Model-Prediction-Based Speed Decision Making And Motion Planning Methods For Autonomous Land Vehicles

In recent years,intelligent driving technology of vehicles has been extensively studied by relevant scholars,organizations and institutions at home and abroad.In the ongoing decades,the development of autonomous vehicles will definately change existing driving habits and travel modes of human beings,and the traffic safety will also be continuously improved due to the development of intelligent driving technology.Model prediction is the modeling and predicting process of complex decision-making and planning problems,including data-driven and mechanism's model prediction.Human beings expect that the vehicle revolutionizes into an autonomous system in the future,to achieve autonomous decision making and planning in various scenes,conditions and tasks.The model-predition-based speed decision making and motion planning have been studied for autonomous land vehicles(ALVs)in this paper.The main achievements and innovations of this paper are as follows:(1)Aiming at the description of vehicle kinematics and dynamics with respect to autonomous vehicle motion planning,the kinematics model,tire model and vehicle dynamics model of the vehicle are analyzed and established.At present,the motion planning based on model-free does not consider the model information of the vehicle or that based on the vehicle model does not satisfy the constraints and parameter-setting.In terms of the above problems,the motion mechanism of the vehicle is analyzed and modeled,and the dynamics model that can be applied to the laboratory vehicle platform and retain the vehicle dynamics characteristics well is proposed.(2)Aiming at the speed decision-making behavior of the skilled driver,a speed decision model based on kernel-based extreme learning machine has been proposed.At present,the speed decision-making of autonomous driving vehicles still faces the problem of complex environment adaptability.To learn modeling and prediction of the human driver's speed decision mechanism undoubtedly improve the autonomous vehicle's ability to cope with complex road scenes while ensuring the rationability of the speed decision making and the comfort for passengers.The model is built by the supervised learning method and samples are from a typical dataset of speed decisions taken by an experienced driver.The results of modeling are compared with those of the machine learning algorithms of support vector machine and extreme learning machine.The simulation experiment results show that the speed decision modeling method based on kernel-based extreme learning machine has the advantages of short training time and high precision,and verify the effectiveness of the proposed method.(3)In order to test the practicability of the proposed speed decision model,a corresponding speed decision-making software module is designed on the experimental platform of autonomous vehicles.The module covers three steps,including the data collection and processing,the use of the speed decision model and the optimization of output results.The online vehicle experiment on an autonomous vehicle experimental platform shows that the obtained speed decision making model is feasible;furthermore,the validity of the proposed autonomous vehicle speed modeling and prediction method based on kernel-base extreme learning machine is verified.(4)Aiming at the problem of ALVs' motion planning in typical and complex scenes,a heuristic variable weight motion planning method based on model predictive control is proposed.At present,the motion planning method based on model predictive control relies on empirical adjustment in the weight-setting.This paper focuses on the problem that bending radius is pretty large under the curved typical scene of autonomous driving,and proposes an improved motion planning method based on model predictive control.The motion planning method is improved based on the heuristic information of the quadratic weight vector of the state variables in the objective function of the model prediction control.As the rolling optimization is performed ceaselessly,the weight vector can ensure that the control sequence converges faster in the U-bend scenario after the quadratic programming solution and reduces the concussion.The simulation results verify the effectiveness of the proposed method in the U-turn scene.