In this paper, we propose to use an alternative to the classical Kalman filter (KF) to slave the movement of a robot manipulator to the measurements delivered by a far-infrared reflectometry sensor. This alternative, called Guess filter (GF), uses fuzzy rought set theory and possibilistic inference. It is particularly suited to measurements that are both imprecise and inaccurate. We compare the ability of GF and KF approaches to be used to control a robot on a simulated experience that favors neither approach.