A new data processing architecture for table tennis robot

Playing table tennis with human is a challenging work for robot. The common architecture for table tennis robot in the past researches are to gather information at first, then predict the ball status at hitting point, and the motion mechanism moves according to the prediction at last. In order to save the time for the later two steps, the gathered information is limited. In this paper, we propose a new architecture, in which the robot would keep gathering the information and amending the prediction as the motion mechanism moves. To combine the gathered information and obtain reliable prediction of hitting point, a data processing algorithm is proposed. The algorithm is based on the kinetic analysis of the ball and the extended Kalman filter. The estimation of angular velocity measurement is also improved in this algorithm. Finally, the validity of the algorithm is confirmed by conducting experiments.