State estimation under quantized measurements: a Sigma-Point Bayesian approach

Sensors providing only quantized or binary measurements are present in several automation contexts. A remarkable example is the Radio Frequency IDentification technology when only the detection of the tags is used as information for robot localization. In this paper we propose an algorithm which merges some concepts of the Unscented Kalman Filter (UKF) with some aspects of the Particle Filter (PF). The prediction step of the proposed method is like the prediction step of a standard UKF. On the contrary, the correction step of the UKF can not be trivially implemented due to the presence of binary measurements. For this reason a different correction step is proposed here where the sigma-points weights are modified according to their agreement with the measurements, like it is done for particles of a PF. The main advantage of the proposed algorithm with respect to a PF is that much less particles are needed. Moreover, the way to generate particles in the proposed approach is not random but deterministic. A simulative comparison of the proposed approach with respect to a PF and with respect to a Quantized Kalman Filter is reported in the paper.