Research On Key Algorithms Of Autonomous Parking In Dynamic And Complex Scenarios

Intelligence is an inevitable trend of automobile technology and industry development,and its development pace is also accelerating.With the increasing urban congestion,it is difficult to find parking spaces and parking becomes extremely inconvenient.Therefore,autonomous parking is widely welcomed by the market and is also one of the research hotspots in the field of intelligent vehicles.Although there are many researches on autonomous parking,most of them are based on relatively simple scenarios.However,with the gradual expansion of the application scenario of autonomous parking,many scenarios present dynamic and complex characteristics,which makes the traditional path planning algorithm difficult to adapt.Therefore,the study of autonomous parking algorithm in dynamic and complex scenarios becomes the key to improve the adaptability of autonomous parking scenarios and promote the wide application of automatic parking。In this paper,the key algorithms of autonomous parking in dynamic and complex scenarios are studied,which is verified by simulation and real car experiment.The specific research content includes the following five parts:(1)This paper proposes a dynamic obstacle tracking and prediction algorithm based on extended Kalman filter,a dynamic obstacle location algorithm based on NDT normal distribution probability density point cloud registration,dynamic obstacle association method based on dynamic obstacle size consistency and motion constraint,and dynamic obstacle motion prediction algorithm based on extended Kalman filter to reduce dynamic obstacle tracking The tracking error caused by lag and its influence on trajectory planning;(2)With the dynamic obstacle movement prediction completed,this paper proposes a predictive description method of dynamic scenario.Based on approximate grid decomposition,hierarchical cost map updating method is adopted and the prediction cost layer of dynamic obstacles is integrated to complete the predictive description of dynamic scenario and provide dynamic input for the path planning algorithm.(3)In order to achieve the replanning in dynamic environment and improve the planning algorithm of adaptive dynamic scenario at the same time satisfy the real-time performance,this paper puts forward the Hybrid dynamic A * mixed with Reeds-Shepp curve planning algorithm,adopts serial search method.Hybrid A* adapts to the dynamic costmap changes;Reeds-Shepp curve with simple calculation,satisfing the vehicle constraints,can be used to accelerate the search in A dynamic environment.A algorithm based on Euclide distance allocating the search weight of Hybrid A* and Reeds-Sheep curves is proposed to improve the overall search efficiency.Simulations are designed to verify the search efficiency of the algorithms.(4)In order to verify the practical effect of the proposed key algorithm in dynamic environment,the pure tracking algorithm is used for lateral control and the course Angle deviation is used for feedback control.Longitudinal velocity tracking is accomplished by using discrete PID considering dynamic obstacles in the longitudinal direction.(5)The above key algorithms are verified by real car test.Firstly,the dynamic complex scenarios is analyzed and typical scenarios are selected.Secondly,the components of autonomous parking experiment platform are introduced.Finally,based on the platform,the experiment is designed to verify the real-time effect of trajectory planning and trajectory following algorithm in the dynamic complex scenarios.The results show that the Hybrid A* and Reeds-Shepp curve mixed planning algorithm proposed in this paper has a good replanning performance.