Semantic Localization System for Robots at Large Indoor Environments Based on Environmental Stimuli.

In this paper, we present a new procedure to solve the global localization of mobile robots called Environmental Stimulus Localization (ESL). We propose that the presence of common facts on the environment around the robot can be considered as stimuli for the procedure. The robust performance of our approach is supported by two concurrent particle filters. A primary particle filter estimates and tracks the robot position, while a secondary filter is fired by environmental stimuli, helps to reduce the influence of measurement errors and allows an earlier recovery from localization failures. We have successfully used this method in a 5000 m 2 real indoor environment using as inputs the available environment information from a Geographical Information System (GIS) map, the robot's odometry and the output of an algorithm for the perception of facts from the environment. We present a case study and the result of different tests, showing the performance of our method under the influence of errors in real applications.