Abstract
A critical challenge in the design of robots that operate while interacting with humans is to ensure mutual understanding, which contributes to build reliable human-robot interactions. It is an arduous task since interactive scenarios are often uncertain, exposing robots to exogenous situations that affect their ongoing activities. In those cases, robots shall perceive and recognize unexpected changes in the environment, represent and reason about them, and decide how to adapt to them. This will certainly modify robots’ internal knowledge, and it is fair to assume that part of the new robot beliefs might be hidden from other agents such as humans. Hence, robots shall also be capable of communicating or explaining the relevant knowledge about those beliefs updates. In this context, this thesis investigates the use of ontologies as an integrative framework for the construction of robot explanations, particularly within interactive settings involving humans. To this end, the thesis starts formulating the scope of the relevant domain knowledge to conceptualize, and it continues proposing novel ontological models and methods for ontology-based robot explanation generation. The first part of the thesis discusses two main contributions: a systematic review and classification of the state-of-the-art that narrows down the target set of reality phenomena to be conceptualized, and the investigation and development of novel robot perception methods to extract from realistic robot experiences the common patterns of the target conceptualization. The second part discusses the two remaining contributions: ontological analysis and modeling of the target domain knowledge, and the design and development of algorithms to construct ontology-based robot explanations. Note that the different ontological models and algorithms were mainly validated in collaborative and adaptive robotic scenarios. However, they were conceived from a foundational perspective, and we think that their scientific value extrapolates to other application domains (e.g. assistive robotics or non-robotic agents). Overall, the scientific contributions of this thesis set a solid foundational basis for the ontology-based explainable robots domain, boosting the design of trustworthy interactive robots.