{shortcode-a70e313b6f8665428b919ac45d1e26f585ba203d}

Researchers at the Harvard School of Engineering and Applied Sciences have developed a wearable sensor system capable of estimating braking and propulsion forces while running outdoors — a breakthrough that could transform how scientists analyze running mechanics beyond controlled laboratory settings.

In a study published last month in the Public Library of Science journal PLOS One, researchers from the lab headed by SEAS professor Connor J. Walsh — in collaboration with human and evolutionary biology professor Daniel E. Lieberman ’86 — used the sensor system and an accompanying machine-learning model that estimates key biomedical metrics linked to performance and injury risk during real-world running.

Current studies on running mechanics mostly focus on data collected in carefully controlled laboratory environments, like treadmills. Fabian C. Weigend, a co-author of the study, said the team’s primary aim was to bridge the gap between the lab and outdoor conditions that are often more variable.

“A big challenge in the field is that lots of running data are collected in the lab — mostly on treadmills or overground force plates,” Weigend said. “But it’s very difficult to make this transition to running outside in the real world, because outside you don’t have force plates.”

To overcome this challenge, the researchers turned to inertial measurement units — small motion sensors found in smartphones that detect acceleration, angular velocity, and orientation. Each runner in the study wore four sensors on one leg, and the data collected under lab conditions were used to train a machine-learning model capable of predicting braking and propulsion forces during outdoor running.

Lauren M. Baker, the lead author of the study, said that the cyclical nature of the running data made it particularly well-suited for the researchers to apply machine learning methods.

“Running things are very cyclic, and there probably are plenty of parallels that models and like machine learning type models could start to learn better than we can even explain mechanistically,” Baker said.

The researchers looked to the Harvard outdoor track as their ultimate testing ground, where they found the model could accurately estimate running forces based on the sensor-captured motion data. Weigend said he was surprised by how well the lab-trained model translated to real-world conditions.

“The lab over ground measurements actually helped us to translate to the outdoor running measurements,” Weigend said. “To me, it was not directly intuitive that this is possible or that this would work as well as it did in this paper.”

While the findings were promising, Baker noted that their current setup required a more user-friendly design before it could be widely adopted by everyday runners.

“I had double sided tape, would stick the sensor on, and then could wrap it down with varying other types of tape or wraps to make sure it’s not moving,” Baker said. “But obviously, if you want somebody to wear something like this more regularly, that kind of process is probably not gonna fly.”

Baker said that future wearable devices collecting metrics similar to those examined in the study could help runners prevent injuries by providing real-time insights into how to improve their form.

“One of these metrics is over-striding, which is correlated to breaking forces, which could be correlated to different stress related injuries,” Baker said.

“I think generally people are trying to not get injured, right? People are out running around because they’re trying to get some exercise,” she added.