Many people have been inspired by the football World Cup in France and now artificial intelligence is learning to play too.
Karol Kurach and colleagues at Google Research in Zurich, Switzerland have made a virtual football training pitch聽for AIs to use to learn how to play.
Because football requires a balance between short-term control and high-level strategy, it is challenging for AIs to master, says Kurach.
Advertisement
The goal is to develop AIs that can better interact with their environment and solve complex tasks, including in robots and self-driving cars.
The training pitch simulates a full standard football game, with features including goals, corner kicks, offsides and penalty kicks. An AI controls one or all 11 players on a team and tries to defeat another AI opponent.
Google has tested their own AI on it and another from research company DeepMind, which it owns, but anyone can download the program to train their own AIs.
To master football, an AI must learn to deal with unpredictability 鈥� for example, when a player kicks the ball, it may land in different locations or could be intercepted by the opposing team. 鈥淯nlike games such as chess or Go, there is not a set model of moves,鈥� says Kurach.
The AIs are provided with two inputs: what the gameplay looks like visually at a point in time, including the score, and information on the position and speed of players.
The approach differs from the computer opponents used in games such as FIFA, which are manually programmed by game designers to use specific rules and strategies. 鈥淪uch bots can do only what they were programmed for, and always follow the same pattern,鈥� says Kurach.
Google鈥檚 training pitch also has a football academy option, which provides specific scenarios of varying difficulty. 鈥淓xamples of the academy scenarios include settings where agents have to learn how to score against an empty goal, how to run towards a keeper, how to quickly pass the ball between players, or how to execute a fast counterattack,鈥� says Kurach.
