Socially assistive robots primarily provide useful functionality through their social interactions with user(s). An example application, used to ground work throughout this thesis, is using a social robot to guide users through exercise sessions. Initial works have demonstrated that interactions with a social robot can improve engagement with exercise, and that an embodied social robot is more effective for this than the equivalent virtual avatar. However, many questions remain regarding the design and automation of socially assistive robot behaviours for this purpose. This thesis identiﬁes and practically works through a number of these questions in pursuit of one ultimate goal: the meaningful, real world deployment of a fully autonomous, socially assistive robot. The work takes an expert-informed approach, looking to learn from human experts in socially assistive interactions and explore how their expert knowledge can be reﬂected in the design and automation of social robot behaviours. It is taking this approach that leads to the notion of socially assistive robots needing to be persuasive in order to be effective, but also identiﬁes the difﬁculty in automating such complex, socially intelligent behaviour. The ethical implications of designing persuasive robot behaviours are also practically considered; with reference to a published standard on ethical robot design. The work culminates with use of a state of the art, interactive machine learning approach to have an expert ﬁtness instructor train a robot ‘ﬁtness coach’, deployed in a university gym, as it guides participants through an NHS exercise programme. After a total of 151 training sessions across 10 participants, the robot successfully ran 32 sessions autonomously. The results demonstrated that autonomous behaviour was generally comparable to that of the robot when controlled/supervised by the ﬁtness instructor, and that overall, the robot played an important role in keeping participants motivated through the exercise programme.
|Date of Award||29 Sep 2020|
- The University of Bristol
|Supervisor||Arthur G Richards (Supervisor)|