Context Awareness in shared human-robot Environments: Benefits of Environment Acoustic Recognition for User Activity Classification

Abstract

Context awareness is a key element in human-robot interaction. Being able to recognize user activity improves robot decision making when facing ordinary situations in home-like environments, as well as robot overall performance. In robotics applications, context recognition is usually performed using time of day and three subsystems: localization, perception, and dialog. The proposal described in this paper adds to this approach a fifth item to classify user activities: an environmental recognition component. The Environment Recognition Component (ERC) described in this article uses convolutional neuronal networks to classify ordinary acoustic signals present in indoor environments. This information is used by a second element, the Context Recognition Component (CRC) that infers the user activity using propositional calculus. The empirical evaluation of the framework presents an 86% of accuracy at ERC level, and the CRC inference system provides three times more contexts than the approach without ERC.