“We’ve taken cheap, large-scale data and learned how to map that to a domain where it’s expensive to collect it for a specific system. And this could be repurposed for any exoskeleton.”
Young likened it to a translator, taking information in one “language” and making it possible for any specific robot to understand the information and use it.
The new technique builds on his team’s previous work developing highly capable AI exoskeleton controllers. But those efforts required years of collecting data on people moving with the exoskeleton before the controller could provide useful assistance. This new class of AI-powered model makes it possible to skip all that.
Like the earlier “task-agnostic” controllers, the AI model isn’t predicting what the user is trying to do — climb stairs or step off a curb, for instance. Instead, it’s instantaneously detecting and estimating how the user’s joints are moving and how much effort they’re exerting. Then the exoskeleton boosts those efforts by as much as 20%.
“One of the things I’m most excited about is seeing how this accelerates not just our lab’s research but also opens the door for our types of controllers to be deployed by roboticists who don’t have access to the equipment that we do,” said Scherpereel, now a senior controls engineer at Skip. “This has the potential to increase the speed and the number of researchers who can work on this. And with that combination, who knows what cool things can be built on this foundation.”
The Science Robotics study proved the AI translation worked with a leg exoskeleton that provided power at the hip and knee joints. But the implications reach much further.
“Broadly speaking, we’ll be able to take the advances in this paper and apply them to upper limb systems, prosthesis systems, and potentially even autonomous robots,” Young said. “That is the big advance. And this is where Matthew’s team has really helped us by creating the AI that does this translation. Now we have the ability to collaborate with industry partners and get these controllers deployed in real systems that people use, hopefully in the near future.”
