Inspection of an underwater structure using point‐cloud SLAM with an AUV and a laser scanner

Abstract

This paper presents experimental results using a newly developed 3D underwater laser scanner mounted on an autonomous underwater vehicle (AUV) for real‐time simultaneous localization and mapping (SLAM). The algorithm consists of registering point clouds using a dual step procedure. First, a feature‐based coarse alignment is performed, which is then refined using iterative closest point. The robot position is estimated using an extended Kalman filter (EKF) that fuses the data coming from navigation sensors of the AUV. Moreover, the pose from where each point cloud was collected is also stored in the pose‐based EKF‐SLAM state vector. The results of the registration algorithm are used as constraint observations among the different poses within the state vector, solving the full‐SLAM problem. The method is demonstrated using the Girona 500 AUV, equipped with a laser scanner and inspecting a 3D sub‐sea infrastructure inside a water tank. Our results prove that it is possible to limit the navigation drift and deliver a consistent high‐accuracy 3D map of the inspected object.

Underwater Object Recognition Using Point-Features, Bayesian Estimation and Semantic Information

“…As shown in the encircled area of the left Figure 4 , two pipes where detected, one appearing in red (the long one) and the other in blue (small section of a pipe). This happens due to small deformations of the scan caused by the motion induced distortion present in the underwater laser scanner [ 41 ]. Therefore, it is necessary to identify and fuse the point clouds that correspond to the same pipe segment (Algorithm 1) in order to provide a set of non duplicated pipes as input to the next module.…”

State of the Art of Underwater Active Optical 3D Scanners

“…Apart from all these, a key ability of an autonomous robotic system is sensing its environment. For UUVs it is especially important to acquire 3D data of its surroundings in order to perform tasks such as object recognition [16], inspection [17], manipulation [18] or navigation [19].…”

Active SLAM for Autonomous Underwater Exploration

“…On the one hand, if the scene does not have enough features, the ICP algorithm is not capable of registering the point clouds, or, in the worst case, registers them poorly (e.g., when registering two flat surfaces). To solve this problem we propose extracting the 3D features of the point clouds first, to make a first coarse alignment, and only if this succeeds, completing the registration with the ICP algorithm [36]. A second problem has been observed when the vehicle has to travel long distances close to obstacles in order to move from one viewpoint to another.…”