Mimicking and Evaluating Human Motion to Improve the Imitation Skill of Children with Autism Through a Robot

In this paper, we report techniques for mimicking and evaluating the human motion in real time by a therapeutic humanoid robot to improve the imitation skill of children with autism. For realizing the mimicking technique, we propose a method of selecting key frames using a Q-Learning approach to remove the significant noises. Then, in order to evaluate human motion in real time, we introduce a method of cluster-based framework of Mixture Gaussian and an Expectation-Maximization algorithm using parameters which are converted by Principal Component Analysis. Practical experiments have been performed to test the interaction of children with autism with the robot and evaluate the possibility of improving their imitation skills by training them to perform specific tasks through a robot.