Search papers, labs, and topics across Lattice.
This paper analyzes the observability and consistency of visual-inertial navigation systems (VINS) that leverage anchored feature representations, revealing that the unobservable subspace is independent of the landmark state but dependent on the navigation state. The findings indicate that using anchored feature parameterizations enhances estimator consistency without requiring modifications, although additional techniques are necessary to address navigation state dependencies. Simulation and real-world experiments show that VINS with anchored features outperform traditional global reference frame methods, particularly in challenging initialization scenarios, while achieving comparable performance to enhanced consistency estimators.
Anchored feature parameterizations can significantly boost the consistency of visual-inertial navigation systems, even in poor initialization conditions.
This paper presents an analysis of the observability and consistency properties of filtering-based visual-inertial navigation systems (VINS) that utilize anchored feature representations. The unobservable subspace of VINS with anchored landmark parameterizations is shown to be independent of the estimated landmark state, which leads to improved estimator consistency properties without any additional modifications. However, the unobservable subspace is still found to depend on the estimated navigation state, necessitating additional consistency-enforcing techniques. Two methods to improve the consistency of VINS with anchored feature representations are presented. Simulation results showcase that all estimators employing anchored feature paramterizations exhibit improved consistency properties compared to algorithms that estimate features resolved in a global reference frame, especially in scenarios where feature initialization may be poor. Real-world experiments on the TUM-VI dataset showcase that the use of anchored feature representations alone can yield comparable performance to consistency-improved estimators employing a global feature representation, demonstrating the benefit of using anchored feature parameterizations for VINS.
