Abstract

Robots are more and more present in our lives, particularly in the health sector. In therapeutic centers, some therapists are beginning to explore various tools like video games, Internet exchanges, and robot-assisted therapy. These tools will be at the disposal of these professionals as additional resources that can support them to assist their patients intuitively and remotely. The humanoid robot can capture young children’s attention and then attract the attention of researchers. It can be considered as a play partner and can directly interact with children or without a third party’s presence. It can equally perform repetitive tasks that humans cannot achieve in the same way. Moreover, humanoid robots can assist a therapist by allowing him to teleoperated and interact from a distance. In this context, our research focuses on robot-assisted therapy and introduces a humanoid social robot in a pediatric hospital care unit. That will be performed by analyzing many aspects of the child’s behavior, such as verbal interactions, gestures and facial expressions, etc. Consequently, the robot can reproduce consistent experiences and actions for children with communication capacity restrictions. This work is done by applying a novel approach based on deep learning and reinforcement learning algorithms supported by an ontological knowledge base that contains relevant information and knowledge about patients, screening tests, and therapies. In this study, we realized a humanoid robot that will assist a therapist by equipping the robot NAO: 1) to detect whether a child is autistic or not using a convolutional neural network, 2) to recommend a set of therapies based on a selection algorithm using a correspondence matrix between screening test and therapies, and 2) to assist and monitor autistic children by executing tasks that require those therapies.

Citation

I. Salhi, M. Qbadou, S. Gouraguine, K. Mansouri, C. Lytridis, V. Kaburlasos, “Towards robot-assisted therapy for children with autism – the ontological knowledge models and reinforcement learning-based algorithms”, Frontiers in Robotics and AI, 06 April 2022, vol. 9, Article 713964. doi: 10.3389/frobt.2022.713964 (Special Issue on Emerging Technologies for Assistive Robotics: Current Challenges and Perspectives – Section “Biomedical Robotics”. Guest Editors: Lyuba Alboul, Jacques Penders, Peter Mitrouchev, Maya Dimitrova, Anna Lekova, Vassilis Kaburlasos)

View on Frontiers