Bayesian space conceptualization and place classification for semantic maps in mobile robotics

摘要: The future of robots, as our companions is dependent on their ability to understand, interpret and represent the environment in a human compatible manner. Towards this aim, work attempts create hierarchical probabilistic concept-oriented representation space, based objects. Specifically, it details efforts taken towards learning generating concepts classify places using gleaned. Several algorithms, from naive ones only object category presence more sophisticated both objects relationships, are proposed. Both inference use information encoded underlying representation-objects relative spatial between them. approaches exemplars, clustering Bayesian network classifiers. generative. Further, even though they not ontology specific; i.e. do assume any particular ontology. presented algorithms rely robots inherent high-level feature extraction capability (object recognition structural element extraction) actually form concept models infer Thus, report presents methods that could enable robot link sensory increasingly abstract (spatial constructs). Such conceptualization results thereof would be cognizant surroundings yet, us. Experiments place classification reported. theme is-conceptualization for cognition.