Olympic figure skating looks easy. The players sail across the ice, spinning like tops as they soar into the air and land on a single blade just 4 to 5 millimeters wide. To help figure skaters land quad axels, salchows, lutzes, and perhaps the elusive quintuple without any stress, Jerry Lu MFin ’24 developed the following optical tracking system. woof skate It uses artificial intelligence to analyze videos of figure skaters’ jumps and suggest ways to improve them. Mr. Lu is a former researcher. MIT Sports Labsupports the technical performance of Team USA’s elite skaters and will work with NBC Sports during the 2026 Winter Olympics to help commentators and television viewers better understand the complex scoring systems in figure skating, snowboarding and skiing. He plans to apply AI technology to explain nuanced refereeing decisions and demonstrate how technically difficult these sports are.
Meanwhile, Professor Annette “Peko” Hosoi, co-founder and dean of the MIT Sports Lab, is embarking on new research aimed at understanding how AI systems evaluate aesthetic performance in figure skating. Hosoi and Lou recently chatted Massachusetts Institute of Technology News About applying AI to sports, whether AI systems will be used to judge Olympic figure skating, and when we’ll see skaters land quintos.
question: Why apply AI to figure skating?
Lou: Skaters can always keep pushing higher, faster, and harder. OOFSkate aims to help skaters figure out how to spin a little faster or jump a little higher on their jumps. This system helps skaters capture things that might pass the eye test, but it could also allow them to target high-value areas of opportunity. The artistic aspects of skating are much more difficult to evaluate than the technical elements because they are subjective.
To use the mobile training app, all you have to do is take a video of your athlete jumping, and it spits out physical metrics that determine how much they can rotate. Track these metrics and incorporate all other current elite and former elite athletes. You can look at the data and think, “That’s how the Olympic champions did this element, maybe I should try that too.” You’ll see a comparison and auto-classification feature that tells you that if you performed this trick at a world championship and it was judged by an international judge, this is the approximate grade of execution score you would receive from the judge.
Hosoi: Many AI tools are emerging online, especially pose estimation tools that can approximate skeletal configurations from videos. The challenge with these pose estimators is that when there is only one camera angle, they work very well in the plane of the camera, but very poorly at depth. For example, if you are trying to criticize someone’s fencing form and they are moving toward the camera, you will get very bad data. But when it comes to figure skating, Jerry has found one of the few areas where depth issues are less important. In figure skating, you need to understand the following: “How high did this person jump, how many times did he turn, and how well did he land?” None of that depends on depth. He found an application where the pose estimation function worked very well and didn’t pay any penalties for doing it wrong.
question: Have you ever seen a world where AI is used to evaluate the artistic aspects of figure skating?
Hosoi: When it comes to AI and aesthetic evaluation, new research is underway thanks to an MIT Human Insight Collaborative (MITHIC) grant. This research is a collaboration between Professor Arthur Bahr and IDSS graduate student Eric Liu. When you ask the AI platform to make an aesthetic evaluation, such as “What do you think about this painting?”, it responds with a voice that sounds like a human. What we want to understand is, to arrive at that evaluation, does the AI go through the same kind of reasoning paths or use the same intuitive concepts that humans go through to arrive at “I like that picture” or “I don’t like that picture”? Or is it just a parrot? Are they just copying what others are saying? Or is there a concept map that shows aesthetic appeal? Figure skating is judged aesthetically, so this is a great place to look for this map. And then there are the numbers. You can’t go around a museum and find a score that says, “This painting is a 35.” But in skating, we have data.
That raises another, more interesting question. It’s the difference between a beginner and an expert. It is known that experienced humans and novice humans react differently to the same thing. Someone who is a professional judge may have a different opinion about skating performance than the general public. We are trying to understand the differences in responses from experts, novices, and AI. Do these reactions have something in common in terms of their origins, or does AI come from a different place than both experts and novices?
Lou: Figure skating is interesting because everyone working in the field of AI is trying to understand AGI, or artificial general intelligence, and trying to build this very healthy AI that replicates humans. Efforts to apply AI to sports like figure skating can help us understand how humans think and approach decisions. This ultimately has implications for AI research and companies developing AI models. A deeper understanding of how current state-of-the-art AI models perform in these sports, and how these models need to be trained and fine-tuned to work in specific sports, will help us understand how AI needs to advance.
question: Now that you’ve been studying and working in the figure skating competition at the Milan-Cortina Olympics, what are you paying attention to? Do you think someone will land Quint?
Lou: At the Winter Games, we’re working with NBC on figure skating, skiing and snowboarding competitions to help tell data-driven stories to the American public. The goal is to make these sports more accessible. Skating looks slow on TV, but it’s actually not. Everything should look effortless. If you find that difficult, you will probably be penalized. Skaters need to learn how to spin very fast, jump very high, get airborne, and land beautifully on one foot. The data we’re collecting helps show us how difficult skating actually is, even when it looks easy.
We are happy to be working in the Olympic sports field. This is because Olympic sports, which are watched by the world once every four years, are traditionally coaching-intensive and talent-driven sports. Unlike a sport like baseball, without an elite-level optical tracking system, you won’t be able to maximize the value you currently have. It’s great to be able to work with these Olympic sports and athletes and make an impact here.
Hosoi: Ever since I was able to turn on a TV, I’ve been watching the Olympic figure skating competitions. they are always great. One of the things I will be practicing is identifying jumps, which is very difficult when you are an amateur “umpire.”
I also did some behind-the-scenes calculations to see if Quint is possible. I am now completely convinced that it is possible. We will see such scenes in our lifetimes, if not relatively soon. Not at this Olympics, but in the near future. When I saw us so close together in Quinto, I thought, how about six people? Can you do 6 rotations? Probably not. This is where we begin to face the limits of human physical ability. But I think 5 is within reach.
