Research On Optimal Trajectory Planning And Tracking Control For Autonomous Parking

In recent years,the rapid development of the economy has driven a sharp increase in vehicle amount,which in turn leads to various problems on energy,environment and traffic.For this case,autonomous driving system and intelligent transportation system can be reliable solutions.Relevant technology in this field has already been a hotpot of research where certain breakthroughs have been made.But unfortunately,due to the limitations of current techniques and laws,autonomous driving technology is not completely applicable in practical scenarios,needing further improvement and development.Autonomous parking technology is one of the components of autonomous driving.As the increasingly complex parking environment leads to increased parking difficulty,demand for autonomous parking technology becomes more and more strong.In addition,it has the feature of the more fixed scenes and low driving speed,indicating a higher possibility in practical applications for related research results.Therefore,research on autonomous parking technology is of great research value and practical significance.This paper conducts in-depth research on trajectory planning of autonomous parking and corresponding tracking control,where the local environment parking and autonomous valet parking are taken as the research objects.The specific research contents are as follows:(1)The trajectory planning of single vehicle in local parking is described in the form of an optimal control problem.A vehicle kinematic model is established as well as constraints of limitation on acceleration and steering wheel speed and kinematic characteristics of the vehicle.The environmental constraints and vehicle collision avoidance constraints in the parking process are established according to the parking scenario,and a single or multi-objective cost function is constructed.For the solving of this optimal control problem,state variables and control variables are discretized at collocation points by pseudospectral method,transforming it into a nonlinear programming problem,which is solved in the next step to obtain the optimal parking trajectory and optimal control sequence as the final result.(2)To improve the accuracy and convergence of the solving of optimal control problem,hp adaptive pseudospectral method is utilized when the accuracy requirement is not satisfied.In this method,the number of time domains or collocation points is dynamically adjusted according to the computation error.To improve the calculation speed of algorithm,the parking scenario is separated based on human parking experience,then a trajectory planning strategy for parallel parking with segmented pseudospectral method is proposed for trajectory planning in each separated part.Aiming at the difficulty of parking trajectory planning caused by narrow parking space width in vertical parking scenario,a trajectory planning method combined with the homotopy method and pseudospectral method is proposed.With the homotopy method,an optimization problem with simple constraints is constructed at first and gradually approaches the original problem to obtain the optimal solution in the process of iteration.In the process of approaching simpler problem to the original problem,the solution of simpler problem serves as the initial guess of next step,improving the convergence of solving.To meet the real-time requirement in practical application,offline optimal trajectories are collected to build the initial value data set for online solving,improving the convergence and solving speed of the algorithm.With simulation,it is shown that the time cost of this method is about 1 second,satisfying the real-time requirement of autonomous parking trajectory planning? These combined into a parking trajectory replanning with free sampling frequency as a whole,which achieves the online trajectory planning with optimal initial guess.(3)Detailed work is carried out for the global trajectory planning of the vehicle in autonomous valet parking.A proposed dynamic heuristic function is introduced to the traditional A* algorithm,adjusting the weight in the heuristic function according to the parking environment.With this flexible weight,the number of extended nodes in the search process can be reduced to help speed up the computation and successfully find the feasible path.Optimal control problems are constructed between waypoints and discretized by the pseudospectral method.Considering solving failure caused by excessive curvature of the path between waypoints which violates kinematic constraints,a trajectory planning strategy with segmentation,rewind and iteration is proposed to improve the success rate of iteration.In addition,for the confliction between trajectories of multiple vehicles,a computation framework based on heuristic initial value is proposed,in the confliction time domain,centralized motion control with the form of optimal control is taken for multiple vehicles to avoid confliction.In this optimal control problem,the control area is minimized to reduce the constraint scale after discretion,aiming at lower computational complexity.(4)According to the optimal trajectory obtained by solving optimal control problem as the reference trajectory,the model predictive control method is utilized for trajectory tracking,where the linear time-variant kinematic model of vehicle is established to predict the future state and the cost function in prediction time domain is the minimization of state error and control increment.Then the local optimal problem in rolling time domain is solved to obtain optimal speed and steering wheel angle for the tracking control of the vehicle.Considering the large curvature of parking trajectory and the limited length of the reference trajectory,a dynamically adjusted prediction time domain strategy based on vehicle speed is proposed for the problem of tracking failure caused by the mismatch of end trajectory parameter.(5)A real vehicle test platform for autonomous parking is established and the parking the controller is developed based on an embedded platform to verify the solving of the proposed algorithm for optimal parking trajectory.The driving simulation experiment platform is built based on Car Sim and A&D5436 and the virtual simulation scenario is constructed,through the hardware-in-the-loop test,the experimental results verify the effectiveness of the proposed multi-vehicle cooperative trajectory planning and tracking control method.