Virtual planning and testing of AUV paths for underwater photogrammetry

Abstract

We introduce a system for automatically generating paths for autonomous underwater vehicles which optimize views of a site of interest. These paths can then be used to survey and map underwater sites of interest using photogrammetry. Paths are generated in a virtual world by a single-query probabilistic roadmap algorithm that quickly covers the configuration space and generates small maps with good coverage. The objective function used to compute the paths measures an approximate view coverage by casting rays from the virtual view to test for intersections with the region of interest, with added weight for views with high information gain. The motion planning algorithm was implemented in a virtual world that includes the ability to test paths and acquire views of the virtual scene for evaluation prior to real world deployment. To measure the effectiveness of our paths versus the commonly used pre-packaged lawnmower paths, photogrammetry reconstructions were compared using CloudCompare. The 3D reconstructions created from the views along the paths generated by our algorithm were more detailed and showed better coverage, creating point clouds with a mean distance between points ranging from 1.5 to 2.3 times better than that of the lawnmower pattern.