Can a robot keep up with Serena Williams?<\/p>\n
Researchers have taught a humanoid robot<\/a> to play tennis with humans \u2014 and it can hold its own.<\/p>\n Chinese AI robotic company Galbot designed software to teach a Unitree G1 humanoid robot to play tennis against a human engineer.<\/p>\n The company posted a video to social media<\/a> showing a white robot holding what appears to be an unmodified tennis racket and using it to return the ball as it shuffles across the court.<\/p>\n \u201cYour humanoid tennis player is here!\u201d Galbot wrote on X. \u201cFor the first time, a humanoid robot can sustain high-dynamic, long-horizon tennis rallies with millisecond-level reactions, precise ball striking, and natural whole-body motion.\u201d<\/p>\n \u201cThis marks a leap from mechanical motion imitation to intelligent, decision-driven athletic interaction.\u201d<\/p>\n Researchers have taught a humanoid robot to play tennis with humans. Galbot<\/p>\n The software is dubbed LATENT (Learning Athletic Humanoid Tennis Skills from Imperfect Human Motion Data), and the company claims it\u2019s the world\u2019s first real-time whole-body planning and control algorithm for athletic humanoid tennis.<\/p>\n According to a yet-to-be-peer-reviewed paper<\/a>, the system had to rely on \u201cimperfect human motion data\u201d consisting only of \u201cmotion fragments that capture the primitive skills used when playing tennis\u201d rather than clean motion capture from \u201creal-world tennis matches.\u201d<\/p>\n The short fragments of human movement used were made up of things like forehand swings, backhand strokes and basic footwork. These motion fragments become a library of movement building blocks which the robot stitches together and figures out how to combine them in real time.<\/p>\n A white robot is holding what appears to be an unmodified tennis racket and using it to return the ball. Galbot<\/p>\n \ud83c\udfbeYour humanoid tennis player is here!\ud83e\udd16<\/p>\n Introducing LATENT (Learning Athletic Humanoid Tennis Skills from Imperfect Human Motion Data) \u2014 the world\u2019s first real-time whole-body planning and control algorithm for athletic humanoid tennis.<\/p>\n For the first time, a humanoid robot can\u2026 pic.twitter.com\/gCi38wxHVQ<\/a><\/p>\n \u2014 Galbot (@GalbotRobotics) March 16, 2026<\/a><\/p>\n When it comes to wrist control, the robot\u2019s high-level controller directly adjusts the wrist during play rather than using the \u201cimperfect\u201d data.<\/p>\n The robot can sustain multi-shot tennis matches with humans, reacting to balls traveling over 15 meters per second, which is about 33.5 miles per hour, and manages to produce coordinated strokes and footwork.<\/p>\n The movements produced look relatively natural \u2014 especially for a robot. It\u2019s not exactly fluid like a human, but it\u2019s not rigid and robotic either.<\/p>\n The robot can react to balls traveling over 15 meters per second, which is about 33.5 miles per hour. Galbot<\/p>\n \u201cOur key insight is that, despite being imperfect, such quasi-realistic data still provide priors about human primitive skills in tennis scenarios,\u201d the researchers found. <\/p>\n \u201cWith further correction and composition, we learn a humanoid policy that can consistently strike incoming balls under a wide range of conditions and return them to target locations, while preserving natural motion styles.\u201d<\/p>\n In simulation tests, the system achieved up to 96% success in forehand shots.<\/p>\n The short fragments of human movement used for data were made up of things like forehand swings, backhand strokes and basic footwork. Galbot<\/p>\n However, the engineers said that the software could be useful beyond the ability to play tennis.<\/p>\n \u201cAlthough this work primarily focuses on the tennis return task, the proposed framework has the potential to generalize to a broader range of tasks where complete and high-quality human motion data are unavailable,\u201d they noted.<\/p>\n If a robot can learn a complicated physical skill like tennis from imperfect data, it suggests that similar approaches can work for real-world tasks<\/a> as well.<\/p>\n