Design and validation of a linear parameter varying localization system

This paper proposes and experimentally validates a landmark-based absolute mobile robot localization system, composed by two filters, one for attitude estimation and the other for position estimation. The estimation is carried out in the body-frame allowing for the model kinematics to be LPV (Linear Parameter Varying), thus using no approximate linearisation. The resultant estimators respect GAS (globally asymptotically stable) error dynamics, are parametrized by odometry data and corrected by landmark position and attitude measurements provided by an on-board RGB-D (red, green, blue and depth) sensor. Experiments were carried out, making use of a wheeled mobile robot and a Qualysis Motion Tracking System for ground-truth validation. Attitude and position as well as linear and angular slippages, both proven observable, are estimated, resulting in an effective real-time localization system without requiring the landmark to be always visible. Error convergence is achieved regardless of the initial estimate of both position and attitude, validating the system global stability.