A deep learning approach in robot-assisted behavioral therapy for autistic children

• Published in: International Journal of Advanced Trends in Computer Science and Engineering
• Main Author: 2-s2.0-85078254537
• Format: Article
• Language: English
• Published: World Academy of Research in Science and Engineering 2019
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85078254537&doi=10.30534%2fijatcse%2f2019%2f6381.62019&partnerID=40&md5=8148b109ba08890237c6f80bcac77ed0

A significant percentage of the world's children are being diagnosed with Autism Spectrum Disorders (ASD every day. According to the most recent reports for Disease Control Data (DCD), ASD affects one in 68 children in the US only. It has been recognized as a neurological disorder characterized by difficulties in social communication and social interaction; abnormal body posturing; repetitive movements and self-abusive behavior. There is no cure for ASD but efforts to mitigate difficulties in social functioning, learning, and to improve quality of life of persons with ASD is usually through behavioral therapy. Robot-assisted behavioral therapy is one emerging field that provides intervention mainly for children with ASD, so far, only to augment traditional rehabilitation approaches. In this approach, robots have been used for different purposes such as for behavior eliciting, rehearsing skills, and improving interaction and socialization skills. Nonetheless, there are still a lot to be done in developing robots that can effectively work towards improving social and emotional confidence in children with ASD. This paper sheds light on recent studies that utilize deep learning technique and sets out to propose a deep learning-based emotion detection system for humanoid robots to enhance robot awareness during therapy sessions. We present a model of the emotion-aware robot-assisted therapy which is expected to ease the prediction and recognition for the emotion and behaviors of autistic children and enhance robot intervention during rehabilitation. It was found that the proposed DL model when tested on an improved trial dataset of normal subjects has increased the accuracy of detection. However, while new deep learning technologies for facial expression recognition algorithms could lead to higher detection accuracy, it is clear from that the size and reliability of the data will be the success factor in this study. © 2019, World Academy of Research in Science and Engineering. All rights reserved.