Scientists at Google DeepMind trained the Robotis OP3 robot to play soccer using an AI training technique called "deep reinforcement learning."
This machine learning training technique is the result of combining different AI training methods. The algorithm works like artificial neurons, and is arranged similarly to the human brain.
Compared to robots not trained with AI, robots trained with this technique moved 181% faster, turned 302% faster, kicked the ball 34% faster, and took 63% less time to recover from falls in simulated matches.

From the video provided by scientists, it can be seen that after being trained by AI, robots can perform complex movements such as dribbling, blocking, passing, scoring... relatively proficiently and at high speed.
The researchers added that the robot has developed extremely difficult to program emergency movements in football, such as spinning around to fool opponents, turning the angle of the foot...
The findings suggest that this AI training technique can help humanoid robots in general create simple, but relatively safe movements. From there, robots can be advanced with more complex movements, and applied to more real-life situations.
Similar technology was applied and successfully enabled the robot to make coffee after a training period of only 10 hours.