An almost globally convergent observer for visual SLAM without persistent excitation.
Bowen Yi,Chi Jin,Lei Wang,Guodong Shi,Ian R. Manchester
DOI: https://doi.org/10.1109/CDC45484.2021.9683624
2021-01-01
Abstract: In this paper we propose a novel observer to solve the problem of visual simultaneous localization and mapping, using the information of only the bearing vectors of landmarks observed from a single monocular camera and body-fixed velocities. The system state evolves on the manifold $SE(3)\times \mathbb{R}^{3n}$, on which we design dynamic extensions carefully in order to generate an invariant foliation, such that the problem is reformulated into online parameter identification. Then, following the recently introduced parameter estimation-based observer, we provide a novel and simple solution to address the problem. A notable merit is that the proposed observer guarantees almost global asymptotic stability requiring neither persistent excitation nor uniform complete observability, which, however, are widely adopted in the existing works.