This paper proposes a method for the on-site localization of autonomous vehicles in large-scale outdoor environments, where a single sensor cannot achieve the high precision for localization. The method is based on an improved Kalman filter by fusion of odometry and LiDAR, and it is intended to address the challenge of localization in large-scale environments. Given the complex nature of such environments and the difficulty of identifying natural features at the worksite accurately, the paper uses artificial landmarks to model the working environment. The Iterative Closest Point (ICP) algorithm matches local features of landmarks that were scanned by LiDAR at the current time with local landmark features from the past to obtain the vehicle’s on-site pose. Within the extended Kalman filter (EKF) framework, odometry information is fused with the pose information obtained by the ICP algorithm to further enhance the accuracy of the vehicle’s localization. Simulation results demonstrate that the localization accuracy of unmanned vehicles optimized by the EKF algorithm improves by 9.21% and 53.91% compared to the ICP algorithm and odometry, respectively. This reduces the noise error of measurements, which improves the precise movement and on-site localization performance of unmanned vehicles in large-scale outdoor environments.
Hanxiao Zhou (Fri,) studied this question.