AI robots learn how to play football
What AI Robots do?
The about knee-height small humanoid robots were adept at making tackles, scoring goals, and quickly rising to their feet after being knocked to the ground.
How researchers working on it ?
AI researchers initially trained virtual versions of the robots in decades’ worth of match simulations using DeepMind’s cutting-edge MuJoCo physics engine to learn how to play. If the simulated robots’ actions boosted performance, such as taking the ball away from an opponent or scoring a goal, they were rewarded.
Why agents are the best for Robot’s basic strategic understanding?
The agents also gained a fundamental comprehension of the game’s strategy, learning, for example, how to predict ball motions and deflect opponent strokes. Despite the robots’ inherent fragility, small hardware adjustments and basic behaviour regularisation during training allowed the robots to learn secure moves while still performing in a quick and nimble manner.
Previous DeepMind football-playing AI research has employed various team configurations, increasing the number of players to train simulated humanoids how to function as a team. According to the researchers, this approach will not only improve AI system coordination but also open up new avenues for creating AGI that is on par with or perhaps beyond that of humans.