Google DeepMind's new AI model could help soccer teams take the perfect corner


Working with player-tracking data from 7,176 corners taken in the Premier League during 2020 and 2021, the researchers began to represent the players' arrangements as a graph, plotting the players' positions, gaits, height and weight. Encoded as nodes on the graph. , and the relationships between players as lines between them. They then used an approach called geometric deep learning, which takes advantage of the symmetry of the football field to reduce the amount of processing required by the neural network. (This isn't a new strategy – a similar approach was used in DeepMind's influential AlphaGo research.)

The resulting model led to the creation of several tools that may be useful to football coaches. Based on the arrangement of the players when the kick is taken, TacticAI can predict which player is most likely to make first contact with the ball, and whether a shot will be taken as a result. It can then generate recommendations for the best ways to adjust player positions and movements to either maximize the chance of taking a shot (for the attacking team) or minimize it (for the defending team. For) – moving a defender to cover the near post, for example, or placing a man on the edge of the field.

Velicovic says Liverpool's football experts particularly liked how TacticAI's recommendations could identify attackers who were crucial to the success of a particular tactic, or defenders who were “asleep at the wheel”. Analysts spend hours sifting through video footage looking for weak points in their opponents' defensive setup that they can target, or trying to find loopholes in their team's performance to double down in training. “But in different situations, it's really hard to track 22 people,” Veliskovic says. “If you have a tool like this it immediately helps you see which players are not moving correctly, which players need to do something different.”

TacticAI can also be used to find other corners that have similar patterns of players and activities, saving hours of time for analysts. According to DeepMind, the suggestions provided by the model were considered twice as useful by Liverpool coaches as current techniques, which are based only on players' physical coordinates and do not take into account their movement or physical characteristics. (The two corners may look alike, but if the tall striker is on the edge of the box in one and running towards the near post in the other, this is probably significant.)

One thing it's also doing, according to DeepMind's Zhe Wang, another major contributor to the paper, is overcoming the lack of a suitable language to describe the vast range of different things happening in a corner. Unlike American football, which has a deep and historical nomenclature for various plays and running routes, the choreography of football set pieces in such detail is a relatively new phenomenon. “Different coaches may have their own expression for the corner kick pattern,” says Wang. “So with TacticAI, we hope to harness the power of deep learning to establish a common language for describing the patterns of corner kicks.”

According to the paper, in the future the researchers hope to build TacticAI into a natural language interface so that coaches can query it in text and get answers to the problems they are trying to solve on the field. Veliskovic says this model can be used during games to help coaches improve their corner routines, but it is most likely to be useful in the days before a match, where it helps coaches Will free up time. “We don't want to create AI systems that replace experts,” says Veliskovic. “We want to create AI systems that enhance the capabilities of experts so they can do their work more efficiently and have more time for the creative part of coaching.”