Human-Robot Interaction Based on Introspective Capability

The paper describes a mechanism of introspection of a humanoid robot based on self observation of its running processes and sensory data. The robot builds is internal state representation by a suitable ontology: a working knowledge of its physical structure and functionalities, that it can exploit to interact with humans in order to describe, in natural language, his state and the manner in which it perceives. We report experiments performed on a humanoid robot, describing all the components necessary to implement the proposed architecture of introspection.