AdaPT Framework for Professional Tennis Motion in Humanoid Robots

AdaPT uses a hierarchical design to learn professional tennis styles from broadcast videos, separating stylistic kinematic planning from execution tracking. The framework addresses sim-to-real gaps by learning to track randomized execution speeds to improve robustness against noisy perception.