Customarily, characters and their movement are crafted manually in video games. Sports games, akin to FIFA, make use of a motion capture – the process of recording the movement of objects or people. However, the possibilities are restricted by the movement that’s been recorded, and code nonetheless must be written to animate the character.
Through automation of the animation process, along with other elements of game design and development, AI could save firms that specialise in games million of dollars while making games more efficient and appear more authentic. This is to ensure a complex game can run on a smartphone, for example.
Reinforcement learning has ignited excitement in recent years by allowing computers to learn in order to play complex games and solve vexing problems without any instructions. In 2013, researchers at DeepMind used reinforcement learning to design a computer programme that learned to play several Atari video games to a superhuman level. The program learned to play through experimentation and feedback from pixels and the game score. The same technique was later employed to construct a programme that accomplished the fiendishly complex and subtle board game Go, amongst other things.
To make the character, a machine-learning model was first trained to identify and reproduce statistical patterns in motion-capture data. Reinforcement learning was then used to train another model to reproduce realistic motion with a specific objective, such as passing a ball to another player. This primarily means the program learns how the soccer player moves and can then animate the moves accordingly.