The AI sped up through matches to learn years' worth of knowledge in no time.
Nowadays, AI is capable of doing many things, and playing soccer is now one of them. DeepMind managed to teach digital people how to play in a computer simulation through an athletic curriculum.
As reported by NewScientist, the first phase of the curriculum trained the players to run naturally by imitating motion-capture videos of humans playing soccer. Then they practiced dribbling and shooting the ball through a form of trial-and-error machine learning.
These steps represented about 1.5 years of simulation training time, which the AI managed to cover in about 24 hours. More complex behaviors appeared after five simulated years of soccer matches.
“They learned coordination, but also they learned movement skills that we didn’t have explicitly set as training drills before,” said Nicolas Heess at DeepMind.
The third phase made the digital players score goals in two-on-two matches. This stimulated teamwork skills, which emerged throughout about 20 to 30 simulated years of matches, or two to three weeks in the real world. "This led to demonstrated improvements in the digital humanoids’ off-ball scoring opportunity ratings, a real-world measure of how often a player gets in a favourable position on the pitch."
This research can be used to create a team of robotic soccer players, and that's what DeepMind is working on. Maybe we will be watching robot sports in the future if experiments in this field continue.
If you'd like to learn more, read the From Motor Control to Team Play in Simulated Humanoid Football paper here. Also, don't forget to join our 80 Level Talent platform and our Telegram channel, follow us on Instagram and Twitter, where we share breakdowns, the latest news, awesome artworks, and more.