Markov-localization through color features comparison

Self-localization plays a fundamental role in all the activities of a service mobile robot, from simple point-to-point navigation to complex fetch-and-carry tasks. In particular, in presence of an environment which changes dynamically, a trade-off must be found between apparently opposite characteristics: uniqueness (i.e. the ability to univocally recognize every location in the environment) and ductility (i.e. the ability to recognize a location of the environment in spite of small changes). The paper shows a vision-based approach which exploits color analysis and clustering to match perceptions with a pre-stored model of the environment, and relies on a Markovian model to update a probability density over the possible robot's configurations.