12/10/2026
Fifty kilometres per hour is a solid club serve, and a humanoid robot just returned one live on a global broadcast.
Galbot's machine played a full autonomous rally on 22 August at Beijing's National Speed Skating Oval, during the second World Humanoid Robot Games, per LeiPhone.
It served, drove forehands and backhands and volleyed against Zheng Jie, a former Grand Slam doubles champion, then got up on its own after a hard fall.
Galbot says its AstraBrain model runs brain, cerebellum and control as one system, with inference latency below 1.5 milliseconds after a 92.58 per cent zero-shot result in the lab.
The interesting part is not the rally.
Galbot skipped perfect motion capture from elite athletes and trained on messy clips from ordinary players, letting a model find the physics that makes robot movement cheap.
Cheap data is what turns a lab stunt into a product line, and tennis stands in for the factories and kitchens where spacing and surprise change every second.
Making a robot swing a racket is easy, and making it get up after a fall is what changes the industry.
Full story 👇
Galbot's Humanoid Robot Plays Tennis, a Milestone for Embodied AI