Toward a New Paradigm for Imitation Learning and Explainable AI Inspired by Cognitive Memory

We introduce an algorithm for learning structures of symbolic concepts in an ontology through on-line observations of the environment. The algorithm recognises scenes and sub-scenes based on a general-purpose taxonomy we formalised, which should be implemented for each application. The algorithm is used to implement cognitive-like abilities, such as: consolidating, retrieving and forgetting. We also introduce tools we developed to evaluate the algorithm in a human-robot collaboration scenario.