Automatic training method applied to a WiFi+ultrasound POMDP navigation system

This paper presents an automatic training method based on the Baum-Welch algorithm (also known as EM algorithm) and a robust low-level controller. The method has been applied to the indoor autonomous navigation of a surveillance robot that utilizes a WiFi+Ultrasound Partially Observable Markov Decision Process (POMDP). This method uses a robust local navigation system to automatically provide some WiFi+Ultrasound maps. These maps could be employed within probabilistic global robot localization systems. These systems use a priori probabilistic map in order to estimate the global robot position. The method has been tested in a real environment using two commercial Pioneer 2AT robotic platforms in the premises of the Department of Electronics at the University of Alcala. Some experimental results and conclusions are presented.