Posts Tagged bionic arms
Advances in the control of prosthetic arms, or even exoskeletal arms, continue to amaze. Yet someone with a severe neck injury doesn’t need any such device since the greatest arm they could imagine is sitting right there hanging off their shoulder — but unable to perform. Efforts to control an artificial arm may seem impotent to these folks, when a bridge spanning just a couple centimeters of scar tissue in the spinal column can not even be made. A way forward is now taking shape at Case Western University in Ohio. Researchers there are gearing up to combine the Braingate cortical chip developed at Brown University with their own Functional Electric Stimulation (FES) platform.
It has long been known that electrical stimulation can directly control muscles. The problem is that it is fairly inaccurate, and can be painful or damaging. Stimulating the nerves directly using precisely positioned arrays is a much better approach. One group of Case Western researchers recently demonstrated a remarkable device called a nerve cuff electrode that can be placed around small segments of nerve. They used the cuff to provide an interface for sending data from sensors in the hand back to the brain using sensory nerves in the arm. With FES, the same kind of cuff electrode can also be used to stimulate nerves going the other direction, in other words, to the muscles.
The difficulty in such a scheme, is that even if the motor nerves can be physically separated from the sensory nerves and traced to specific muscles, the exact stimulation sequences needed to make a proper movement are hard to find. To achieve this, another group at Case Western has developed a detailed simulation of how different muscles work together to control the arm and hand. Their model consists of 138 muscle elements distributed over 29 muscles, which act on 11 joints. The operational procedure is for the patient to watch the image of the virtual arm while they naturally generate neural commands that the BrainGate chip picks up to move the arm. (In practice, this means trying to make the virtual arm touch a red spot to make it turn green.) Currently in clinical trials, the Braingate2 chip has an array of 96 hair-thin electrodes that is used to stimulate a small region of motor cortex.
The trick here is not just to find any sequence that gets the arm from point A to point B, but to find sequences similar to those that real arms actually use in particular tasks. This is important because each muscle has not only a limited contraction range, but also a limited range where it can actually deliver significant force, and generate feedback signals about those forces. When muscles contract they obviously change shape, but less obvious perhaps, is that their shape at any given moment affects how the other muscles leverage the joints they work. Just as important is the effect of the opposing muscles that control counter movements.
Few movements that we make, even low-force movements, consist of pure contractions of the active muscle and pure inhibition of the opposing muscle. In actuality, muscle units on both sides can be firing in alternating bursts to quickly ratchet joint angles open, particularly when the vector of end-point movement is oblique to the axes of individual arm segments. In other words, even in a simple movement like a bench press, both the biceps and triceps generate forces alternately at various points in the lift, despite the fact that the weight rises uniformly in the upward direction.
If artificial methods of control are going to be used for flesh-and-blood systems, particularly ones that have been idle for some time, overstimulation (or mis-stimulation) when lifting anything even slightly heavy is something to be guarded against. Many sports injuries, such as those in older people performing unfamiliar moves, happen not because they reach too far or too hard, but because their nervous system is not sufficiently practiced to be able to protect the muscle.
While no model for limb movement can be perfect, for the majority of everyday tasks, close may be good enough. The eventual plan is that the patient and the control algorithm will learn together in tandem so that the training screen will not be needed at all. At that point, we might say that Case Western will have a pretty slick interface to offer.