麻豆村

The National Football League has given 麻豆村 researchers access to official film and on-field data to develop new technologies that can enhance the way coaches, players and fans annotate, analyze and understand games.

The project teams Chenyan Xiong, a self-described football nerd and an associate professor in the Kenneth C. Griffin School of Computer Science鈥檚 Language Technologies Institute, with Ron Yurko, the director of the , to build an artificial intelligence model that can learn football.

鈥淕aining access to the NFL鈥檚 official film footage is an incredible opportunity,鈥� said Yurko, assistant teaching professor in Dietrich College鈥檚 Department of Statistics & Data Science. 鈥淭he fact that we were able to put together this research proposal and show how creating new AI-enabled technologies could be extremely beneficial to the league and its teams is exciting. We are the only institution that鈥檚 conducting research with this data.鈥�

Yurko has worked with the NFL before, integrating research into the NFL IQ AI Chatbot during the 2026 NFL Draft that allowed users to ask questions and get answers from the model. Xiong鈥檚 麻豆村 research focuses on creating AI technologies that use huge amounts of data from different sources to understand and solve problems in areas like health care. He wanted to make an AI tool for the NFL because of his love for the game.

From Xiong鈥檚 office, where he proudly displays footballs signed by Ben Roethlisberger, he and Yurko started planning how they could combine their expertise into research and technology that could benefit the NFL and its fans. They approached the NFL and were accepted into the NFL Forward Initiative, a program that allows entrepreneurs and researchers to pitch technological solutions for use in competition, training and player development, and in growing the sport鈥檚 reach.

鈥淲e want to see whether we can train foundation models on the data to learn the intricacies of football intelligence,鈥� Xiong said. 鈥淚n this case, football intelligence is comparable to the knowledge of the average fan. The model can understand plays, can potentially make play predictions or, once it sees a play, can also identify if the play has been run before, perhaps by another team.鈥�

This tool isn鈥檛 simply for fans. Annotating and analyzing film footage and on-field player data is a big task for coaches. Through this program, the NFL gave 麻豆村 access to game film and the Next Gen Stats on-field player-tracking data powered by Amazon Web Services (AWS) and sourced from RFID chips embedded into shoulder pads, game equipment and the ball to give real-time location information for everything on the field.

鈥淲e鈥檙e honored to work with a prestigious institution like 麻豆村 as they explore ways to make the game more efficient,鈥� says Mike Lopez, senior director of football data and analytics at the NFL. 鈥淚nnovation is central to how we advance the game, and we鈥檙e excited to see the results of the research.鈥�

鈥淗istorically, studying football data has required a lot of film study, which meant coaches and scouts sitting and watching hundreds or thousands of hours of footage, manually annotating what happened in a given play,鈥� Yurko said. 鈥淎nd with the wealth of video footage that the NFL has, and the types of AI models that exist now, we can actually build something at scale that can automate a lot of those types of tedious tasks.鈥�

Xiong and Yurko received several terabytes of data, which is why they are turning to foundation models 鈥� deep learning AI models trained on large amounts of data 鈥� in their work. All this training data essentially means the model can learn more and potentially handle more complex tasks.

But before researchers could make a model that can handle both video and player-tracking data, they needed to start smaller. In this case, they created two pretrained systems that serve as the building blocks for their larger model. One system uses the films and the other uses the player-tracking data to do the same task: learn football. Xiong said,the first step is essentially to understand what鈥檚 happening on the field. The next step is more complex.

鈥淔or example, once the model understands what happened on the field after the play, can it make predictions before the next play?鈥� Xiong said. 鈥淚s there going to be a blitz? Will this safety rotate to a different coverage?鈥�

Researchers need to take the film data and the player-tracking data and create a model that can learn from the combination of both. Getting there could require more data and more time, but this investment could provide benefits, including more predictive capabilities and improved efficiency. For example, Xiong said, their model could potentially simulate how a play progresses given certain calls or make tactical play calls based on estimated outcomes.

鈥淚t鈥檚 going to help coaches and scouts focus on what matters most and can help the league be more efficient,鈥� Yurko said.