Researchers have developed an AI framework called AdaPT that enables humanoid robots to learn professional tennis serving and rally styles from broadcast videos. The framework consists of two main components: a planner that generates stylistic kinematic motions and a tracker that executes them with minimal interference. However, the team found that there is a significant gap between simulation and real-world performance, which they addressed by introducing an adaptation mechanism to improve tracking robustness. They tested AdaPT on two humanoid robots, including a full-size robot, and demonstrated its effectiveness in serving without motion capture. The researchers also shared videos and code for their project online. --- Why it matters: This research is significant because it tackles the challenge of achieving professional motion styles in humanoid robots while maintaining strong task performance. It has implications for future ball-sports systems and could potentially be applied to other areas where human-like movement is required. Source: https://arxiv.org/abs/2608.20087