A fast LiDAR registration algorithm for autonomous racing: track constrained Generalized Iterative Closest Point (tc-GICP)

This paper presents a novel point cloud registration approach tailored to meet the requirements of autonomous racing, where rapid and precise LiDAR-based localization is essential in high-speed environments. The proposed method, track constrained GICP (tc-GICP), is a variant of the well-established Generalized Iterative Closest Point (GICP) algorithm, where we increase the computational efficiency by lowering the degrees of freedom of the optimization problem from six to three. This modification, that accelerates convergence significantly without affecting localization accuracy, is accomplished by restricting vehicle pose estimations to a predefined track plane. The performance of tc-GICP is evaluated using real-world data from competitions attended by the PoliMOVE racing team, including the Goodwood Festival of Speed and the 2024 Indianapolis Autonomous Challenge. The results demonstrate the capability of our algorithm to preserve high localization accuracy while achieving a notable speed improvement.