In this paper we combine robot control and data analysis techniques into a system aimed at early detection and treatment of autism. A humanoid robot - Zeno is used to perform interactive upper body gestures which the human subject can imitate or initiate. The result of interaction is recorded using a motion capture system, and the similarity of gestures performed by human and robot is measured using the Dynamic Time Warping algorithm. This measurement is proposed as a quantitative similarity measure to objectively analyze the quality of the imitation interaction between the human and the robot. In turn, the clinical hypothesis is that this will serve as a consistent quantitative measurement, and can be used to obtain information about the condition and possible improvement of children with autism spectrum disorders. Experimental results with a small set of child subjects are presented to illustrate our approach.