Humanoid Robot Demonstrates Tennis Skills with Human Players

On March 16, 2026, Galbot Robotics unveiled a video showcasing their Unitree G1 humanoid robot engaging in real-time tennis rallies with a human player. This demonstration highlights the capabilities of the LATENT system (Learning Athletic Humanoid Tennis Skills from Imperfect Human Motion Data), developed in collaboration with researchers from Tsinghua University and Peking University. The LATENT system enables the robot to respond to fast-moving tennis balls, navigate the court, and sustain rallies, marking a significant advancement in robotic athletic interaction.

Interesting EngineeringNew York Post

The development of the LATENT system faced challenges due to the scarcity of accurate human movement data in sports like tennis, where players must cover large areas and react to high-speed balls. To overcome this, researchers collected short fragments of essential movements—such as forehands, backhands, and footwork—rather than full match recordings. This data was gathered using a motion-tracking system on a compact court, significantly smaller than standard tennis courts. Five players contributed approximately five hours of recorded motion data, which the system utilized to train the robot in both individual movements and coordinated gameplay.

Interesting Engineering

In simulation tests, the LATENT system achieved a success rate of up to 96% in executing forehand shots. When tested on the Unitree G1 robot, it demonstrated the ability to maintain rallies with a human opponent and consistently return balls to the opponent's side of the court. The robot can react to balls traveling at speeds exceeding 15 meters per second (approximately 33.5 miles per hour), showcasing its capacity for high-dynamic athletic interaction.

Interesting Engineering

The researchers noted that the framework developed for tennis could be applied to other sports and tasks where capturing complete human motion data is challenging, such as football and badminton. They emphasized that the ability to learn complex physical skills from imperfect data suggests potential applications in various real-world scenarios, including household tasks like laundry folding and service activities.

New York Post

Galbot Robotics stated, “For the first time, a humanoid robot can sustain high-dynamic, long-horizon tennis rallies with millisecond-level reactions, precise ball striking, and natural whole-body motion.” The researchers added that their approach, despite relying on imperfect data, provides valuable insights into human primitive skills in tennis scenarios.

Interesting EngineeringNew York Post

While the demonstration has been met with enthusiasm, some experts caution that the robot's movements, although improved, still lack the fluidity of human players. Critics argue that while the technology shows promise, it may not yet be ready for competitive play against elite human athletes.

New York Post

• “For the first time, a humanoid robot can sustain high-dynamic, long-horizon tennis rallies with millisecond-level reactions, precise ball striking, and natural whole-body motion.” — Galbot Robotics

New York Post

• “Our key insight is that, despite being imperfect, such quasi-realistic data still provide priors about human primitive skills in tennis scenarios,” — Researchers from Tsinghua University and Peking University

Interesting Engineering

• “Although this work primarily focuses on the tennis return task, the proposed framework has the potential to generalize to a broader range of tasks where complete and high-quality human motion data are unavailable,” — Researchers

Interesting Engineering

This innovative development in humanoid robotics not only showcases the potential for athletic interaction but also opens avenues for future applications across various domains.

New York Post