Integrated Perception, Planning, and Control for Autonomous Racing: A Case Study on Formula Student Autonomous China

This paper presents the design, implementation, and comprehensive validation of an autonomous driving modular software system tailored for a Formula Student Autonomous China (FSAC) racing platform. The system is architected around four tightly integrated modules: perception, localization and mapping, trajectory planning, and vehicle control ${ }^{[1]}$. The perception module incorporates an enhanced YOLOv8 network for high-precision detection and spatial localization of multiple traffic cone types. A dense point cloud map is constructed using an RGB-D camera, and IMU data is fused via factor graph optimization to ensure robust and globally consistent localization. For trajectory planning, cone landmarks are geometrically sorted and smoothed using Bezier curve fitting, followed by cubic spline interpolation for continuity and smoothness. The control framework combines Model Predictive Control (MPC) and PID control to meet the rapid dynamic response requirements of highspeed autonomous driving. The system has been rigorously validated through both closed-loop simulation and real-world field tests. Experimental results confirm that the system meets real-time performance and accuracy requirements, successfully achieving end-to-end closed-loop autonomous driving for high-dynamic FSAC scenarios.