ABSTRACT
This chapter describes an integrated architecture for robots that combines the complementary strengths of knowledge-based and data-driven methods for transparent reasoning and learning. Specifically, the architecture builds on the principle of step-wise iterative refinement to support non-monotonic logical reasoning and probabilistic reasoning with tightly coupled transition diagrams of the domain at different resolutions. Reasoning with prior domain knowledge triggers and guides the interactive learning of previously unknown domain knowledge in the form of axioms governing domain dynamics. Furthermore, the interplay between these components is used to embed the principles of explainable agency, enabling a robot to provide on-demand relational descriptions of its decisions and beliefs in response to different types of questions. The architecture's capabilities are evaluated in the context of visual scene understanding and planning tasks performed in simulation and on physical robots.
