TWOR: Improving Modeling and Self-Localization in RFID-Tag Networks Under Colored Noise

Abstract

This paper discusses the three-wheeled omnidirectional robot (TWOR) self-localization in radio frequency identification (RFID) tag environments. The nonlinear TWOR model is significantly improved by using geometric interpretation and incremental time representation in discrete time. The TWOR position and heading are self-estimated using distance measurements to RFID tags and a digital gyroscope in the presence of typical colored measurement noise (CMN). The extended unbiased finite impulse response (EFIR) is developed along with the extended Kalman filter (EKF) and their versions, cEKF and cEFIR, modified for Gauss-Markov CMN. A particle filter is used as a benchmark. It is shown that the cEFIR filter is more robust than the cEKF and almost as robust as the particle filter, although the latter is less accurate in real time.