Action-Embedded Human-Robot Interaction for Hands-Off Assistive Domains

Contents

How can we make robots effective at helping people to do things they do not really want to do? This question is at the heart of assistive robotics, a new field that concerns itself with enabling robots to empower people in need, including those who are convalescing from an illness or medical procedure, those with physical and/or cognitive disabilities, those who are rehabilitating from stroke or serious trauma, those attempting to avoid some natural or man-made disaster or cope with its aftermath, and many others. In assistive domains, achieving and maintaining user engagement is only part of the larger picture, and is often in conflict with other goals to be achieved. The challenges for assistive human-robot interaction are found at the juxtaposition of enabling robots to be both helpful and engaging. Good human teachers, coaches, and nurses are rare and given the challenges of real-time on-board perception and action, good robot teachers, coaches, and nurses are not yet within reach, thus making them a compelling research topic. Our work in assistive robotics is based on action-embedded representation, the use of body language and movement as a powerful means of interaction, and for implicit conveyance of information and intent. By using a set of built-in behavior primitives as the underlying "social vocabulary" for the robot, we enable our robots to understand and interpret observed activity of the human user by mapping it onto their own social vocabulary, and using that vocabulary to convey the intended content (coaching, instruction, etc.), as well as receiving aid and clarification as needed. The social vocabulary thus becomes the core of the human-robot interaction, enabling activity recognition, classification, and generation, as well as learning. The ability to expand the social vocabulary is critical for adapting to the user and improving the robot’s (and the user’s) performance. We will describe the foundations for this work, demonstrate them on mobile robots learning behaviors and tasks from human users, and show how we are adapting the approach to enable robots to become effective teachers, trainers, and coaches in assistive contexts.