Knowledge Representation and Interactive Learning of Domain Knowledge for Human-Robot Interaction

This paper describes an integrated architecture for representing, reasoning with, and interactively learning domain knowledge in the context of human-robot collaboration. Specifically, Answer Set Prolog, a declarative language, is used to represent and reason with incomplete commonsense knowledge about the domain. Non-monotonic logical reasoning identifies knowledge gaps and guides the interactive learning of relations that represent actions, and of axioms that encode affordances and action preconditions and effects. Learning uses probabilistic models of uncertainty, and observations from active exploration, reactive action execution, and human (verbal) descriptions. The learned actions and axioms are used for subsequent reasoning. The architecture is evaluated on a simulated robot assisting humans in an indoor domain.