe driving of AV.This paper addresses the shortage of path tracking methods in some complex scenarios,and conducts research in three aspects: adaptive lane-change path tracking method,adaptive high-speed cruise path tracking method,and adaptive path tracking method under strong transient side wind disturbance.The main works of this paper are as follows.(1)A heuristic-based 3D multi-objective path planning method(H3DM)is proposed for path planning with good trade-off between distance and energy consumption.A weight decay model is designed in this method to solve the technical problem that heuristic planning methods are hard to plan paths on 3D maps.According to the multi-objective planning requirement of weighing distance and energy consumption,a path evaluation model based on the integrated distance and energy consumption in 2D and 3D scenes is constructed,solving the technical problem of unreasonable planning.In addition,a multiobjective optimization algorithm is used to optimize the parameters of the proposed method,which improves the global performance of the results.Finally,the H3 DM is tested on a random map without road network and a local map of the mountain city of Chongqing with road network,respectively.(2)For higher planning efficiency than the heuristic-based planning method,a DQNbased 3D multi-objective path planning method(D3DM)is proposed.In this work,a deep convolutional network is used to construct the DQN,a multi-objective reward function is constructed based on the reward shaping theory,and a training process based on imitation learning combined with stochastic exploration is designed to solve the technical problems of difficult and poor convergence of DQN.In addition,a 3D path planning simulation platform based on Gym framework is built.Finally,it is verified that the D3 DM can plan paths with a significantly high efficiency in comparison with other methods.(3)An adaptive lane-change path tracking method(ALPT)is proposed to improve the path tracking accuracy of AV under the emergent scenario.The method consists of two parts: the course angle optimal reference model(CAORM)and the neural networkbased model prediction controller(NNMPC).In the CAORM,the driving experience of human experts is analyzed,and the fuzzy inference system is used to fuse the lateral speed and longitudinal speed of the vehicle to build a new preview following model,which solves the problems of inaccurate estimation of desired coarse and poor adaptation.In the NNMPC,a neural network predictor is designed to achieve high-accuracy prediction of vehicle state and coarse deviation;a PID controller based on the neural network optimizer is built,and the neural network optimizer is used to provide optimization parameters for the PID controller,which eliminates the coarse deviation of AV.Simulations prove that the ALPT can achieve high-accuracy adaptive lane-change maneuver under the emergent scenario.(4)An adaptive high-speed cruise path tracking method(AHCC)is proposed to improve the path tracking accuracy of the AV during high-speed cruise.In order to improve the adaptivity of the algorithm,the design is carried out in two aspects: first,a dynamic multi-point preview model(DMPM)is proposed to improve the adaptivity of the preview follower to the vehicle speed;second,an adaptive fuzzy steering controller(AFSC)is built to enhance the adaptivity of the steering control.In the DMPM,an adaptive multi-point preview model using vehicle speed is proposed,and an error integration feedback model is designed to improve the estimation accuracy of the desired coarse deviation.In the AFSC,the analysis utilizes the human expert driving experience and combines the control requirements of AV,and adopts the desired coarse deviation and the curvature of the path to achieve adaptive inference of the control signal.In addition,a parameter fitting curve is constructed by curve fitting method to solve the parameter optimization problem of the AFSC.(5)An adaptive path tracking method under side wind disturbance(APTW)is proposed for the problem of strong side wind disturbing with the driving of AV.The method is designed based on the idea of neural network predictive control and contains three parts: a neural network predictor,a steering controller based on fuzzy inference and a speed controller based on PID.The neural network predictor is used to predict the state of the AV under side wind disturbance and the coupled interference between steering and speed,and the error integrated model for steering control and speed control is constructed to solve the problem of mixed modeling of control error and interference error;The steering controller and speed controller,based on the error estimation provided by the error integrated model,realize the course and speed control in high accuracy,eliminate the disturbance of side wind,and improve comfort and safety of passengers.