Computational Behaviour Generation for a Robotic Coach for Well-being

Social robots are continuously being researched and applied to new domains, with applications being developed for socially complex environments such as hospitals, classrooms, and homes. Building these types of applications requires that social robots' interaction and social capabilities be continuously researched to increase their smoothness, usefulness and adaptation capabilities. A central part of human-to-human social interaction is the sensing and expression of emotions. This makes Emotional Intelligence (EIQ) an important research area in the field of Human-Robot Interaction (HRI), as endowing robots with emotion skills can help make them feel more natural as interaction partners to humans. In applications which aim to increase human well-being, researching affective behaviours for robots is important, due to their potential to enable a robot to express empathy toward its user.


Social robots have been previously researched as tools to increase well-being, such as in therapy with children with autism, and in stroke rehabilitation. Robots with EIQ as tools to increase and maintain well-being in the context of coaching is an unexamined research area. Research shows that coaching can have a role in maintaining subjective and psychological well-being, and preventing mental health issues. By creating a robotic coach that utilizes adaptive affect (i.e. a robot that adapts its emotional expression according to the ongoing interaction), new technology can be used to improve both people's well-being, and to maintain the strength of the workforce. A robotic coach could be used to address existing challenges with both self-coaching and working with a human coach, such as issues of lacking motivation and embarrassment.


To apply EIQ in a coaching context, the robot will incorporate state-of-the-art methods of affective computing and machine learning. The computational challenge for this research will be creating a model for adaptive affective (displaying emotion) robot behaviour. The goal is to enable the robot to adapt its own emotional state over time during HRI interactions, by sensing the coachee's immediate emotional state, as well as by creating a model of the ongoing interaction and its emotional context. To build such a model, different methods of machine learning, such as Reinforcement Learning, Bayesian Methods, and Markov Models will be compared for suitability.


The research will focus on (1) understanding what behaviours are useful in HRI in a coaching context, and (2) understanding the effects of a robotic coach on the user's well-being. The research aims to work together with experts (such as psychologists and potential coachees), applying participatory design methods to understand what type of robot and HRI scenarios would be most useful. Findings of the participatory design process will be applied to create appropriate HRI scenarios that incorporate EIQ, and examined for their effects.