A Close Look at the Imitation Performance of Children with Autism and Typically Developing Children Using a Robotic System

Deficit in imitation skills is one of the core symptoms of children with Autism Spectrum Disorder (ASD). In this study, we have tried to look closer at the body gesture imitation performance of 20 participants with autism, i.e. ASD group, and 20 typically developing subjects, i.e. TD group, in a set of robot-child and human-child gross imitation tasks. The results of manual scoring by two specialists indicated that while the TD group showed a significantly better imitation performance than the ASD group during the tasks, both ASD and TD groups performed better in the human-child mode than the robot-child mode in our experimental setup. Next, to introduce an automated imitation assessment system, we present different mathematical models of the children's imitation performance using some State-Image based algorithms including Acceptable Bound, Mahalanobis Distance, and Signals' Cross-Correlations as well as Hidden Markov Models based on the time-dependent kinematics data of the participants' joints. Among the different studied models, we observed that the "State-Image Acceptable Bound method with position, velocity, and acceleration features" is the best one. This method has a mean Pearson correlation of similar to 45%, which is fairly comparable to the related works (out of autism field) in assessing the quality of dynamic actions. Finally, for a treatment application of using artificial intelligence algorithms in automated evaluation of children's behaviors as an unbiased and quantifiable measurement in HRI, we propose a reciprocal gross imitation human-robot interaction platform with the potential to aid in the cognitive rehabilitation of children with autism.