A vision based algorithm for active robot localization

This paper describes a method to localize a mobile robot in a structured indoor environment from visual information provided by omnidirectional images. In particular, an approximate algorithm for active robot localization in partially observable environments based on a POMDP model is introduced. Results from many different tests are presented to prove the setting of algorithm parameters. They show that a localization accuracy of less than 25 cm can be achieved even with a partial knowledge of the environment. Besides, the proposed approximated algorithm reduces the computational cost without degrading results.