Adhering to terrain characteristics for position estimation of mobile robots
Abstract
Outdoor environments bear the problem of different terrains along with changing driving properties. Therefore, compared to indoor environments, the kinematics of mobile robots is much more complex. In this paper we present a comprehensive approach to learn the function of outdoor kinematics for mobile robots. Future robot positions are estimated by employing Gaussian process regression (GPR) in combination with an Unscented Kalman filter (UKF). Our approach uses optimized terrain models according to the classification of the current terrain - accomplished through Gaussian process classification (GPC) and a second order Bayesian filter (BF). Experiments showed our approach to provide more accurate estimates compared to single terrain model methods, as well as to be competitive to other dynamic approaches.
