11-19 更新 实时
首页/ 算力/ 正文

研究人员教 AI 模型打橄榄球

算力 来源:TechXplore 2026-11-06 10:40 0 阅读 机器翻译
研究人员教 AI 模型打橄榄球

作者:Marylee Williams、Jason Bittel,卡内基梅隆大学

编辑:Gaby Clark,审校:Andrew Zinin

本文已按照 Science X 的编辑流程与政策进行审核。编辑在确保内容可信度的同时,强调了以下属性:

添加到首选来源

卡内基梅隆大学(CMU)的研究人员希望打造一款 AI 工具,能够预测即将进行的战术、模拟结果,并改善专业人士与球迷理解橄榄球比赛的方式。图片来源:卡内基梅隆大学

NFL 已向卡内基梅隆大学的研究人员开放官方比赛影片与场上数据,用于开发新技术,以改善教练、球员和球迷对比赛的标注、分析和理解方式。

该项目将自称橄榄球迷的 Chenyan Xiong 与卡内基梅隆体育分析中心(CMSAC)主任 Ron Yurko 组队。Chenyan Xiong 是 Kenneth C. Griffin 计算机科学学院语言技术研究所的副教授。两人将共同构建一个能够学习橄榄球的人工智能模型。

“能够获得 NFL 官方比赛影片的访问权限是一个难得的机会,”Yurko 说。他是 Dietrich 人文与社会科学学院统计与数据科学系的助理教学教授。“我们能够整理出这份研究提案,并展示创建新的 AI 技术如何能为联盟及其球队带来极大益处,这令人兴奋。我们是唯一利用这些数据开展研究的机构。”

Yurko 此前曾与 NFL 合作,在 2026 年 NFL 选秀期间将研究成果整合进 NFL IQ AI Chatbot,让用户能够向模型提问并获得回答。Xiong 在 CMU 的研究专注于打造 AI 技术,利用来自不同来源的海量数据来理解和解决医疗健康等领域的问题。出于对这项运动的热爱,他想为 NFL 打造一款 AI 工具。

在 Xiong 的办公室里,醒目地摆放着 Ben Roethlisberger 签名的橄榄球。他和 Yurko 就是从那里开始规划,如何将各自的专长结合起来,形成能够惠及 NFL 及其球迷的研究与技术。他们接洽了 NFL,并被接纳加入 NFL Forward Initiative——该项目让创业者和研究人员提出技术解决方案,用于比赛、训练和球员发展,以及扩大这项运动的影响力。

“我们想看看能否在这些数据上训练基础模型,从而学习橄榄球智慧的种种精妙之处,”Xiong 说。“在这里,橄榄球智慧相当于普通球迷的知识水平。模型能够理解战术,有可能做出战术预测,或者一旦看到某个战术,也能识别出这一战术此前是否被跑过——也许是被另一支球队。”

这款工具并不只是给球迷用的。对教练来说,标注和分析比赛录像与场上球员数据是一项繁重的工作。通过该项目,NFL 让 CMU 获得了比赛录像,以及由 Amazon Web Services (AWS) 提供支持的 Next Gen Stats 场上球员追踪数据;这些数据来自嵌入护肩、比赛装备和橄榄球中的 RFID 芯片,可实时提供场上所有对象的定位信息。

“我们很荣幸能与 CMU 这样享有盛誉的机构合作,共同探索让这项运动更高效的方法,”NFL 橄榄球数据与分析高级总监 Mike Lopez 表示。“创新是我们推动这项运动发展的核心,我们也很期待看到这项研究的成果。”

“过去,研究橄榄球数据需要大量观看录像,这意味着教练和球探要坐着看数百甚至数千小时的影像素材,并手动标注某个具体回合中发生的事情,”Yurko 说。“而有了 NFL 拥有的海量视频素材,以及如今存在的各类 AI 模型,我们实际上可以大规模构建出一些东西,把许多这类繁琐任务自动化。”

Xiong 和 Yurko 收到了数太字节的数据,这也是他们在工作中转向基础模型——即用大量数据训练的深度学习 AI 模型——的原因。如此规模的训练数据,本质上意味着模型可以学到更多,并有可能处理更复杂的任务。

但在研究人员能够做出一个同时处理视频和球员追踪数据的模型之前,他们需要从更小的规模起步。在本例中,他们先构建了两个预训练系统,作为更大模型的基础模块。一个系统使用比赛影像,另一个系统使用球员追踪数据,来完成同一项任务:学习橄榄球。Xiong 表示,第一步本质上是理解场上正在发生什么。下一步则更为复杂。

“例如,一旦模型理解了这一回合之后场上发生了什么,它能否在下一回合开始前做出预测?”Xiong 说。“会不会有突袭?这名安全卫会不会轮转到不同的防守覆盖?”

研究人员需要获取比赛录像数据和球员追踪数据,并创建一个能够从两者结合中学习的模型。要实现这一点可能需要更多数据和更多时间,但这项投入可能带来收益,包括更强的预测能力和更高的效率。例如,Xiong 表示,他们的模型有可能模拟在特定战术指令下某个回合如何推进,或基于预估结果做出战术安排。

“这将帮助教练和球探专注于最重要的事情,也能帮助联盟提高效率,”Yurko 说。

这篇报道背后是谁?

英语硕士,自 2021 年起担任文字编辑,拥有高等教育和健康领域内容经验。致力于可信赖的科学新闻。完整简介 →

物理学硕士,拥有科研经验。长期的科学新闻爱好者。在 Science X 的编辑工作中发挥关键作用。完整简介 →

引用:研究人员教 AI 模型踢足球(2026 年 11 月 5 日),2026 年 11 月 6 日检索自 https://techxplore.com/news/2026-11-football-ai.html

本文件受版权保护。除为私人学习或研究目的的合理使用外,未经书面许可,不得复制任何部分。内容仅供参考。

获取即时摘要文本(Gist)

两个预训练 AI 系统分别使用 NFL 比赛录像和球员追踪数据来学习场上事件。将这两个系统整合为一个多模态模型仍是一项开发目标;自动标注和战术预测尚未得到验证。

本摘要由 LLM 自动生成。完整免责声明

译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(Researchers teach football to AI models)

Researchers teach football to AI models

Researchers teach football to AI models

by Marylee Williams, Jason Bittel, Carnegie Mellon University

edited by Gaby Clark, reviewed by Andrew Zinin

This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility:

Add to Preferred Sources

Researchers teach football to AI models CMU researchers aim to create an AI tool that can predict upcoming plays, simulate outcomes and enhance how professionals and fans understand football games. Credit: Carnegie Mellon University

The NFL has given Carnegie Mellon University 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's Language Technologies Institute, with Ron Yurko, the director of the Carnegie Mellon Sports Analytics Center (CMSAC), to build an artificial intelligence model that can learn football.

"Gaining access to the NFL's official film footage is an incredible opportunity," said Yurko, assistant teaching professor in the Dietrich College of Humanities and Social Science's Department of Statistics & Data Science. "The 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's conducting research with this data."

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's CMU 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's 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's reach.

"We want to see whether we can train foundation models on the data to learn the intricacies of football intelligence," Xiong said. "In 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."

This tool isn't 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 CMU 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.

"We're honored to work with a prestigious institution like CMU as they explore ways to make the game more efficient," said Mike Lopez, senior director of football data and analytics at the NFL. "Innovation is central to how we advance the game, and we're excited to see the results of the research."

"Historically, 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," Yurko said. "And 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."

Xiong and Yurko received several terabytes of data, which is why they are turning to foundation models—deep learning AI models trained on large amounts of data—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's happening on the field. The next step is more complex.

"For example, once the model understands what happened on the field after the play, can it make predictions before the next play?" Xiong said. "Is there going to be a blitz? Will this safety rotate to a different coverage?"

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.

"It's going to help coaches and scouts focus on what matters most and can help the league be more efficient," Yurko said.

Who's behind this story?

Gaby Clark

MA in English, copy editor since 2021 with experience in higher education and health content. Dedicated to trustworthy science news. Full profile →

Master's in physics with research experience. Long-time science news enthusiast. Plays key role in Science X's editorial success. Full profile →

Citation: Researchers teach football to AI models (2026, November 5) retrieved 6 November 2026 from https://techxplore.com/news/2026-11-football-ai.html

This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.

Get Instant Summarized Text (Gist)

Two pretrained AI systems use NFL game film and player-tracking data separately to learn on-field events. Combining these systems into a multimodal model remains a development goal; automated annotation and play prediction have not yet been demonstrated.

This summary was automatically generated using LLM. Full disclaimer

This site uses cookies to assist with navigation, analyse your use of our services, collect data for ads personalisation and provide content from third parties. By using our site, you acknowledge that you have read and understand our Privacy Policy and Terms of Use.

原文为英文,正文由自动抓取并保留原文。

本文由采集器自动同步

原文发布于 TechXplore(Researchers teach football to AI models),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗