Developing Peer-Trained Autonomous Companion Robots to Tackle Isolation

Abstract: There has been a sharp rise in the loneliness and social isolation felt amongst young people, resulting in an increased risk of developing many harmful conditions. Quarantining and remote working due to the recent COVID-19 pandemic have made the situation worse. Therefore, this thesis aims to develop a companion robot that provides social support and reduces the social isolation and loneliness felt by students in their dorm rooms. A Wizard-Of-Oz style study is used to capture the ideal behaviour of the robot. By allowing the human to take full control of the robot and deploying it with a user, the actions and behaviour the human chooses for the robot can be recorded and later used to create autonomous behaviour for the robot. The autonomous behaviour is controlled by two machine learning classification models. Information collected from the environment and the interactions with the user are combined to create a statespace. Given an input statespace, the first model determines if an action should be performed, and the second model determines which action this should be. The results show that the companion robot successfully reduced the amount of social isolation and loneliness felt by the students it was deployed with. The robot exhibited both socially engaging and autonomous behaviour, and the behaviour was personalised to each individual user.