Scene-Dependent Intention Recognition for Task Communication with Reduced Human-Robot Interaction

Abstract

In order for assistive robots to collaborate effectively with humans, they must be endowed with the ability to perceive scenes and more importantly, recognize human intentions. These intentions are often inferred from observed physical actions and direct communication from fully-functional individuals. For individuals with reduced capabilities, it may be difficult or impossible to perform physical actions or easily communicate. Therefore, their intentions must be inferred differently. To this end, we propose an intention recognition framework that is appropriate for persons with limited physical capabilities. This framework determines and learns human intentions based on scene objects, the actions that can be performed on them, and past interaction history. It is based on a Markov model formulation entitled Object-Action Intention Networks, which constitute the crux of a computer vision-based human-robot collaborative system that reduces the necessary interactions for communicating tasks to a robot. Evaluations were conducted on multiple scenes comprised of multiple possible object categories and actions. We achieve approximately 81% reduction in interactions overall after learning, when compared to other intention recognition approaches.