Harvard's Smartwatch Makes You Run Faster & Safer

In this episode, we explore how researchers at Harvard developed an AI-powered wearable system that uses inexpensive motion sensors to estimate running forces that previously required specialized biomechanics labs. The technology could make professional-level gait analysis accessible to everyone, helping runners improve efficiency, reduce injuries, and receive personalized coaching in real time using affordable wearable devices.

Daniel: Anyone out there a runner or a smartwatch user? Well, what if your smartwatch could tell you not just how fast you're running, but why you're getting tired or even warn you that you're about to get injured? On today's episode we're talking about some researchers from Harvard University that are using cheap wearables to get a lot closer to that future than you might expect.

What's up, friends? Welcome back to the Next Byte. As we discussed, today we're talking all about electronics that help athletic performance in terms of running data. But before we jump straight into running, let's talk about the general theme of designing electronics around the human body as wearables in general. If you've been listening to this podcast for more than probably two or three episodes, you're aware that our favorite place to get electronics for a hobby project or a work project that's Mouser Electronics. And we're also proud to partner with them on the podcast because they do a great job of making complex technical topics easy to understand for the layperson. And they've done just that again. So we're linking an article in the show notes about designing electronics for the body, not the bench. And it talks about the history of designing printed circuit boards, PCBs. And in the past, we've had rigid PCBs that were designed for manufacturing and designed for testing on the benchtop. These cause failures if they're a part of wearables in the real world outside of the lab. 

So the new age of engineering is saying things are going to have to be curved, they're going to have to be different shapes, they're going to have to deform, they're going to have to flex, and we have to make materials and sensors and circuit boards that accommodate that if we want to make wearables that are actually going to work on the human body. And so they go in this article, talk about all different types of technologies that have helped advance wearable technology and wearable electronics about wearing the electronics on the body. But the main theme is finding a way to mechanically and electrically comply to the real world conditions. and you you move the primary constraint from being like what can we design to being like how can it be used functionally in the real world. And that's a great primer for what we're talking about today, in terms of wearable technology that can be used for improving running form.

Farbod: Well, I was gonna say, do you have your Oura ring on or no? There you go. We're big, big fans of wearables here on the Next Byte Podcast and Daniel has been rocking with the Oura since like what, 2020? I was gonna say like right around COVID is when I got into it. I got the Whoop strap and then my Apple Watch, and then you were the one that got the Oura ring, but it's so impressive what they've managed to pack in such a small package. And it's all thanks to these, you know, innovations like the flexible circuit boards, the small batteries that can flex. It's really cool.

Daniel: Yeah, do my do my AirPods count as wearables? I guess I'm wearing them. All right. So if that's the case, I'm wearing AirPods, I'm wearing an aura ring and I'm wearing a garment. So I've got four wearables on me right now.

Farbod: You're like the poster child of wearables. A walking ad, some might say.

Daniel: Well, how about a running ad? Because l let's talk about running here. I personally am a runner, so this like this article when Farbod texted me about it, I'm like, Yes, let's do it. But I think it'll be interesting to lots of folks who exercise and enjoy activity of all types. So there's a focus on running here, but I think an analogy to lots of different types of activity.

Farbod: I was gonna say running is like my least favorite activity and it's probably one of your favorite activities. But the wearable aspect of it for exercising in general speaks to me. So I was like, This is gonna be nice; he's gonna like this.

Daniel: Well well let's stop being around the bush, let's jump into it. Have you ever, I assume you have not done a gait analysis before?

Farbod: I have not, but I know my gait's probably messed up.

Daniel: It probably is. And maybe why you don't enjoy running. It was a big reason why I didn't enjoy running when I first switched from sprinting to long distance is like, hey, my running form, if you're doing that over a course of like, you know, a hundred meters, you're sprinting, that's great. You're gonna stop a couple hundred times over the course of like a sprint race, having that form be suboptimal is not a huge deal. Right. It makes you slower. You can improve your form to be faster. But a lot of the things around running form, around speed, around injury risk come into play when you start talking about endurance running. So you think about running in a marathon in which you're taking several tens of thousands, if not a hundred thousand steps. And your form being one or two percent off while you're running a hundred thousand steps or a million steps over the course of like lots of training blocks and stuff like that to get ready for an endurance race, that can add up in your body. And smartwatches are really they're already I would say, pretty much mainstream in terms of endurance sports, using smartwatches to track your pace, to track your heart rate, to track your recovery. But what they can't measure is the form of the person who's running. And they can't measure the ground reaction forces like braking and push-off forces when you're stepping, and you're bouncing your foot off the ground. And when I first started doing some longer distance running. I tapped a mutual friend of mine who's like all his family's like professional runners and Olympic trials and all this stuff and he went on a full-ride scholarship to Stanford to run. I was like, dude, can you just like take a video of me running and look at my form because it feels like I'm doing something wrong. And like, turns out I was. And he told me, here's what you can do differently. 

The thing is is like, that's just a snapshot in time. And even just that one short conversation I had with my buddy Brandon who like helped straighten up my running form, that unlocked a lot for me in the long term of being able to like run several marathons with Nelly and start to enjoy endurance running. But, these researchers from Harvard that we're talking about today are saying like if you've got this constant tracking on your smartwatch of your pace, of your heart rate, and I've even got some biometric or biomechanical data in there. Like they try and see how long my stride length is, seeing like between arm swings and my uprightness, how much I'm bouncing up as opposed to forward during a run. They can give you some rough indicators on what your bio biomechanical form looks like while you're running. But, there's a lot of information that's missing. So they're saying can we take these things that are currently being done only in biomechanics labs, only during a gait analysis, can we constantly track that as a part of a wearable that someone's wearing 24/7 they can wear it during all their runs. And the idea there being we can give you constant feedback and constant collection of data the same way that you've got of your heart rate, but also for your biomechanics, which I would say arguably biomechanics are probably a bigger contributor in the long run to your terms of injury risk as well as speed in as it relates to your running form.

Farbod: And one thing I'll note is that, like you said, in a laboratory setting they've been able to accomplish this sort of analysis. It is also possible to do it outside of the lab, but the barrier is cost. It'll be incredibly expensive. So one of the challenges that these researchers had is, can we do this with affordable off-the-shelf components? And that's the part that really speaks to me, right? It means it can be possible and affordable. And I think what was it like the Google new Fitbit or whatever that came out at a hundred dollars, made it super accessible, everyone's adopting it. That's the kind of like price point I want for these types of wearable technology so that everyone can access it and benefit from it.

Daniel: And spoiler alert, but yes, they were able to achieve that using low-cost, commercially available off-the-shelf IMUs, which stands for inertial measurement units. Those don't directly measure the forces that your feet are experiencing when you're stepping off the ground hundreds of thousands of times during a long run, but they do measure their own motion and their orientation and they can tell the rate at which they are accelerating and decelerating. And so what they did is they trained a machine learning model to be able to predict the forces that an IMU is experiencing based on its installation on a runner. So they used force plates, which were, let's say the laboratory-grade source of truth, ground truth that they could compare the IMU data against and also motion capture. And then they installed these low-cost IMUs. I think what were they doing? One at the hip and one at the ankle. And try and track the forces or or the motion and orientation and acceleration that these IMUs are experiencing and then correlate that back to the force plates, the laboratory ground truth. They found that this machine learning model was really, really good at predicting force from the IMU data alone. So you didn't need laboratory-grade motion capture, you didn't need laboratory-grade force plates, you only needed these low-cost IMUs. And then they took it off the I think it was fifteen runners that they trained it on. They took it to five new runners and they were able to get it to work within just eight steps of the new runner running. So it was like not only does this work for the fifteen runners we trained on, it works on new runners and it works pretty much right away, which is pretty sweet.

Farbod: That's the incredible bit, right? Like it's it's wonderful if you have this technology and it works from fifteen participants to whatever your experimental group is. But as an end user, most folks aren't even reading like the manual that comes out of the box. They just want to slap something on and get on with it. So if within eight steps you're able to fine-tune it to the point that it's pretty much custom to them, custom. That's incredible. And that is the type of user experience that like I would want researchers to focus on when it comes to consumer goods. I love it.

Daniel: For sure. And I will say, we've been complimenting these researchers a lot. we should give them a dual compliment because this is a research group we've recently covered on the podcast and maybe it's because of my bias towards where technology meets endurance sports, 'cause that's interesting to me. But episode 243, when we said cycling super suit moment is here. This is from the same research group at Harvard's School of Engineering and Applied Sciences. It's Connor Walsh and Daniel Lieberman. I think we've we've covered them on the podcast before, and then researcher Lauren Baker's also included in this script.

Farbod: I I when when you're thinking about, like, the impact of something like this, I mean you're a runner, so you're you're gonna have a different takeaway than I do. But for someone like me, like I I've tried jogging and again I have not enjoyed it very much. I don't get the hype. I do a mile, I'm like I could have done so many more things that's gives me more enjoyment than that. I run a forty five minute mile. So what? but it does make me wonder, like you were saying at the beginning, if this can be almost like a training tool, like a real-time trainer. So instead of having your friend who goes to Stanford, there's just an assistant with me that's like, Hey, try adjusting your foot this way or that way. So even if I am hating what I'm doing, at least I'm doing it right. Or, you know, better yet, I actually just learn how to do it and enjoy it.

Daniel: Well, that's what I was gonna say is for me personally, and I think some of my friends have experienced something similar, is like a lot of the thing that made it very unenjoyable as opposed to tolerable, let's say, is having something that was wrong with their running form. And the two things that I think are interesting here. This would be cool if this could democratize like a gait analysis or something like that. If you don't have an expert friend whose entire bloodline is professional runners, like this helps you get access to that information, but also it gives you constant feedback on that information. So I have no idea if my running form today is bad or if it's continuing to creep and especially there's a warning about if you have like a small nagging injury and you run your way through it, you could be compensating for that in ways that you don't know. And so like, say my toe is bruised or something like that, or I broke my toe, and it's gonna persist over several weeks of running. I could be slightly tweaking my running form in a way that is not conscious to me that someone looking at me with the naked eye before or after, they may not be able to notice it, but by having something like these IMUs installed could help me keep tabs on how my form is changing over time, could keep me prevent me from causing a repeated use injury, like a stretch fracture in my shin because I was changing my form because of my broken toe. It could also help give you performance on the efficiency of your running form as well. So a lot of times people are creating a lot of extra movement that burns energy and it's actually counterproductive to them moving forward as fast as they can. And so giving people feedback on something like this, one, could help them prevent them from getting injured, but two, help them better spend their energy while exercising towards moving. And then for about I think if you have both of those things unlocked, right? It doesn't hurt really, really bad when you run because you're not trying to make your body do something it wasn't designed to do. And you're also a lot more efficient at it. Maybe you would enjoy running more if you started banging out like some six-minute miles and they didn't hurt.

Farbod: Maybe. Or maybe that's what Big Running wants me to believe. Jokes aside though. Yeah. Jokes aside though, the other application that I was thinking about is like, you know, joint issues are very common, especially as the population ages. I do wonder, for the non-running crowd, if there could also be like a continuous monitor that's feeding to a doctor or whatever, that's watching for like minor differences as days, as weeks, as months progress, to be like, Hey, I think you might have some sort of a knee issue or whatever issue.

And you know, that's contingent on so many different things. One of them being that like there's gotta be mass data collection and categorization. But I think this could unlock that if folks are open to it and willing to share that information. I know everyone has their own, you know, comfort levels with data privacy, but the health data is always one that I'm like more willing to share because it can have so many positive benefits for other people as well.

Daniel: There's an outsize impact on potentially finding something that was hidden and improving your life. I I agree with you there and just like broad strokes. Well, actually I'm gonna do quick pros and cons on this specific topic, right? So pros I think it works great outside the lab, which is really exciting for me to see them within the first eight steps on those five new runners that they trained it on outside of the training group that it worked, that it uses inexpensive, commercially available off-the-shelf sensors. Gives runners real-time feedback on form and injury risk. So, like I said, not just a snapshot in time of getting a gate analysis one time. It's giving you constant information as your body and your form and your training evolves. Cons, still trained on a small set of data, like twenty people or whatever, however many folks were involved. Like this is great. That's awesome. I would love to see it trained on hundreds of runners, and I'm sure that's part of their next step. But then also another thing that I know that they mention is next steps is: what is the biomechanical information and inferences that my running watch is making today? Can I somehow integrate the IMUs that they're putting on my hip or on my ankle? Can they somehow integrate that with the biomechanical information they're getting from the IMU on my wrist and use that for a more complete picture of my running form? Because again, you wouldn't know it until you've done it and until you've tweaked it just a little bit and you feel the difference. But like the way your arms work and the way your hips work and the way your feet work, those are all pretty much equally important to your running form, even though you think it might be dominated by what your feet are doing. Like changing something small with your arms can actually massively impact your running form. I think right now anyone's only looking at a small part of the total picture. It would be cool if they can somehow access the smartwatch information to get a complete total picture using IMU's.

Farbod: That's a really good point. The only thing I'll contest on your pros and cons list is the small set of data. Because although they trained on fifteen or twenty-three individuals for the running algorithm, the running algorithm is based on a previous model for understanding the impact on walking data. So this is like a evolution of a model. And I think that's why they were able to get to a generalized solution so much quicker, especially with, like, this is basically fine tuned and then after eight steps, it's kind of custom to you. That's not to say that more data isn't gonna refine it even more, but I wouldn't necessarily consider that a con, just as is.

Daniel: No, I'm with you. Good point. what I was gonna say. Yeah, just blind squirrel finds a nut every once in a while. I did want to highlight too, like just in general, I'm pretty bullish on this trend of taking laboratory scale, biomechanics and really biomedical data into the hands of consumers. Like there's a small portion of the world that would say like giving people more information about their health is bad. because people tend to like Google stuff and come jump to the worst conclusion. And people who are hypochondriacs, they don't need more information to like cause themselves to spiral, or people who are perfectionists, they don't need to like over optimize every single aspect of their life. But I think in general, having more biomedical data come out of a lab and be commercially accessible to lots of people in a way that it's super reliable and understandable is going to do net good for a lot of people in the world. Not just as it relates to doing exercise, which is an important part of life. But like I'm I'm very bullish on like my Oura ring and stuff like that. Helping people get a more complete picture of their body. And I think what this team from Harvard is doing is doing a great job of taking the commercially available, inexpensive stuff and integrating that very in a very unique way that makes that something that you could probably buy for less than a hundred dollars off of Amazon to make that little piece of hardware that already exists out there make that super, super valuable and super useful to someone like you and I. Obviously I'm biased there because that's a lot of what we're trying to do during our day-to-day work in the manufacturing realm. Like take commercially available off-the-shelf hardware and integrate that in a system that makes it move the needle a lot for manufacturers, but I'm, generally speaking, a huge fan of the work that they're doing, huge fan of this segment of work. And if anyone's listening and they hear of more technology applications like this, I don't even know if we'll cover them in the podcast, but I'd personally love to read them, read into them myself.

All right, anything else before we wrap up?

Daniel: Alright. Well, most run most running watches do a good job of telling you how fast you're running, maybe how well your heart is beating, but they can't tell you how efficiently you're running. So this team from Harvard developed an AI system that changes that. By combining a few inexpensive wearable motion sensors with machine learning, they can estimate the braking and propulsion forces. These were typically only previously measurable in biomechanics labs.

And these two forces are also highly correlated to both your efficiency and your future injury risk as a runner. This means that future cheap wearables can help runners improve their form, avoid injuries, and train more effectively without ever having to step foot into a research facility.

Farbod: Boom. Solid, man. You really landed that. You like that. That joke at the end, too.

This episode was brought to you by Mouser, our go-to source for electronics parts for any hobby or prototype. Click HERE to discover how modern engineering has revolutionized the construction and durability of wearable tech.

Become a founding reader of our newsletter: http://read.thenextbyte.com/

As always, you can find these and other interesting & impactful engineering articles on Wevolver.com.