Galbot Robots Play 100 Consecutive Tennis Rallies and Tiangong Ultra Runs 9.39s 100m at World Humanoid Robot Games

Two milestones from the second World Humanoid Robot Games, which opened at Beijing's National Speed Skating Oval on August 22, redraw the line between research demos and real-world humanoid capability. Galaxy General Robotics' Galbot ET1 completed more than 100 consecutive tennis rallies against former Chinese professional Zheng Jie, while the Beijing Humanoid Robot Innovation Center's Tiangong Ultra ran a 9.39-second 100-meter dash — faster than the 9.58-second world record Usain Bolt set in Berlin in 2009. Both events were refereed and broadcast live, with timing and rally counts verified by independent technical officials.

The Tennis Match

Galbot ET1, a 1.65-meter humanoid with 28 degrees of freedom, played a regulation tennis match against Zheng Jie, who retired from the WTA tour in 2024 after 12 seasons. The robot completed 103 successful rallies before a service-return error ended the longest continuous sequence, according to PRNewswire reporting on August 23. Each rally required real-time ball trajectory prediction, footwork adjustment and racket swing timing under the same physical conditions a human athlete would face. Galaxy General Robotics said the demo was a stepping stone to "AstraTennis," a consumer robotics platform the company plans to launch in 2027 at a sub-$5,000 price point. PRNewswire quoted the team as saying the system ran on a single onboard NVIDIA Jetson Thor module with no cloud connectivity.

The 100-Meter Sprint

Tiangong Ultra, developed by the Beijing Humanoid Robot Innovation Center — a state-backed consortium that includes UBTech, Xiaomi and the Beijing Institute of Technology — finished the 100-meter preliminary race in 9.39 seconds, eclipsing the previous Games record set at the inaugural event in 2025. The robot stands 1.78 meters, weighs 52 kilograms and uses a hydraulic actuation system optimized for sprint gait. Engineers told the Global Times the timing was independently verified by World Athletics-certified timing equipment, with the robot starting from a real starting block and running in an outdoor stadium. The performance was revised from 38.16 to 38.15 in one secondary event, the 400-meter, where Tiangong Ultra took third.

Why the Two Stories Matter Together

Sports is a useful proxy for embodied AI because it demands both balance and reactivity. A robot that can return a 200 kph tennis serve must combine visual perception, motor control and decision-making in milliseconds; a robot that can sprint 100 meters in under 10 seconds must combine battery density, actuator torque and gait stability at the limits of mechanical engineering. The Galbot and Tiangong Ultra results suggest that Chinese labs have closed the gap with US peers in both dimensions — a shift with implications for the commercial humanoid roadmap that BYD, Tesla Optimus and Unitree are racing to define.

What's Actually New in 2026

Three advances made both performances possible. First, end-to-end neural policies trained in NVIDIA Isaac Lab replaced the cascaded perception-planning-control stacks that powered earlier-generation humanoids; the new policies collapse the typical 50-millisecond pipeline to under 5 milliseconds. Second, a new generation of frameless permanent-magnet motors from Xiaomi and UBTech deliver torque densities above 25 Nm/kg, a 3x improvement over the servos used in the 2025 games. Third, on-board inference hardware from NVIDIA — specifically the Jetson Thor module shipping in production since April — provides 40 TOPS of INT8 compute within a 25-watt power envelope, enough to run the full policy without offloading.

What to Watch Through Year-End

Three checkpoints follow. The semi-finals and finals on August 24 and 25 will reveal whether the 9.39-second time was a peak performance or a typical result; Tiangong Ultra's coach told reporters the robot has hit 9.31 seconds in training. The AstraTennis commercial launch at CES 2027, if Galaxy General Robotics delivers on its sub-$5,000 promise, will be the first major test of whether a humanoid robotics capability demo translates to consumer demand. And Unitree's post-IPO roadmap — the company closed its Shanghai listing on August 19 at a $61 billion valuation — will reveal whether the world's first publicly listed humanoid maker can match the Galbot and Tiangong Ultra performance benchmarks in commercial units by year-end.