“One small step, one giant leap…” How AI is changing the way we think about movement

May 10, 2022
FacebookLinkedInTwitter

In 1935, two gymnasts at the University of Iowa developed what is now regarded as the first modern trampoline. A tool for training, rather than just recreation, it even became part of the preparation for NASA astronauts prior to the moon landing – a novel way to simulate the weightlessness of space.

So, why mention this in a blog about AI? It’s to show that technical innovations can have unexpected benefits – and that the opportunities, when we are prepared to approach problems in unconventional ways, are limited only by imagination.

But back to the trampoline.

Over the past 85 or so years, the trampoline has remained relatively unchanged, its improvements evolutionary, rather than revolutionary. Unchanged, that is, until manufacturer Berg asked the simple question, “What if…?” – how do we take the trampoline experience that we all take for granted and try to improve it?

The solution was technological, rather than engineering based. Berg had already reached the pinnacle of physical design and quality – what they needed was a way to offer something more to the experience itself. And the answer is poetry in motion. What if we could measure movement? What if we could then analyse those movements and interpret them in a way that could bring insights? And more simply, what if we could just make bouncing more fun?

But measuring movement is not an engineering problem – it’s a mathematical one. Which is where Wrlds comes into the story…To measure movement – and interpret it – you need a LOT of things to go right. You need sensors that can measure force, rotation, acceleration, impact. You need a way to communicate these metrics for external analysis. You need software that can make sense of the data itself. And only then can you communicate insights to the person doing the bouncing…

As you can imagine, sporting goods manufacturers (no matter how expert within their own industry) don’t tend to have all these specialist capabilities in-house. And assembling an R&D and manufacturing team carries both a significant cost and lengthy time to market. There are of course consultants who can put these teams together, but there’s no guarantee of a successful outcome. It’s all on a project-by-project basis, and the results (and costs) are worryingly unpredictable.

Wrlds is different. All of the skills necessary to bring a motion capture, analysis, and interpretation project to life are assembled under one roof. So all Berg needed to do was explain the idea and let Wrlds explain the possibilities and take it from there.

Well, almost.

I was never going to be able to do a backflip for reference, so we got in contact with one of Sweden’s best jumpers –  this 14 year old kid – he was amazing! Quadruple backflips, five backflips in a row…and we used this motion data as the basis for training the algorithm

Fredrik Bitén, AI and machine learning engineer, Wrlds

In fact, it turns out that catching something as simple as a backflip and generating insights is actually head-scratchingly complex. Broken down step-by-step, It looks something like this:

‘How to train your dragon algorithm’

With Wrlds AI toolbox app and a video enabled phone, record motion data from the sensors alongside time-stamped video of the trick being performed…
Review the video of the trick – say, someone doing a backflip on a trampoline. Label this trick ‘backflip’. Yes, it’s that simple…
Now that we know what a backflip looks like (in terms of data) compare it with the rest of the unlabelled data you’ve recorded. The AI tries to find other backflips. When it gets it right ‘pat on the back’, when wrong, gentle correction…
Once the AI can successfully recognize all the tricks you want to interpret, it can then be either built onto a device’s chip, or installed in an Android or ios app…
So, how many backflips have you done? How high did you jump? How fast can you spin? Are you improving? How do you rate against the competition? Gamification brings improvements (and fun) to motion…

The key to Wrld’s success lies in the fact that the hardest part – building a complete AI pipeline from data capture to interpretation – has already been done. It’s a white label solution that can be used for any activity that involves motion. Think of the core algorithm as a child – all the potential is there, it just needs education, information (and the occasional kind but firm correction) to learn. 

And (as with children) the results are both delightful and surprising. By showing us in an easily comprehensible visual format exactly how we move, we can challenge ourselves to improve, to go higher, faster – and above all, to put the fun back into motion.

The thing about moonshots? They take a lot of things to go right, but the results are worth it.