Efficient Gaussian Process Grid Map Approximation for Mobile Robot Exploration Using 2D LiDAR

An efficient local kernel-based Gaussian Process (GP) grid map approximation method is proposed for mobile robot exploration using 2D Light Detection and Ranging (LiDAR) sensors. While the conventional GP grid map captures structural correlations, its prediction performance is affected by the training data from LiDAR measurements, and the computational burden increases with the number of training and test points. To reduce computational complexity and improve prediction consistency, we construct a local kernel using trained hyperparameters and perform convolution operations with a modified local grid map derived from LiDAR measurements. Experiments using LiDAR measurements obtained from a robot navigating an indoor environment verified that the method efficiently computes the GP grid map and frontier probability, making it suitable for practical frontier-based exploration.