Exosuit adjusts to individual wearers using machine learning algorithm
Image credit: The Wyss Institute at Harvard University
Researchers at Harvard University have developed a machine-learning algorithm that tailors control strategies for an exosuit and which could help injured patients learn to walk again.
The algorithm is intended to be used with an exosuit under development at the Harvard Biodesign Lab; a lightweight, soft, tethered suit which moves with the body and offers support via an external actuation unit.
Exosuits are wearable robots which support the wearer with their movement. They are most commonly used in clinical settings to help rehabilitate seriously injured patients struggling to walk again, but can also be used to help emergency workers in hazardous environments, or to give soldiers superhuman endurance and strength such that they can march faster and longer while carrying heavy loads.
Despite the potential of exosuits in rehabilitation and other settings, these devices are far from perfect. All humans move differently and we constantly adjust our movements in order to save energy. At present, the Harvard exosuit does not account for this. Tailoring an exosuit for each individual user would take an impractical amount of time.
“Before, if you had three different users walking with assistive devices, you would need three different assistive strategies,” said Dr Myunghee Kim, co-first author of the Science Robotics study.
“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.”
In order to allow exosuits to adjust for each wearer, researchers at the John A Paulson School of Engineering and Applied Sciences developed an algorithm that can identify the best control parameters for a particular wearer.
To achieve this, they used ‘human-in-the-loop’ optimisation. This uses real-time measurements of human physiological signals, such as breathing rate, in order to adjust the exosuit’s control parameters. As the algorithm hones in on more suitable parameters, it sends feedback to the exosuit directing it when and where to deliver an assistive force to support the wearer’s gait.
The researchers found that combining the algorithm with the exosuit could reduce metabolic cost by 17.4 per cent compared to walking without the device - a 60 per cent improvement on the team’s previous attempts.
“This new method is an effective and fast way to optimise control parameters settings for assistive wearable devices,” said Dr Ye Ding, co-first author. “Using this method, we achieved a huge improvement in metabolic performance for the wearers of a hip-extension assistive device.”
“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,” said Professor Conor Walsh, who led the research. “With these online optimisation algorithms, systems can learn how to achieve this automatically in about twenty minutes, thus maximising benefit to the wearer.”
Next, the researchers intend to apply the machine-learning algorithm to a more sophisticated device which gives assistance to multiple joints simultaneously.