Efficient Machine Learning Algorithm for Customizing Wearable Devices

Machine Learning Algorithm for Customizing Wearable Devices" />

In the case of soft, assistive devices, such as the exosuit being developed by the Harvard Biodesign Lab, the robot and the wearer have to be in sync. However, the movements of every person differ slightly; therefore, customizing the parameters of the robot for an individual user is an ineffective and time-consuming process.

At present, scientists at the Harvard John A. Paulson School of Engineering and Applied and Sciences (SEAS) and the Wyss Institute for Biologically Inspired Engineering have designed an effective machine learning algorithm with the ability to quickly customize personalized control methods for soft, wearable exosuits.

The study has been reported in the Science Robotics journal.

This new method is an effective and fast way to optimize control parameter settings for assistive wearable devices. Using this method, we achieved a huge improvement in metabolic performance for the wearers of a hip extension assistive device.

Ye Ding, Postdoctoral Fellow, SEAS

While people walk, the way they move is constantly modified to save energy, the so-called metabolic cost.

Before, if you had three different users walking with assistive devices, you would need three different assistance strategie. Finding the right control parameters for each wearer used to be a difficult, step-by-step process because not only do all humans walk a little differently but the experiments required to manually tune parameters are complicated and time consuming.

Myunghee Kim, Postdoctoral Research Fellow, SEAS

Headed by Conor Walsh, the John L. Loeb Associate Professor of Engineering and Applied Sciences, and Scott Kuindersma, Assistant Professor of Engineering and Computer Science at SEAS, the scientists created an algorithm with the ability to minimize that variability and quickly recognize the optimal control parameters for reducing the walking.

The scientists adopted the well-known human-in-the-loop optimization, which uses real-time evaluations of human physiological signals (for example, breathing rate) to tweak the control parameters of the device. Since the algorithm identifies the optimal parameters, it controls the exosuit to provide its assistive force at the right time and at the right place to enhance hip extension. The Bayesian Optimization strategy adopted by the team was first described in a study published last year in PLoSONE.

The combination of the suit and the algorithm minimized the metabolic cost by 17.4% as against walking without the device, which was an improvement of over 60% compared to the earlier study of the researchers.

“Optimization and learning algorithms will have a big impact on future wearable robotic devices designed to assist a range of behaviors,” stated Kuindersma. “These results show that optimizing even very simple controllers can provide a significant, individualized benefit to users while walking. Extending these ideas to consider more expressive control strategies and people with diverse needs and abilities will be an exciting next step.”

With wearable robots like soft exosuits, it is critical that the right assistance is delivered at the right time so that they can work synergistically with the wearer. With these online optimization algorithms, systems can learn how to achieve this automatically in about twenty minutes, thus maximizing benefit to the wearer.

Conor Walsh, John L. Loeb Associate Professor of Engineering and Applied Sciences, SEAS

In the future, the goal of the researchers is to apply the optimization to a highly complicated device that assists different joints (for example,  ankle and hip) simultaneously.

“In this paper, we demonstrated a high reduction in metabolic cost by just optimizing hip extension,” stated Ding. “This goes to show what you can do with a great brain and great hardware.”

The Defense Advanced Research Projects Agency, Warrior Web Program, the Wyss Institute and the Harvard John A. Paulson School of Engineering and Applied Science supported the study.