Can AI Make Wearable Technologies and Assistive Devices More Helpful?

Patrick Slade is an Assistant Professor of Bioengineering at the Harvard John A. Paulson School of Engineering and Applied Sciences (SEAS). His research combines wearable robotics, biomechanics, sensing, and artificial intelligence to help people move more easily and safely in everyday life. His lab develops technologies that can measure human movement, adapt assistance to individual users, and support people with mobility or navigation challenges. The following Q&A was developed from interviews with Slade. It has been edited for clarity, length, and context. Q: How does AI improve wearable technologies and assistive devices?A: One thing we have learned in wearable robotics is that putting a motor on someone does not necessarily make it easier for them to move. In some cases, it can actually make movement harder because the device is not working in harmony with the person.We all move a little differently. People vary in their strength, balance, motor control, and movement patterns, and those differences become even more important in patient populations. A device that helps one person may not help another in the same way.AI gives us tools to better understand those differences. We can combine wearable sensors with machine-learning models to estimate things that are difficult to measure outside the lab, such as energy expenditure, walking speed, movement asymmetry, or joint loading. We can then use that information to personalize the assistance — adjusting when it is delivered, how much is provided, and what outcome we are trying to improve.Q: What are the limitations of current wearables, like smartwatches?A: Consumer wearables are very good at collecting data, but some of the metrics they report can be surprisingly inaccurate. Estimates of energy expenditure, or calories burned, can sometimes have errors on the order of 40 to 80 percent.Part of the problem is that many devices rely on indirect signals such as heart rate or wrist motion. Those signals can be useful, but they do not always reflect the mechanical work the body is doing. When you walk or run, much of the energy expenditure comes from the muscles in your legs, not from the movement of your wrist.Our approach is to use signals that are more closely connected to biomechanics. For example, a smartphone carried in a pocket can measure acceleration. We can combine that information with machine-learning models and established principles from physics to estimate meaningful measures such as energy expenditure more accurately.The goal is not simply to collect more data. It is to identify the signals that matter most and use AI in a way that reflects how the body actually works.

Raccontata daseas.harvard.edu