Title:
Ultra-wideband Localization on Manifolds for Autonomous Metal Structure Inspection
Ultra-wideband Localization on Manifolds for Autonomous Metal Structure Inspection
Author(s)
Starbuck, Bryan Edward
Advisor(s)
Pradalier, Cédric
Declercq, Nico F.
Declercq, Nico F.
Editor(s)
Collections
Supplementary to
Permanent Link
Abstract
A robot that can probabalistically infer its state and uncertainties while exploiting differential geometry is capable of achieving more consistent, more accurate, robust state
estimation. It is being proposed that ultra-wideband, a cutting-edge technology, that is also highly unpredictable, can be used to give autonomy to a magnetic-wheeled crawler robot for the application of metal structure inspection. Thus, ultra-wideband technology is evaluated based on its sensitivity to metal surfaces at varying heights, as well as its response to varying grid sizes between receivers in experiments featuring a Turtlebot and an RTK-GPS. Then, a novel methodology for ultra-wideband grid initialization is presented featuring a simulation of a ship hull with an ultra-wideband grid. Finally, a metal structure is considered as a parallelizable manifold with a bivariate b-spline representation, and the matrix exponential correspondence between a Lie group and its Lie algebra for the Special Orthogonal Group is applied within the Extended Kalman Filter framework. These considerations constitute the Manifold Invariant Extended Kalman Filter (M-IEKF), a novel approach to more robust state estimation. The filter is derived, presented, and evaluated in comparison with a modified standard approach: the Manifold-Constrained Extended Kalman Filter (MC-EKF), which uses zero-noise virtual measurements to constrain the state estimate. Then, for a real proof of concept, an experiment using a magnetic-wheeled crawler robot
with ultra-wideband localization on a surface consisting of curved metal plates is carried out giving viability to the approach in the real-world application of autonomous metal
structure inspection.
Sponsor
Date Issued
2021-04-28
Extent
Resource Type
Text
Resource Subtype
Thesis