MIT News
Q: Why use AI in figure skating?
Lu: Skaters can at all times push further, higher, faster, stronger. OOFSkate is all about helping skaters discover a approach to spin somewhat faster or jump somewhat higher during their jumps. The system helps skaters catch things that may pass an eye fixed test but may allow them to focus on some helpful areas of opportunity. The artistic side of skating is far harder to guage than the technical elements since it is subjective.
To use the mobile training app, all you have got to do is record a video of an athlete's jump and it’s going to display the physical metrics that determine what number of revolutions you’ll be able to perform. It tracks these metrics and takes into consideration all other current elite and former elite athletes. You can see your data after which see, “This is what an Olympic gold medalist did this item, perhaps I should do this.” You get the comparison and the automated classifier that shows you that in case you had performed this trick at World Championships and it was being judged by a world jury, this is able to be roughly the grade for the execution you’ll receive.
Hosoi: There are many AI tools available online, particularly things like pose estimators that help you approximate skeleton configurations from videos. The challenge with these pose estimators is that in case you only have one camera angle, they work thoroughly within the camera plane, but very poorly in depth. For example, in case you attempt to critique someone's form in fencing and that person moves toward the camera, you'll get very bad data. But with figure skating, Jerry has found considered one of the few areas where depth difficulties don't play a task. In figure skating it’s essential to understand: how high did this person jump, how again and again did he go around and the way well did he land? None of those depend on depth. He found an application that does pose estimators very well and doesn't pay a penalty for the things it does badly.
Q: Could you ever see a world where AI is used to guage the artistic side of figure skating?
Hosoi: When it involves AI and aesthetic evaluation, because of an MIT Human Insight Collaborative (MYTHICAL) grant. This work is in collaboration with Professor Arthur Bahr and IDSS graduate student Eric Liu. If you ask an AI platform for an aesthetic rating, equivalent to “What do you think that of this painting?” It responds with something that feels like it's coming from a human. What we wish to know, to return to this assessment, is: Do AIs follow the identical considering paths or use the identical intuitive concepts as humans to reach at “I like this painting” or “I don’t like this painting”? Or are they simply parrots? Are they simply imitating what they heard from an individual? Or is there an idea map of aesthetics? Figure skating is the proper place to search for this card because there may be an aesthetic judgment involved in ice skating. And there are numbers. You can't walk through a museum and find scores: “This painting is a '35.” But in relation to skating, you have got the info.
This raises one other, much more interesting query: the difference between beginners and experts. It is understood that experienced people and inexperienced people react in another way after they see the identical thing. Someone who’s an authority judge can have a special opinion a couple of skating performance than a member of the final population. We try to know the differences between the responses of experts, novices and AI. Is there a commonality in these reactions where they arrive from, or does the AI ​​come from a special place than each the expert and the novice?
Lu: Figure skating is interesting because everyone who works in the sector of AI is attempting to work out AGI or artificial general intelligence and develop this extremely solid AI that replicates humans. Working on applying AI to sports like figure skating helps us understand how people think and approach judging. This has far-reaching implications for AI research and corporations developing AI models. By gaining a deeper understanding of how current cutting-edge AI models work with these sports and how you can train and refine these models to work for specific sports, you’ll be able to higher understand how AI must evolve.
Q: Having studied and worked on this field, what’s going to you concentrate to within the Milan Olympics figure skating competitions in Cortina? Do you think that anyone gets a fifth?
Lu: For the Winter Games, I'm working with NBC for the figure skating, skiing and snowboarding competitions to assist them tell a data-driven story for the American people. The aim is to make these sports more comprehensible. Ice skating looks slow on TV, however it's not. Everything should look effortless. If it looks harsh, you’ll likely be punished. Skaters must learn to spin in a short time, jump extremely high, float within the air, and land beautifully on one foot. The data we collect will help show how hard skating actually is, even when it's purported to look easy.
I'm glad we're working within the Olympics sports space since the world watches every 4 years and it's traditionally a coaching-intensive and talent-driven sport, versus a sport like baseball where without an elite-level optical tracking system you're not maximizing the worth you currently have. I'm pleased that we are able to work with these Olympic sports and athletes and make a difference here.
Hosoi: I've been watching Olympic figure skating competitions since I could activate the TV. They are at all times incredible. One of the things I’ll practice is recognizing the jumps, which may be very difficult for an amateur “referee”.
I also did some indirect calculations to see if a fifth was possible. I’m now completely convinced that it is feasible. We will see one in our lifetime, if not relatively soon. Not at these Olympics, but soon. When I saw we were so near the fifth, I believed: How about six? Can we do six rotations? Probably not. Here we reach the boundaries of human performance. But five, I feel, is nearby.
