Research on Low-Speed Unmanned Vehicle Obstacle Avoidance Based on FeiTengPai

Abstract

This paper focuses on low-speed unmanned ground vehicles equipped with the Leishen C16 lidar and the FT-Pi embedded platform. Centering on the problems of autonomous navigation and dynamic obstacle avoidance in unknown environments, it systematically expounds the overall scheme design, engineering implementation and real-vehicle test verification of the system. The system takes a 24-dimensional state vector as input — 20 dimensions are lidar ranging data processed by sector partitioning, and the remaining 4 dimensions are the robot's own state information — and outputs continuous speed control commands accordingly. To coordinate the inherent contradictions among approaching the target, avoiding collision and motion smoothness, a new reward function is designed. At the execution level, the system communicates with the underlying motor drive module through a UART serial port with a baud rate of 115200, and a custom service interface is built to receive speed commands. Real-vehicle test results show that the system runs stably, with a maximum linear velocity of 80 mm/s and a maximum angular velocity of 1.57 rad/s. The above results fully demonstrate that the scheme integrating deep reinforcement learning and ROS2 can effectively support robust low-speed autonomous navigation tasks on embedded platforms with limited computing resources.