What Beijing’s Humanoid Robot Games Revealed About the Future of Robots

The second World Humanoid Robot Games turned Beijing’s National Speed Skating Oval into a large-scale test of what humanoid robots can—and cannot yet—do. The machines sprinted, played football and table tennis, attempted strength events and performed work-like tasks. But the most revealing moment came on a tennis court: Galbot’s humanoid could track incoming shots, reposition itself and return the ball, then lose its balance during a live exchange.

That contrast is the clearest takeaway from the event. Humanoids are becoming faster and more coordinated in carefully defined demonstrations, while reliable recovery, long-duration operation and independent performance in unpredictable environments remain much harder problems.

The 2026 games at a glance

The five-day event took place from August 22 to August 26, 2026, at Beijing’s National Speed Skating Oval, widely known as the Ice Ribbon. Organizers reported 2,056 robots, 666 teams and participants from 16 countries and regions. The program included 1,301 competition sessions across 51 events. 23

The scale was substantially larger than the inaugural 2025 edition, according to reporting ahead of the games. The expanded format was designed to test both athletic performance and practical robotic capability rather than simply produce exhibition footage. 3

From sprinting and football to warehouse work

The event combined sports modeled on human competition with scenario-based challenges modeled on work. Athletic events included track disciplines, football, tennis, table tennis and newly added challenges such as long jump, weightlifting and tug-of-war. These events were intended to test movement control, structural design and core robotic components. 13

The scenario-based program moved closer to industrial and service applications. Reported settings included simulated factories, restaurants, offices, warehouses and emergency-response environments. Tasks included connecting cables and handling warehouse operations—activities that demand precise manipulation as well as movement.

That distinction matters. A robot can be highly capable at a rehearsed motion or a narrow athletic task without being ready to work independently in a changing environment. Organizers and partners said more than 40% of the events required robots to operate fully autonomously, making the games a test of decision-making and execution as well as mechanical performance.

Galbot’s tennis demonstration: impressive perception, imperfect balance

During the opening demonstrations, a humanoid developed by Galbot played against or alongside human tennis players, including former champion Zheng Jie. It crouched, shuffled laterally, adjusted its position and swung its racket to return shots. Reports described the robot sensing incoming balls, predicting their trajectories and responding without preset scripts or human control.

Galbot said its robots completed more than 100 consecutive autonomous rallies with Zheng Jie. That is a notable company-reported result, but the claim is not the same as an independently verified benchmark: the rally count, the precise autonomy conditions and the description of the performance rest largely on company or event reporting.

The demonstration also supplied its own caveat. During a fast exchange, the robot chased a ball, lost its balance and toppled backward. The fall did not erase what the system had accomplished, but it showed why tennis is such a demanding test: successful play requires continuous perception, foot placement, racket control, momentum management and recovery when the plan goes wrong.

In other words, the robot could execute impressive sequences while still struggling with the less glamorous but essential question of what happens after an unexpected movement or miscalculation.

What the games revealed about humanoid robots

Athletic skill is not the same as general autonomy

The games showed that humanoids can perform specialized physical behaviors at increasingly impressive speeds. They also showed the limits of judging progress from a highlight clip. Many real-world jobs involve changing layouts, unfamiliar objects, interruptions and errors—conditions that are much less predictable than a marked court or a defined competition lane.

The scenario events were valuable precisely because they introduced those practical demands. A robot connecting a cable or completing a warehouse task must identify the target, plan a sequence, control its body and adapt when the object or environment is not exactly where expected.

Balance and recovery remain central engineering problems

The tennis fall was a visible example of a broader challenge: dynamic balance is not just the ability to stand or walk, but the ability to stop, turn, absorb an error and recover without collapsing. Fast movement increases the consequences of mistakes, particularly when a humanoid must coordinate its legs, torso and arms at the same time.

Battery life limits the workday

An engineer cited operating times of roughly five to six hours for a robot at the event. That may be adequate for demonstrations or scheduled task blocks, but it creates a practical constraint for continuous deployment: systems may need charging breaks, battery swaps or other operating arrangements before they can cover a full work shift.

Connectivity and computing affect responsiveness

The event also highlighted the importance of low-latency connectivity and computing infrastructure. Faster networks can support the rapid exchange of sensor data, coordination and cloud-assisted processing, but network dependence introduces its own operational questions. For tasks requiring immediate reactions, a robot must remain responsive even when conditions are noisy, crowded or less controlled than a demonstration environment.

The bigger meaning of Beijing’s robot competition

The second games were neither proof that humanoids are ready to replace human workers nor merely a publicity spectacle. They functioned as a public stress test: sports made speed and coordination easy to see, while scenario-based tasks exposed the gap between performing a narrow skill and operating independently in the real world.

Beijing’s results point to rapid progress in perception, movement and specialized physical intelligence. They also point to the remaining deployment hurdles—stable whole-body control, recovery from mistakes, dependable autonomy, energy endurance and the infrastructure needed for real-time operation. The most important question is no longer whether a humanoid can produce an extraordinary moment. It is whether it can repeat useful work safely, independently and for long enough to matter.