Efficient Line-of-Sight Viewpoint Sampling in Complex Environments for Autonomous Surface Inspection

This paper presents a novel approach to gen-erate a set of viewpoints in order to maximize the surface coverage for robotic inspection missions with limited field-of-view onboard sensing and operation time. The problem encounters challenges in validating line-of-sight visibility within obstacle-rich environments, achieving feasible optimization to maximize surface coverage with minimal viewpoints, and ensuring tractable computational complexity for onboard durability. The proposed approach leverages sensor and environment geometric information to ensure surface coverage for inspection by visiting a minimal number of viewpoints. A computational complexity analysis reveals the proposed approach alleviates the computational power required compared with former viewpoint samplings in free space. The proposed viewpoint sampling approach is experimentally validated using a quadrupedal robot performing autonomous inspection missions in diverse urban environments, demonstrating better performance toward total surface area coverage, the computational time required, and the surface coverage rate during operation.