Researchers from Tsinghua University and collaborators developed LATENT, a system enabling humanoid robots to learn tennis skills from imperfect human motion data. The approach addresses a fundamental challenge in robotics: the lack of perfect motion data for complex athletic activities. Instead of requiring complete human-tennis sequences, LATENT works with motion fragments capturing primitive tennis skills.

Despite data imperfections, the researchers recognized that quasi-realistic human movement still provides valuable priors about fundamental tennis skills. Through correction and composition, the system learned policies allowing robots to strike incoming balls across various conditions while maintaining natural motion styles. The team implemented specialized designs for sim-to-real transfer to bridge simulation and reality.

Deployed on the Unitree G1 humanoid robot, the method achieved remarkable results in real-world testing. The robots demonstrated the ability to sustain multi-shot rallies with human players, showcasing effective reactive footwork and strategic ball placement. This breakthrough demonstrates progress toward human-robot collaboration in dynamic physical activities.