Galbot’s human-like robots kept up over 100 tennis rallies in a row with no human in control.
The run came as the Second World Humanoid Robot Games (WHRG) opened in Beijing, in what the company depicted as the first live autonomous humanoid tennis match.
Five hours of amateur footage taught the G1 to swing
The demonstration comes with a purposefully loaded label from Beijing Galaxy General Robot Co. It’s calling the run tennis’s “AstraTennis moment,” a reference to AlphaGo’s 2016 win over Go champion Lee Sedol.
Where AlphaGo mastered a game played on a board, this was a physical contest played on a real court, in real time. Galbot said the rally streak set a record for humanoid robot tennis, and the demonstration was done on a live broadcast.
The game was played at the opening ceremony of the WHRG. This second edition hosted 2,056 machines from 666 teams in 16 countries.
Each robot had to take in the incoming shots and determine whether to play a forehand, backhand, serve, return, or a scrambling recovery shot.
Galbot’s robots also played doubles with human tennis champions. The robots altered their positioning and strategies on the court.
In quick exchanges, when the machines were knocked off-kilter, they righted themselves and stayed in the point.
Galbot’s G1 is the physical platform, a 173 centimeter, 85 kilogram humanoid with a 190 centimeter arm span and a torso lift that protracts its reach to 240 centimeters. Those dimensions were chosen to mimic how a human would cover a court.
Each arm has 7 degrees of freedom and interchangeable end-effectors, including two-finger grippers and 22-degree-of-freedom dexterous hands. Visual and tactile sensors feed the perception system, tracking the ball’s speed, spin and path.
The software, named LATENT, stands for Learning Athletic Humanoid Tennis Skills from Imperfect Human Motion Data.
The framework decomposes tennis into small reusable motion primitives, forehands, backhands, lateral shuffles and crossover steps, and then reassembles them into full strokes. Its selling point, according to Galbot, is that it learns on coarse amateur data.
Researchers amassed about five hours of tennis motion from recreational players using a small rig. They then employed reinforcement learning and large-scale simulation to train the robot to pick and mix the right motions.
Tiangong Ultra ran 9.39 seconds, Bolt ran 9.58
Cryptopolitan reported that a headless humanoid named Tiangong Ultra, built by the government-backed Beijing Innovation Center of Humanoid Robotics, ran a 100-meter sprint in 9.39 seconds. That bettered the 9.58 seconds Usain Bolt set in 2009.
The same model then completed a 400-meter dash in an adjusted time of 38.15 seconds.
Cryptopolitan reported that some of the runners failed to start or moved at a walking pace, while one of the boxing robots fell into the ropes and remained still until people carried it away.
At the Beijing venue, one retiree said the sprints were “just a competition of speed. It’s not a test of intelligence.”
Unitree Robotics claimed its Superman humanoid clocked 12.66 meters per second, although Cryptopolitan noted that those figures are from a company video and there is no independent verification.
WORLD RECORD!
Galbot robot completed 100+ consecutive autonomous tennis rallies with zero mistakes!
🎾 Fully autonomous.
📷 Live Broadcast.A new milestone for humanoid robot tennis! #AstraTennis #EmbodiedAI #HumanoidRobots #WHRG #Galbot https://t.co/W6N1b7pmzx pic.twitter.com/c7uTrX09ES
— Galbot (@GalbotRobotics) August 23, 2026
Don’t just read crypto news. Understand it. Subscribe to our newsletter. It’s free.
Be the first to comment