A Nonlinear Least Squares Approach for Nonprehensile Dual-Hand Robotic Ball Juggling

This paper presents a nonlinear least squares approach to deal with dual-hand robotic ball juggling. The task considers the repetitive batting (throwing and catching in a single collision) of a ball between two paddles/hands in a nonprehensile way. In detail, assuming to measure the trajectory of the ball, by solving a sequence of nonlinear minimization problems through a least squares method, the configuration of the paddles at the next impact is computed online to juggle the ball between the hands. Afterwards, an optimal trajectory for the paddles is planned in SE(3). The proposed approach is evaluated on a semi-humanoid robot with 21 degrees of freedom. Numerical tests show the smoothness of the planned trajectories and the precision of the proposed juggling algorithm.