Humanoid Robot Plays Tennis Against a Human — and It’s a Rally, Not Just a Few Shots
On March 16, Galbot Robotics released a video that quickly went viral on Reddit: the Unitree G1 humanoid robot playing tennis against a human opponent. The robot doesn’t just return a couple of shots — it engages in a full rally, moving across the court and responding to fast-paced hits.
The robot is powered by LATENT (Learning Athletic Humanoid Tennis Skills from Imperfect Human Motion Data), a system developed by researchers from Tsinghua University, Peking University, and Galbot.
A major challenge in training robots for sports is the lack of high-quality motion data. Full-body motion capture for tennis swings is expensive and often imperfect — sensors lose accuracy during sharp movements, and some data ends up being unusable. Rather than demanding flawless recordings, LATENT learns to reconstruct complete motion sequences from fragmented, imperfect data, filling in the gaps on its own.
The result is a robot with millisecond-level reaction time, capable of coordinating its legs, torso, and arm simultaneously. In simulations, its shot accuracy reaches 96%; in real-world conditions, it achieves around 90% — an impressive feat for a machine standing just over one meter tall.
“This is the first time a humanoid robot can perform high-dynamic, sustained tennis rallies with millisecond-level reaction time, precise shots, and natural full-body motion,” the Galbot team wrote in a post on X.
Tennis is one of the most challenging tasks in robotics. Unlike walking or simple object manipulation, it demands simultaneous coordination of movement, balance, and precise striking of a moving target. Until now, robots had only mastered table tennis — where the range of motion is minimal — while a full-sized court remained out of reach.
The LATENT system was trained on just five hours of motion data — an order of magnitude less than what is typically required for such complex tasks. If the approach proves scalable to other sports and physical activities, it could dramatically accelerate how robots learn advanced motor skills.