In a major advance in mind-controlled prosthetics for amputees, University of Michigan researchers have tapped faint, latent signals from arm nerves and amplified them to enable real-time, intuitive, finger-level control of a robotic hand.
To achieve this, the researchers developed a way to tame temperamental nerve endings, separate thick nerve bundles into smaller fibers that enable more precise control, and amplify the signals coming through those nerves. The approach involves tiny muscle grafts and machine learning algorithms borrowed from the brain-machine interface field.
One of the biggest hurdles in mind-controlled prosthetics is tapping into a strong and stable nerve signal to feed the bionic limb. Some research groups—those working in the brain-machine interface field—go all the way to the primary source, the brain. This is necessary when working with people who are paralyzed. But it’s invasive and high-risk.
For people with amputations, peripheral nerves—the network that fans out from the brain and spinal cord—have been interesting, but they hadn’t yet led to a long-term solution for a couple of reasons: The nerve signals they carry are small. And other approaches to picking up those signals involved probes that eavesdropped by force. These “nails in nerves,” as researchers sometimes refer to them, lead to scar tissue, which muddles that already faint signal over time.
The U-M team came up with a better way. They wrapped tiny muscle grafts around the nerve endings in the participants’ arms. These “regenerative peripheral nerve interfaces,” or RPNIs, offer severed nerves new tissue to latch on to. This prevents the growth of nerve masses called neuromas that lead to phantom limb pain. And it gives the nerves a megaphone. The muscle grafts amplify the nerve signals. Two patients had electrodes implanted in their muscle grafts, and the electrodes were able to record these nerve signals and pass them on to a prosthetic hand in real time.
The researchers say that this worked the very first time they tried it. There’s no learning for the participants. All of the learning happens in their algorithms. That’s different from other approaches.
News Source: University of Michigan
