Inspired by online music streaming services, a team of engineers from the University of Michigan, Google and Georgia Tech University have developed a particularly simple way to program their own settings when playing music. An exoskeleton comes to their aid.
Of course, experts point out that what is simple for the average user is more complex than it seems, while machine learning algorithms keep offering help profiles that could be more comfortable for an exoskeleton user.
The user then selects one of the two profiles offered and the prediction system suggests a new help profile that they think would be even better. This approach allows users to determine the support an exoskeleton provides according to their preferences, all through a very simple interface that could be implemented on a connected watch or smartphone.
“It looks like the online music service Pandora,” says Elliott Rouse, one of the authors of this study published in Science Robotics. “You give it feedback, a thumbs up or a thumbs down, and it suggests a radio station based on your feedback. Here the idea is similar, but it is about the supporting parameters of an exoskeleton. In both cases, we build a model of user preferences and use it to optimize the individual experience. »
The team tested this approach with 14 participants, each attaching a pair of exoskeletons to their ankles while walking at a brisk speed of about 4 km/h. Volunteers could spend as much time as they wanted between two choices, although they were limited to 50 choices. We would like to point out that most participants repeatedly chose the same help profile in their 45th decision.
After 50 rounds of decision-making, the study team began surveying users to determine whether the final help profile was truly the best and evaluated it against 10 other random and plausible profiles. On average, participants chose the parameters selected by the algorithm about nine out of ten times, highlighting the precision of the proposed approach.
“Using intelligent algorithms and a touch of artificial intelligence, our systems determine what users want with simple yes or no questions,” said Ung Heel Lee, lead author of the study. “I am pleased that this approach will ensure that wearable robots are comfortable and easy to use, bringing us closer to the day when their use will be an integral part of our daily lives.” »
The control algorithm helps manage four parameters of an exoskeleton: the amount of force provided to support the user, the delay between power peaks, and the way the exoskeleton increases and decreases its support on either side of these power peaks. This approach is based on how our calf muscles add force with each step to propel us forward.
According to Mr. Rousse, few exoskeleton users can set their own parameters.
“In most cases, control mechanisms are based on biomechanical or physiological results. Researchers adjust the settings on their computers, minimizing users’ metabolism. “This is currently the standard for the assessment and control of exoskeletons,” said the specialist.
“I think in our field we rely too much on measuring metabolic rate. Humans are actually very insensitive to changes in their own metabolism, so we’re developing exoskeletons to achieve something that humans can’t really sense. »
In contrast, the study authors write, user preference-based approaches not only look at what users can perceive, but also allow them to prioritize the qualities they deem important.