Study on an assistive robot for improving imitation skill of children with autism

In this paper, we report the effectiveness of a therapeutic humanoid robot for children with autism to improve his/her imitation skill. In order to realize this system, a robot has to provide two functions: mimicking and evaluating the child's motion in real time. The former function is achieved by selecting key frames using the Q-Learning approach to remove noisy camera data. The latter is established via a cluster-based framework on Mixture Gaussian and Expectation-Maximization algorithm using parameters which are converted by Principal Component Analysis. Practical experiments are shown in which autistic children interact with a robot and are trained by executing specific tasks for improving imitation skill.