LiDAR-based vehicle detection and tracking for autonomous racing

Operating at the physical limits of vehicle dynamics requires perception systems with near-zero latency and high spatial reliability. High-speed autonomous racing represents an extreme edge case for autonomous driving, where multi-agent interactions at velocities exceeding 250 km/h leave no margin for processing delays. Because this emerging domain lacks the large-scale datasets required for modern data-driven approaches, standard urban autonomous driving models are insufficient. This paper details the LiDAR perception architecture utilized by Team PoliMOVE to secure multiple Indy Autonomous Challenge championships. Rather than proposing isolated novel heuristics, our contribution lies in the system-level integration of a domain-specific pipeline: a high-speed Point Cloud Segmentation technique, a specialized Vehicle Pose Estimation framework, and a variable-step Multi-Target Tracking algorithm. Experimental validation confirms that this architecture provides the computational efficiency and robustness necessary for safe, fully autonomous overtaking under extreme-speed constraints.