A Robust Method of Localization and Mapping Using Only Range

In this paper we present results in mobile robot localization and simultaneous localization and mapping (SLAM) using range from radio. In previous work we have shown how range readings from radio tags placed in the environment can be used to localize a robot and map tag locations using a standard Cartesian extended Kalman filter (EKF) that linearizes the probability distribution due to range measurements based on prior estimates. Our experience with this method was that the filter could perform poorly and even diverge in cases of missing data and poor initialization. Here we present a new formulation that utilizes a polar parameterization to gain robustness without sacrificing accuracy. Specifically, our method is shown to have significantly better performance with poor and even no initialization, infrequent measurements, and incorrect data association. We present results from a mobile robot equipped with high accuracy ground truth, operating over several kilometers.