MINS: Tightly coupled MultiBeam EchoSounder Inertial Navigation System for 3D bathymetric underwater inspection

摘要

By fusing a MultiBeam EchoSounder (MBES) with an Inertial Measurement Unit (IMU) and a Doppler Velocity Log (DVL) we present MINS: a graph-based, tightly coupled, and featureless MBES-Inertial Navigation System that enables accurate and real-time robot trajectory estimation, map building, and sensor bias estimation. MINS formulates an MBES-Inertial odometry problem for an Autonomous Underwater Vehicle (AUV) using a factor graph. The relative displacement estimated from the joint IMU and DVL preintegration is used to chain keyframes and to build 3D scans from a Mechanical Scanning MBES. This sensor allows the AUV to be immersed within the inspected scene to build a 3D point cloud map — which we call 3D bathymetry — compared to conventional systems that use an AUV navigating above the inspected area to collect a 2.5D bathymetry. The obtained scans are aligned to set sonar odometry or loop closure factors, by applying a probabilistic registration algorithm that uses Gaussian Mixture Models to represent the scans and quantify the alignment uncertainty. A rigorous on-manifold formulation is provided, which properly models the AUV state uncertainty in a compact Lie group. This system is evaluated in a field experiment demonstrating its ability to produce an accurate and consistent 3D bathymetry of a shipwreck. • Rigorous formulation that models the robot state uncertainty in a compact Lie group. • Tracking of the alignment uncertainty, getting it from the geometry of the scans. • Tightly coupled estimator that holds all the existing correlation between all the measurements. • Graph-based estimator that allows to set loop closure constrains. • Solution that can be executed at run time on the robot.