Tacticai: AI assistant for football tactics

As part of our many years of cooperation with Liverpool FC, we develop a full AI system, which can advise buses after corners

“Corner made quickly … Origi!”

Liverpool FC returned historically in the semi-finals of the UEFA Champions League 2019. One of the most iconic moments was throwing the horn of Trent Alexander-Aarnold, who set up in Divock Origi in to get what went out in history as Liverpool FC's greatest goal.

Corner kicks have great goals potential, but the development of routine is to combine human intuition and game design to identify patterns in competitive teams and react in flight.

Today, in Nature communicationWe introduce tactics: the artificial intelligence system (AI), which can provide experts with tactical observations, especially on the corner, through predictive and generative artificial intelligence. Despite the limited availability of standard data in corner throws, Tacticai achieves the latest results, using a geometric approach to deep learning to help create more generalized models.

We developed and evaluated tactics with experts from Liverpool Football Club as part of many years of research cooperation. Tacticai suggestions were preferred by human expert rescuers in 90% of cases from tactical configurations observed in practice.

Tacticai shows the potential for supporting AI techniques to revolutionize sport for players, coaches and fans. Sport like football is also a dynamic field for developing artificial intelligence, because they contain real interactions with many agents, with multimodal data. AI procedure for sport can translate into many areas on the pitch and outside – from computer games and robotics to traffic coordination.

Development of a game plan with Liverpool FC

Five years ago, we started working with Liverpool FC for many years to develop AI for sports analysis.

Our first article, the Game plan, checked why AI should be used in the help of football tactics, emphasizing examples such as analysis of criminal kicks. In 2022, we developed Chartwho showed how AI can be used with the prototype of the predictive system to down in football analysis. The system can predict players' movements outside the camera when the tracking data was not available-otherwise the club would have to send scouts to personally watch the game.

We have now developed tactics as a full AI system with connected predictive and generative models. Our system allows trainers to try alternative players' configurations for each routine of interest, and then directly assess the possible results of such alternatives.

Tacticai is built to answer three basic questions:

1. What happens in the case of a given tactical configuration of the corner kick? For example, who will probably receive the ball and will there be a shot attempt?
2. Can we understand what happened after restoring the configuration? For example, did similar tactics work well in the past?
3. How can we adjust the tactics to make a specific result? For example, how should defenders be changed to reduce the probability of shot attempts?

Predicting the results of a corner kick with geometric deep learning

A corner kick is awarded when the ball passes over the line line after touching the player defending the team. Anticipating the results of corner kicks is complex due to the randomness of the game of individual players and the dynamics between them. It is also difficult to model artificial intelligence due to available data with limited golden-only-only projects about 10 corner throws are played in every match in the Premier League every season.

Tacticai successfully predicts a corner game using a geometric approach to deep learning. First of all, we directly model the closed relationship between players, representing the corner kick configurations as charts in which the nodes represent players (with functions such as position, speed, height, etc.), and the edges represent relationships between them. Then we use the approximate symmetry of the football field. Our geometric architecture is a variant Equivariant Convolutional Network Group This generates all four possible reflections of a given situation (original, fastened with H, from V, saved in HV) and forces our forecasts regarding receivers and shot attempts to be identical to all four. This approach reduces the search space of possible functions that our neural network can represent for those that respect the symmetry of reflection – and gives more generalized models, with smaller training data.

Providing constructive suggestions for human experts

Using their predictive and generative models, Tacticai can help trainers, finding similar corner throws and testing various tactics.

Traditionally, to develop tactics and counterattack, analysts would again watch many games from games to look for similar examples and study competing bands. Tacticai automatically numerical faces of player representations, which allows experts to easily and efficiently search for appropriate previous procedures. In addition, we approved this intuitive observation through extensive qualitative research with football experts, who stated that the highest 1 recovery of Taticai was significant in 63% of time, almost twice twice as much 33% of reference points in approaches that suggest couples based on direct analysis of the player's position.

The Tacticai generative model also allows people to redesign the corner dump tactics to optimize the likelihood of some results, such as reducing the likelihood of a shot attempt to arrow for defense configuration. Tacticai provides tactical recommendations that adapt the positions of all players in a specific team. Based on these proposed corrections, coaches can faster to identify important patterns, as well as key players for success or defeat of tactics.

In our quantitative analysis, we showed that Tacticai was accurate in predicting the corner kick receivers and the situation of the shot, and the repositioning of players was similar to how real games were created. We also assessed these recommendations qualitatively in a blind case, in which the evaluators did not know which tactics came from a real game and which were tactikai. Liverpool FC football experts said that our suggestions cannot be distinguished from real corners and were favored in their original situations in 90% of cases. This shows that tactics forecasts are not only accurate, but useful and possible to arrange.

Artificial intelligence procedure for sport

Tacticai is a full AI system that can give trainers immediate, extensive and accurate tactical view – which are also practical in the area. Thanks to Tacticai, we developed a talented AI assistant in the field of football tactics and achieved a milestone in developing useful assistants in AI Sports. We hope that future research can help develop assistants who are expanding to more multimodal input data outside the player data, and help experts in many ways.

We show how AI can be used in football, but football can also teach us a lot about artificial intelligence. It is a very dynamic and difficult game for analysis, with many human factors, from body structure to psychology. Detection of all patterns is difficult even for experts, such as experienced trainers. Thanks to Tacticai, we hope to take many lessons to develop wider supporting technologies that combine human knowledge and analysis of artificial intelligence to help people in the real world.