Parallel FPFH SLAM for Aerial Vehicles

Abstract

Simultaneous Localization and Mapping (SLAM) is a computational problem in autonomous robotics that has been a domain of extensive research in the past decade. The significance of the algorithm in ensuring the mainstream application of autonomous robot navigation explains the importance placed on it. Aerial vehicles are employed in several critical industries like defence and forestry to provide accurate surveillance results with the help of onboard navigation and visioning systems. While SLAM algorithms are existent for UAV applications, the application of parallelism on 3D aerial SLAM has not been explored. This paper proposes a parallelized LiDAR 3D SLAM algorithm for aerial autonomous robots. Fast Point Feature Histogram (FPFH) descriptors are extracted for the alignment process which is later finetuned by Iterative Closest Point (ICP) registration. The final estimated trajectory is put through 3D pose graph optimization to minimize the overall drift that may be present The final trajectory and point cloud map is displayed by the algorithm and compared with the ground truth data. The number of threads and CPUs utilized for parallelization of the algorithm have been varied and compared. The simulated results show a significant change of 26% in execution time for the parallelized algorithm utilizing 4 CPUs compared to its serial version.