Development and Evaluation of an Autonomous, Virtual Agent Based Social Skills Tutor for Children with Autism

Author: Marissa Milne

Milne, Marissa, 2018 Development and Evaluation of an Autonomous, Virtual Agent Based Social Skills Tutor for Children with Autism, Flinders University, College of Science and Engineering

Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact with the details.


This is a multidisciplinary thesis comprising principally of a computer science component involving the development of educational software for teaching children with autism social skills using virtual humans, with a follow up evaluation of the software using social science methods. It addresses a mixed audience and provides technical detail as expected in the mathematical and computational sciences.

Individuals with autism experience difficulties with social skills and can find understanding the nonverbal cues and social behaviours of other people challenging. This makes building friendships and other appropriate relationships difficult, which can lead to isolation, social anxiety and depression, impacting their overall wellbeing. Further, many individuals with autism report an affinity for technology and exhibit high technology usage patterns. Using virtual humans to teach social skills to children with autism harnesses this preference for technology and provides a tool that can support the development of social skill knowledge and behaviour, ultimately aiming to improve individuals' everyday wellbeing.

Existing research with children with autism suggests that autonomous (self-directed) virtual humans can be used successfully to improve language skills (Bosseler and Massaro 2003) and authorable (researcher controlled) virtual humans can be used to improve social skills (Tartaro and Cassell 2006). The original contribution of this research is to combine these ideas and investigate the use of autonomous virtual humans for teaching basic social skills in the areas of greeting, conversation skills, and listening and turn taking.

The software in this research features three virtual humans who guide the learner through tasks and model social scenarios: a teacher, a peer with strong social skills, and a peer with developing social skills. Thirty one participants were assigned to either the control or experimental group using a matched pairs approach then asked to use the software for 10-15 minutes per day, 3-5 days per week for three weeks, with data collected before software use, at the end of the three week period, and again two and four months later. The data collected included a content quiz testing participant knowledge, the Vineland-II evaluating social behaviours as observed by caregivers, and questionnaires assessing participants' prior experience and expectations, and participant and caregiver perceptions following software use. The software itself also automatically recorded log data reflecting participant interaction with the system.

The Social Tutor was generally well-received by participants and caregivers, although more game-like elements and some adjustments to the virtual humans themselves and the lesson sequencing algorithm were requested for future development. Evaluation data likewise indicated positive trends, with a clear difference between performance of the experimental (social content) group and the control (non-social content) group in the content quiz. Vineland data was less clear with both groups performing similarly overall, although some encouraging trends were seen. Future work should focus on further generalisation support with the aim of converting the changes in knowledge demonstrated by participants to changes in real-world behaviour. A follow up evaluation with a larger sample size and longer software use period would also be beneficial to ascertain if any of the apparent promising trends observed in the data eventuate to significant outcomes.

Keywords: autism, Autism Spectrum Disorder, Asperger Syndrome, high functioning autism, social skills, conversational skills, autonomous virtual tutor, virtual human, pedagogical agent, computer assisted learning, special education

Subject: Computer Science thesis

Thesis type: Doctor of Philosophy
Completed: 2018
School: College of Science and Engineering
Supervisor: David Powers