Home / Technology / Robot serves ace: AI masters tennis in real-time
Robot serves ace: AI masters tennis in real-time
23 Mar
Summary
- Humanoid robot plays tennis, reacting to shots in real time.
- AI learned complex movements from small motion data segments.
- This AI advancement applies to many skills beyond tennis.
A 4-foot-tall humanoid robot is now capable of playing tennis, reacting to shots and rallies in real time without pre-programmed scripts or remote control. This marks a significant advancement, as the robot, powered by LATENT on the Unitree G1, can track fast-moving balls, adjust its position, and return shots with accuracy.
Researchers developed this AI by focusing on small motion data segments rather than full gameplay, gathering about five hours of data from five players on a compact court. The system learns individual movements and combines them into coordinated sequences, demonstrating millisecond-level reactions and full-body coordination.
Simulated training with varied physical conditions helped the robot adapt to real-world unpredictability, achieving up to 96% success on forehand shots in simulation. The system can sustain rallies with human opponents, and its ability to place shots suggests early forms of decision-making, although its motion is not yet as fluid as a trained athlete.
This AI breakthrough has implications far beyond tennis, demonstrating that robots can learn complex human skills without perfect data. This approach could be applied to various tasks lacking complete motion data, indicating a future where robots can train with or challenge professional athletes.
