Human-like Hitting Strategies for a Ball Bouncing Robot

Hand-eye coordination control is one of the dexterous operational skills of mankind. This study reproduces human-like hitting strategies based on hand-eye coordination techniques for bouncing a ping pong ball with a robotic manipulator holding a paddle and an RGB-D camera. The target ball and the background are first separated by using the color information. The target ball’s position in the world coordinate frame can therefore be obtained by incorporating the depth information. Analysis of forces exerted on the ball is able to predict its future motion trajectory. In addition, a to-the-center and an adaptive learning hitting strategies based on manipulation skills of humans are developed to overcome difficulties caused by uncertainties and unknown parameters for the successive bouncing task. The to-the-center strategy tends to bounce the ball towards the center of the paddle, not just vertically upwards in a classical approach, in order to maintain the ball in the reachable region. However, the adaptive learning strategy provides controls of the inclination and the hitting force for the paddle according to previous bounce behavior of the ball. Actual bouncing experiments with a three degrees-of-freedom robotic wrist were conducted using three different bouncing strategies: the classical vertical approach, the to-the-center strategy, and the adaptive learning strategy. Experimental results demonstrate that the proposed adaptive learning hitting strategy displays best bouncing performance in terms of average bouncing number of times and average distance of contact point away from the center.