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. Answer Set Prolog, a non-monotonic logical reasoning paradigm, is used to represent and reason with incomplete commonsense domain knowledge, computing a plan for any given goal and diagnosing unexpected observations. ASP-based reasoning is also used to guide the interactive learning of previously unknown actions as well as axioms that encode affordances, action preconditions, and effects. This learning takes as input observations from active exploration, reactive action execution, and human (verbal) descriptions, and the learned actions and axioms are used for subsequent reasoning. The architecture is evaluated on a simulated robot assisting humans in an indoor domain.