Examination of sampling-based path planning for indoor UAV using voxel grid-based Visual SLAM

Abstract

Simultaneous Localization and Mapping (SLAM) concept solve the problem of unknown map and unknown pose when running a robot in an indoor environment. Amongst SLAM approaches, Visual SLAM can construct 3D map with a camera as the only perception input, as opposed to 2D map of conventional SLAM methods using specialized sensors. That said, applying it to an Unmanned Aerial Vehicles (UAV) in hazardous or dangerous enclosed space reduce certain risk to human. The solution for UAV indoor exploration with an on-board camera consists of using ORB-SLAM2 localisation method, voxel grid- based Voxblox 3D mapping with ESDF layer and path planning algorithm. We examine different planning algorithms under the same VSLAM and resulted in different targeted responses that indicated Optimal RRT (RRT*) as the most efficient sampling-based algorithm for indoor UAV applications. Further results yielded from statistical analysis using a three-way repeated-measure analysis of variance regarding planning quality, computing time, and trajectory feasibility.