This work investigates the relationship between system observability properties and estimator inconsistency for a Vision-aided Inertial Navigation System (VINS). In particular, first we introduce a new methodology for determining the unobservable directions of nonlinear systems by factorizing the observability matrix according to the observable and unobservable modes. Subsequently, we apply this method to the VINS nonlinear model and determine its unobservable directions analytically. We leverage our analysis to improve the accuracy and consistency of linearized estimators applied to VINS. Our key findings are evaluated through extensive simulations and experimental validation on real-world data, demonstrating the superior accuracy and consistency of the proposed VINS framework compared to standard approaches.
Bibliographical noteFunding Information:
This work was supported by the University of Minnesota (DTC) and AFOSR (grant number FA9550-10-1-0567).
- Vision-aided inertial navigation
- estimator consistency
- observability analysis
- visual-inertial odometry