Machine learning humanoids learn to play soccer from scratch

Simulated “dummies” have been taught to play football (soccer) by an international team of machine learning scientists. Yes, you read that right.

Sidenote: I insist on using the term “football” instead of “soccer”.

The results, including endearing videos of the AI ​​models putting their new skills into practice, are published in the newspaper Scientific robotics. Do yourself a favor and watch these videos. If you think the screenshot above is the visual representation of chaos, you’ll love the movies.

But before the AI ​​grudge match of the century, some basic training was needed. To get three-dimensional artificial intelligence humanoids to play football, they first had to be trained in rudimentary locomotion.

The team of computer scientists led by Siqi Liu are based at UK-based artificial intelligence and computer programming company DeepMind, which is owned by Google’s Alphabet Inc.

In their paper, the researchers describe a three-step machine learning framework designed to teach virtual AI humanoids a wide range of basic skills and motor functions.

Video demonstrating the machine learning study. Credit: Liu et al., Sci. Robot. 7, eabo0235.

First, the AI ​​mannequins watched videos of humans in motion to learn by imitation to mimic low-level natural movements such as walking.

Then, a reward-based reinforcement learning approach taught the AI ​​mid-level skills like kicking a ball.

Finally, they learned high-level skills, including teamwork and body control, by working as a team in an advanced version of reward-based reinforcement learning. Also known as a “match”.


Read more: Artificial intelligence tool springs into action, discovering alternative physical variables!


No prior knowledge of football, its rules or its goals has been fed into the machine learning algorithm. Still, the virtual players learned all of these skills and played a hard-fought game against each other, showing cooperation, body control, and decent technique. And, as an occasional futsal administrator for six years, I should know that.

Much like taking a football to a city that loves rugby or Australia rules, the researchers didn’t provide any context to the game itself, but the AI ​​was able to learn how to play football through the machine learning. Goal kicks indeed.

“The result is a team of coordinated humanoid soccer players who exhibit complex behavior at different scales, quantified by a range of analyzes and statistics, including those used in real-world sports analysis,” write the authors. authors.

Some of the familiar football skills the virtual players mastered included jostling for position, clearing the ball away from the goal they were defending, quick turns, passing the ball through, kicking from different heights, slowing down before changing direction with the ball, tackling and running into a defensive position.

Lionel Messi and Cristiano Ronaldo aren’t, but these mock soccer stars would have gone straight into my high school soccer team.

Besides giving a hilarious reprieve and emulating our proud A-League, what do the Seekers hope to accomplish?

“[The simulated players] achieved integrated control in a framework where motor skills and high-functioning goal-directed behavior were intertwined – a framework that reflects many of the challenges that animals and humans face and where solutions would be extremely difficult to devise by hand,” the authors write.

They believe that research aimed at teaching AI how to move and interact like animals or humans can help in the development of more animal-like robots.

“Enabling machines to produce agile, animal-like movements has been one of the goals of robotics research,” they write. “To date, these techniques appear rather distinct from the learning-based solutions developed in the simulation AI community. Yet recent partial successes in transferring simulation results to real robots suggest that simulation-based learning approaches may, in the future, play a greater role in controlling real-world robots.

With the World Cup just a few months away, watching these clumsy AI dummies certainly got me into the footballing mind.

Sherry J. Basler