Simultaneous RSS-based Localization and Model Calibration in Wireless Networks With a Mobile Robot

This paper presents a recursive expectation maximization-like algorithm that can be used to simultaneously locate the nodes of a wireless network and calibrate the parameters of received signal strength vs. distance models. The algorithm fine tunes one model for each node accounting for its local environment and small hardware differences with respect other nodes. In contrast with using a common model for all the nodes, it is not required to artificially inflate the standard deviation of the random variable accounting for uncertainties in order to accommodate differences of signal strength measurements from different nodes. As a consequence, the position estimate is more accurate. We conducted a series of experiments in which a mobile robot with known location was used as a mobile beacon in three environments with different propagation characteristics. The results show a significant decrease of the mean error of the position estimates in all environments when using individual models compared to using a common one. Using a model with a third order polynomial and a mixture of two Gaussians, the algorithm was able to locate the nodes within a meter on average in an office and with less than half a meter in more open environments. The estimated potential accuracy is about half a meter in all the environments.