Swimming: faster thanks to AI

How can the adjustments needed to make a swimmer faster be identified? Science and technology have long been powerless to provide coaches with clear answers. It’s impossible to automatically and precisely reconstruct the movement of a crawl swimmer, for example, by filming it: splashes and limbs that mask part of the body make the task too difficult. Fortunately, in recent years, the miniaturisation of accelerometers, or motion sensors, has made it possible to equip swimmers with them. Except that when they go really fast, things get tough. This is precisely what Jérémie Boulanger, a teacher-researcher at the university’s Lille Research Centre for Computer Science, Signal and Automation(CRISTAL ➊) is working on.

‘One of the questions coaches ask themselves,he says, is the role of the 4 key points in the arm crawl cycle [rising, entering the water, pulling and pushing]. Does the swimmer spend too much time on the first two, which don’t move him forward?’ Does he execute them less well when he wants to go fast? Does the left arm begin its movement just after the right arm has finished its own, or a little before, or a little after?

When the swimmer is going relatively slowly, the sensor data clearly shows the four stages of the stroke. But when he’s going fast, it’s impossible. That’s why Jérémie Boulanger uses artificial intelligence to help the computer learn to recognise the steps automatically. To do this, it must first be told where to find them in the data. It’s a painstaking task carried out by fellow swimming specialists, who scrupulously record them from camera footage. Once this has been synchronised with the data recorded by the accelerometers, the computer can take over. The verdict? After a year of annotating the data, artificial intelligence is beginning to be able to identify the different stages of the crawl with good probability.

With the method now in the process of being validated, it will be possible to study all sorts of things: looking at other strokes such as butterfly, but also understanding the role of pelvic or shoulder roll, a phenomenon that doesn’t help you move forward but is sometimes present, particularly in para-swimmers, etc. These studies are part of a wider project for the 2024 Olympics, the Neptune project,which focuses on many other aspects, such as the swimmer’s flow and the wave he creates. ■