Design And Implementation Of Park-oriented Autonomous Vehicle Computing System

Automated driving technology has important application potential in park patrols,sanitation,etc.However,the park has many traffic elements,complex environments,unstable lighting and unstable positioning signals,which have brought challenges to the autonomous driving perception decision system.The in-depth application of artificial intelligence algorithms in the field of autonomous driving greatly enhances the ability of autonomous vehicles to perceive complex scenes,but at the same time puts forward higher requirements on the computing power of the on-board computing platform.Park vehicles are generally small in size,limited by the capacity and power of the power supply system,and difficult to meet the real-time operating conditions of complex algorithms,which limits the application of artificial intelligence algorithms in small park autonomous vehicles.To this end,this article takes the automatic driving scenario of some roads in the airport as an example.First,for the driving environment in the park,a low-power heterogeneous computing platform is designed.Secondly,according to the characteristics of the computing unit and algorithm and based on artificial intelligence algorithms,lane line detection,obstacle perception recognition,path tracking and obstacle avoidance functions are realized,and automatic driving at low speed(5 ~ 10km/ h)under park conditions is supported.The main work of the thesis is as follows:(1)According to the park environment and the characteristics of the computing unit,complete the overall design of the on-board computing system for the park.Mainly include sensor selection,low-power heterogeneous computing platform design,software algorithm architecture and communication design.Sensors mainly include GNSS(Global Navigation Satellite System),16-line lidar and cameras;heterogeneous computing platforms mainly include CPU,GPU,FPGA and MCU(Microcontroller Unit);software algorithm framework based on ROS(Robot Operation System)is designed on the software,Support real-time operation of each subsystem and communication between systems.(2)Based on the aforementioned overall system architecture,at the implementation level of the vehicle perception subsystem of the park,obstacle recognition and lane line recognition subsystems were deployed based on the calculation unit and algorithm characteristics,respectively.Among them,based on FPGA using yolov3(You only lookonce)target recognition algorithm to achieve obstacle detection.Based on Xavier,combined image filtering and Kirchhoff transform are adopted to realize lane line detection.On this basis,the fusion of lidar and visual perception is achieved to obtain the spatial positioning information of obstacles and lane lines.(3)Based on Xavier and MCU,a decision control subsystem is implemented,which implements trajectory tracking,obstacle avoidance planning,and bottom-level motion control functions to ensure the real-time and safety of trolley driving.The behavior planning of unmanned vehicles is realized through finite state machine,and the trajectory tracking of unmanned vehicles is carried out through pure tracking algorithm.According to the detection results,a multi-track lane-change obstacle avoidance scheme is designed,and the car is controlled by the STM32 bottom layer controller.The control uses motor soft start,smooth steering and other algorithms.(4)Use the TRAXXAS large X chassis to build a lightweight autonomous driving car,and verify the aforementioned research content in real vehicles.The power consumption of the heterogeneous computing platform designed in this paper is about120 W,which realizes the functions of fixed trajectory path tracking,lane line detection,obstacle detection and obstacle avoidance.In the park scene,the trajectory tracking of the fixed route and the automatic driving of obstacle avoidance were tested.The final experimental results reached expectations,indicating that the automatic driving system designed in this paper can meet the automatic driving in the low-speed environment of the park and achieve low power consumption Claim.