Design of a humanoid ping-pong player robot with redundant joints

Jointly learning trajectory generation and hitting point prediction in robot table tennis

“…As a comparison, we also run a traditional method in Fig. 2, which predicts a strike point on the virtual striking plane using a flight model [2] and a linear rebound model [1], then determines the desired racket state with a simplified ball-racket contact model and the desired joint state with inverse kinematics, finally generates robot trajectory in joint space using fifth-order polynomials [3], [8]. We calculate striking rates every 200 trials; Meanwhile, we can statistically analyse actual landing points according to landing errors d. Evaluation results are shown in Fig.…”

“…Especially, the linear rebound model can not characterize the rebound phenomenon well. For some ball trajectories, the virtual striking plane [1], [3] leads to improper striking points for the robot, and it is very hard to hit balls at these points.…”

“…In the previous publications for the ball trajectory prediction, a predefined virtual striking plane is introduced. For the physical model methods [1], [2], [3], [4], [5], the intersection point between the predicted ball rebound trajectory and the predefined virtual plane is considered as the striking point. For the regression methods [6], [7], the striking point at the predefined virtual plane is predicted directly.…”

Design of a Redundant Manipulator for Playing Table Tennis towards Human-Like Stroke Patterns

“…The challenging tasks of playing table tennis performed by robots have attracted many researchers [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16]. They employed the table tennis robot as extensive research platform for artificial intelligence [3,6], advanced control algorithms [1,8], high speed machine vision and prediction [13], fast responses against unforeseen events [17], online trajectory generation within very limited time [2,10,15], target tracking and intercepting [6,14], and so forth.…”

“…Zhang et al [13] built a five-DOF table tennis robot whose racket was restricted to the half-table. Recently, Lai and Tsay [12], Yu et al [14,16], and Zhang et al [15] presented a seven-DOF robot to achieve playing table tennis dexterously. Mülling et al [18] presented a biomimetic trajectory generation method to robot table tennis.…”