A tennis ball flew across the court at Beijing's National Speed Skating Oval on August 22. On the other side of the net, a humanoid robot moved into position, adjusted its steps and swung its racket to return the shot.

The match was part of the opening ceremony of the second World Humanoid Robot Games, where a humanoid robot developed by Chinese robotics company Galaxy General Robotics played against human athletes, demonstrating autonomous movement and real-time interaction on court.

The event highlighted a major challenge for artificial intelligence: moving beyond digital environments and enabling machines to perceive, make decisions and act in the physical world.

Unlike games played on a fixed digital board, sports such as tennis require robots to respond to constantly changing conditions. A robot must track a fast-moving ball, predict its trajectory, maintain balance while moving and coordinate its entire body to complete a precise shot.

During the match, the robot independently judged ball paths in singles play. In doubles, it worked with human teammates, adjusting its strategy based on the situation on court and responding to changes during fast-paced exchanges.

The performance showed why tennis has become a demanding test for embodied intelligence, a field focused on enabling AI systems to interact with the physical world through robotic bodies.

For robots, the challenge is not only understanding what to do but also executing actions in real time.

According to Galaxy General Robotics, the system behind the demonstration combines high-level decision-making with low-level motion control. Its "brain" analyzes the game situation and determines how to respond, while its "cerebellum" coordinates movements across the body, including walking, balance and racket control.

The company said it trained the robot using a combination of human movement data and virtual simulations. In virtual environments, multiple robotic agents could practice against each other under different conditions, helping the system improve skills such as continuous rallies, positioning and shot selection.

The approach reflects a broader shift in AI development. A decade ago, systems such as AlphaGo captured global attention by defeating top human players in board games, demonstrating AI's ability to reason within clearly defined digital environments. By contrast, embodied AI must combine autonomous movement, situational analysis and real-time response in unpredictable physical settings. In a real-world match, every shot requires the robot to assess changing conditions and react immediately rather than follow a fixed sequence of movements.

While humanoid robots still have room to improve compared with professional athletes, the tennis match demonstrated their growing ability to handle longer and more complex tasks in dynamic environments. For researchers developing embodied AI, such tests provide a way to evaluate whether robots can turn perception and decision-making into reliable physical actions – a key challenge in developing the next generation of intelligent machines.