Collaborative HRI and Machine Learning for Constructing Personalised Physical Exercise Databases

Abstract

Recent demographics indicate that we have a growing population of older adults with increasingly complex care-related needs, and a shrinking care workforce with limited resources to support them. As a result, there are a large number of research initiatives investigating the potential of intelligent robots in a domestic environment to augment the support care-givers can provide and improve older adults’ well-being, particularly by motivating them in staying fit and healthy through exercise. In this paper, we propose a robot-based coaching system which encourages collaboration with the user to collect person-specific exercise-related movement data. The aim is to personalise the experience of exercise sessions and provide directed feedback to the user to help improve their performance. The way each individual user is likely to perform specific movements will be based on their personal ability and range of motion, and it is important for a coaching system to recognise the movements and map the feedback to the user accordingly. We show how a machine learning technique, a Nearest Neighbour classifier enhanced with a confidence metric, is used to build a personalised database of 3D skeletal tracking data. This approach, combined with collaborative Human-Robot Interaction to collect the data, could be used for robust and adaptable exercise performance tracking by a collaborative robot coach, using the information to provide personalised feedback.