Robotic extremities are not a complete solution for amputees. Technicians usually spend hours manually adjusting them until they fit completely with the walks of the people who wear them and hours that teach them how to walk with them independently without hindering. Costs really increase – tuning sessions require visits to clinics. But the good news is that thanks to Artificial Intelligence (AI), the horizon may have a better way.
In a recent article published in the magazine IEEE deals with cybernetics, researchers from the North Carolina State University and the University of North Carolina describe a system that implements AI reinforcement training that uses a reward system to drive agents for certain purposes to the task of robotic knee setup In one of the tests, the artificial intelligence system they developed took only 10 minutes to help the prosthetic person walk naturally on an even basis.
"Our body does strange things when we have a foreign object of our body," said Jenny S, professor of electrical, computer and energy at Arizona State University and co-author of the article. IEEE Spectrum, "In a sense, our computerized learning algorithm learns to cooperate with the human body."
How it works? When exercising the robotic edge, the AI takes into account various parameters that determine the link between force and movement using the limb – such as the robot stiffness or the vertical movement range allowed in the front. The basic lines are such that the wearers can walk relatively comfortably but not completely smoothly.
In the experiments of researchers, a dozen parameters require corrections. Training data was recorded from amputated walks in short sessions (15 to 20 minutes) and submitted to the algorithm, which over time learned to recognize models of sensors embedded in the prosthesis. In the interest of safety, researchers have imposed restrictions to avoid situations that could cause the user to fall. But the system has reached the parameter settings for stable, smooth walking patterns alone.
This is by no means perfect – the artificial intelligence system can not "know" whether its corrections improve or worsen a particular walking pattern, co-author Helen Huang, a professor of biomedical engineering at North Carolina State University and the University of North Carolina , he said IEEE Spectrum,
"If you want to make this clinically meaningful, there are many, many steps that we need to get through before it happens," she said. "It's really just to show that it's possible – in itself, it's very, very exciting."
But the team is already planning the future work. They intend to train the algorithm to deal with vertical movement, such as footsteps, and hope to create a wireless version of the prosthesis that can be used to collect training data outside of the laboratory sessions.