How the researchers taught the robot to play tennis

The team condensed five hours of real tennis moves into a simplified form that the robot could learn from.
They used simulations with different ball speeds and conditions so the robot could handle real-life games, even when things got unpredictable.

The robot nailed 96% of forehand shots in simulation tests

The robot nailed 96% of forehand shots in simulation tests and kept up with humans during rallies, reacting in milliseconds.
The best part? This approach could help robots learn other sports too, like football or badminton, especially when perfect training data isn't available.
If you're curious or want to build on their work, it's all open source on GitHub.