Model Predictive Control Approaches for Dynamic Table Tennis Swinging

Abstract

This thesis presents three model predictive control (MPC) formulations for robotic table tennis swinging, addressing the challenge of generating precise, real-time paddle trajectories for dynamic ball interactions. We explore key differences in optimization structure, solver strategy, and real-time implementation, evaluating each approach through hardware experiments that measure strike condition tracking and hit success. The final controller integrates the full task of a table tennis possession by planning the return ball trajectory through the contact dynamics, and generating a swing to achieve it. This controller improves the hit rate of the system from 88.3% to 97.6% and significantly enhances strike condition accuracy and smoothness enabling control over the landing location and spin of the ball.