Visual-Inertial Navigation with Guaranteed Convergence

This contribution presents a constraints-based loosely-coupled Augmented Implicit Kalman Filter approach to vision-aided inertial navigation that uses epipolar constraints as output map. The proposed approach is capable of estimating the standard navigation output (velocity, position and attitude) together with inertial sensor biases. An observability analysis is proposed in order to define the motion requirements for full observability of the system and asymptotic convergence of the parameter estimations. Simulations are presented to support the theoretical conclusions.