This paper describes a learning system that has the ability to estimate the flight trajectory of a spinning ball. This system plays an important role in realizing a robot that plays ping-pong against a human. LWR (Locally Weighted Regression) is employed to learn a forward map from stereo image inputs of an incoming ball to the state outputs of the ball just before hitting. The experimental results show that the ball trajetory can be estimated accurately irrespective of how the ball is spinning.
