Not everything fails in the world of robots, while some videos show machines losing control, others reveal something completely different, a level of precision that begins to impress. Researchers recently demonstrated a human robot was capable of playing tennis in real time with a human and we're not talking about simple movements.

The robot manages to react to fast balls, position itself on the court and return hits with consistency. The demonstration was carried out using a system called Latent, developed in collaboration with Galbot Robotics and researchers from the University of Xuais and Peking. The system was implemented in the Unitree G1 human robot, the same one that scared the lady and you can see it in one of my previous posts.

In simulation tests, the robot achieved up to 96% accuracy in forehand shots, but the most impressive thing is not only the number, but the challenge behind it, because tennis is one of the most difficult sports for a robot, the balls can reach speeds of up to 30 m per second, the contact time between the racket and the ball lasts just a few milliseconds and the player needs to constantly move over large areas.

Instead of capturing entire matches, the researchers divided the learning into small fragments of movement, forehand blows, left blows and changes of direction, this data was collected in a small space and then combined to form more complex sequences, as if the robot learned individual pieces before putting together the complete game, in addition, the training was carried out in a simulation environment where variables such as weight, friction and aerodynamics were constantly modified.

This helped the system to better adapt to real-world conditions, a robot capable of maintaining ball exchanges with a human in a continuous and dynamic way, yes, it was not with a professional player, but between us. Impressive, right? The researchers claim that this methodology can be applied to other sports such as football, badminton, even tasks outside of sports, where capturing complete human movement data is difficult.