Three-Wheeled Omnidirectional Robot Localization in RFID-Tag Environments using UFIR Filtering

cation (RFID) tag-based systems have attracted the interest of many consumers due to low cost and low (or zero) energy consumption and a wide distance range that made them standard for indoor object navigation and tracking [1]–[8]. In practical designs of mobile robot navigation systems, one finds various efficient hybrid solutions, such as the localization scheme combining information available from the RFID tag-based networks and other sensors. In [9], a novel localization method is proposed to combine the RFID tag-based data with laserbased measurements. In [10], a variable power RFID model is proposed for the localization over passive ultra-high frequency (UHF) RFID tag nets in complex environments. In [11], a location system is designed to combines two types of the RFID tag-generated signals with a logical classification strategy and the integration is provided using the Bayesian filter-based algorithms (BFA). The objective of the BFA is to compute the posterior distributions of the states of a dynamic system, given an observation function with noise. This method has many advantages, but the more remarkable is the capability to represent a complex distribution without requiring information about the state-space model or the state distributions, although with a high computational cost. The state estimation problem [12], [13] can be solved for linear Gaussian processes and observations using the Kalman Filter (KF) and for non-linear models using the Extended KF (EKF) or unscented KF (UKF). Another approach is the unbiased finite impulse response (UFIR) filter [14], which can also be applied to linear models and nonlinear models [15] as described in [16]. An advantage is that the UFIR algorithm does not require the noise statistics and has a better robustness. Navigation over the RFID tag nets is typically provided in the presence of the colored measurement noise (CMN) [17]–[19]. To estimate the robot state under CMN, there can be used two well-known approaches developed by Bryson et al. in [20], [21] and Petovello et al. in [22]. In the Bryson algorithm, the CMN is filtered out in two phases: smoothing and filtering. In the Petovello algorithm, only one stage (filtering) is needed. Another solutions were found by Shmaliy et al. to deal with the colored process noise using state differencing [23], Zhou et al. by using the second moment of information [24], and Ding et al. by applying the least squaresbased iterative parameter estimation to dynamical systems with the autoregressive moving average (ARMA) noise model [25]. In this paper, we apply the KF and UFIR filter modified in [26] under CMN to provide an accurate robot navigation over RFID tag networks.