Hitting Flying Objects With Learning From Demonstration

There are four main steps in the flying object hitting problem. Firstly, the position and velocity of the object need to be accurately predicted ahead of time. Secondly, feasible hitting poses need to be calculated. Thirdly, a fast motion planning algorithm for the robot needs to be implemented. Lastly, the inverse kinematics of the robot needs to be derived. In this paper, a six degrees-of-freedom UR5 robot is implemented to hit a freely flying ball. The dynamics of the ball is derived, and the analytical inverse kinematics model of the robot is given. The Gaussian Mixture Model (GMM) and Gaussian Mixture Regression (GMR) are used to encode human hitting demonstrations based on an autonomous dynamical systems model. Experimental results performed in the simulation environment have validated the effectiveness of the proposed method.