Online Motion Concept Learning: A Novel Algorithm for Sample-Efficient Learning and Recognition of Human Actions

Miguel Vasco, Francisco S. Melo, David Martins de Matos, Ana Paiva, Tetsunari Inamura

Abstract: Humans have the remarkable ability to interact with their environment through rich and diverse actions. In order to actuate on shared environments with humans, artificial agents should have the ability to recognize the human's actions. However the creation of an artificial agent that is capable of recognizing all possible human actions from prior training is unrealistic, given the diversity of potential actions and ways to perform them. If we aim at deploying competent artificial agents in such scenarios, this hurdle should be overcome.