my understanding is current mechatronic prosthetics use electric motors with continuous control of monolithic motors for each joint, which is very much not even close to how our brains control our bodies.

Part of the reason for this is that most prosthetics are controlled by just a few myoelectric sensors reading the controlled activity of what healthy muscle groups the patient has left. Most upper limb amputations are from trauma or something like cancer, where there're not a lot of options when it comes to healthy residual tissue.

To control the actions in the terminal device the patient has to learn how to flex those individual muscle groups to activate them. Not only does this take a lot of practice and concentration, but it can cause a lot of fatigue in muscles that are usually atrophying.

one of the papers also showed that while we do use feedback from sensor neurons in our limbs, its actually pretty slow compared to some of the activities people perform, to the point where the action starts and ends before the feedback even reaches the brain. so basically our brains just perform those rapid actions open loop, and only use the feedback for learning after the event. isnt that fucking cool!

Yeah, we have a couple ways to get around motor planning perturbation. Our cns can anticipate predictable events and start prepping the action before it happens on an open loop as you discussed. Or it can take a reflexive shortcut. When dealing with unpredictable stimuli that over stretches muscle spindle, or over stimulates nociceptors, the peripheral nervus system will skip communication with the brain all together and just communicate with the spine.

that suggests that if we can just make actuators that respond similarly to muscles and connect them with existing nerves we could leverage this system to get pretty far along the way to people having fine motor control over prosthetics

I'm not sure if we could ever get to super fine motor control. Really fine motor control isn't performed quickly, it's dependent on the constant feedback of propiceptors for all the small adjustments we make.

What you are absolutely correct about is that nerve integration would make unconscious and even some reflexive movements possible, and it doesn't even really require a new type of actuator, though that would help.

When I was in college I helped build the shoulder and chest housing for a neutral interface for someone who had an operation for targeted muscle reintegration. The surgeons moved his nerve to reintegrate into an area in his chest where they cut the muscle into a grid, so when you touched a part of the grid, it felt to him like you were touching part of the missing limb.

You build a prosthetic that has haptic feedback so when the prosthetic touches something the interface on his chest would create stimuli to the corresponding area on his chest. I think I have a video somewhere of part of the neural training where he was stacking blocks. At one point he knocked the block over and reflexively caught it before it dropped. Which I believe was one of the first times a powered limb showed signs of unconscious reflexive movement.

Couldn't find the video of him with the blocks, but this is the guy I >helped out on. This is more advanced control than pretty much any limb on the market nowadays and this is from the early 00s. Unfortunately there aren't a lot of people doing Targeted muscle reintegration anymore, it's just too costly. So I think most of it is done at Walter Reed now for research purposes.

Neural interface might be the big problem in prosthetics. Many of the limitations of current prosthetics are due to the janky ass workarounds, like myoelectrics, that are needed to try and facilitate communication between the prosthetics and the brain. Without high fidelity, low latency, bidirectional communication they really don't have the fluidity and "natural" feel and so it doesn't feel so much like your body but like a tool that you are using. There's a reason "the claw" is still preferred by most upper body amputees; it's a really good tool!

Fairly normal electric motors can have fast and "jerky" movements similar to biological muscles, however the issue is the neural signals that are being used to control the motors are TRASH. Even with direct electrode measurements the signals still have fairly high SNR and are averaged across a large population of hundreds-thousands of neurons. The "bangbang control scheme" then leads to unpredictable destructive/constructive interference that makes the already noisy reading very hard to interpret and act on appropriately. Modern prosthetics are as "good" as they are thanks to predictive control schemes and lots of long term averaging. This high uncertainty and slow response time results in what I think you refer to as the "continuous control" type movement.

At present the best neural interfaces we have cause immunological responses that rapidly destroy the signal integrity within, at best, ~8-12 months via scarring, encapsulation and material degredation. That's with the best zwitterionic coatings, low young modulus, embedded anti-inflammatories, etc. In the lab optogenetics and other sorts of "biological amplifiers" have resulted in some of the best bidirectional neuronal communication. As far as I'm aware, the last promising candidate of that sort was targeted muscle reinnervation which is unidirectional and still relies on shitty myoelectrics.