In a recent article[bibcite key=volkmar2019improving], the authors kindly cited our works about the comparison of body-powered tuned prosthetic arm technology with myoelectric stuff (body-powered won) by writing: "Therefore, development of prosthetic hands slowed down and stayed behind the modern robotic technology. In the last decade, prosthetic hands with multi-articulated fingers have been commercially introduced. However, due to their high mechanical complexity and poor, under-developed control interface the overall robustness suffered, thus rendering under-actuated prostheses popular to this day ."
"Rendering under-actuated prostheses popular to this day."
Seriously? They write that a body-powered arm with a split-hook is "under-actuated"? Where is that written?
The prosthetic hands and devices with many joints and fingers moving separately: these are underactuated, like, once they only have an "open" and "close" signal, and the rest is maybe just a bit constrained. So there, configuration space options wildly exceed the dimensionality of the control inputs [link].
The totally wild thing here is that the devices these guys write about - body-powered split hook prostheses - are not underactuated: my prosthetic hook as 1 actuated DOF (single ball joint) and a (1) (single) (one) cable to open it, and a spring or rubber that pulls it close. So the single DOF has two actuators for each direction of its single DOF - if I was to state correct terms here, I'd a said that this device is totally on par with actuation. And I'd a let it slipped - but on second thought, this is deeply relevant. So here we are.
In other words: a prosthetic body powered arm with a split hook such as the device that I wear has a configuration space whose options I can fully control and use - and the reverberations of that are non-linear first-order gains: no extra wiggle room mechanically, no extra bulk or material to embody any added DOFs, that is, conceptual sleekness to a very high degree. If you work really hard with these devices that is what you feel, and that feel builds up over days and weeks. Just the bulk is a relevant aspect here: I did compare hand and hook shapes over many attentive use sessions and it dawned upon me that any prosthetic hand just adds visual blockage, visual bulk, adds lack of sleekness to what otherwise is a prosthetic split-hook, the epitome of a skeletonized precision grip, with added shape features that wedge this device far elsewhere in performance space, FAR elsewhere. The more you play with these different prostheses, that is the least that should become clear to you.
In fact, 1:1 actuation and its unseemly first-order gains is one of the hard defining control and build characteristics that make a body-powered split-hook device so massively reliable and robust. And I can tell you that academia still has the hardest time grasping (!) (bwaha) ("grasping" - "prosthetic arm" - get it? bwahaha!) how really low errors in a real everyday body-powerd prosthetic arm usage statistic turns out to be (if the device is built right, that is): we get an error rate a few powers of tens lower. And they never wrote a paper about DOF, underactuation, and how problematic that really is for any truly constrained application domain. As the other grippers that I use - VC, VO, VC/VO - work similar with regard to DOF / DOC ratios, except maybe the widely underpublished Becker hand. So all is sweet and they are not actually underactuated. In addition, I also wear a passively rotating wrist connector, which however has two DOF (turn around longitudinal axis, move out / back in along axis), and accordingly we gave this puppy what amounts to a 2-DOF-lock mechanism. So overall, none of my devices I usually wear to work have DOF that are underactuated.
This distinction of underactuation understood wrongly as here plays a considerable role when designing (really) sleek prostheses, a concept which incidentally also none of the non-users ever really understood: the material instances, the realized works for the not tightly controlled actuations, in fact the extra bulk and weight, as well as a serious associated second-order problem, the dynamical risk (chaos, opportunity, random stuff) is always carried on the prosthesis and that is where it sits as dead weight, and as liability.
So, underactuated devices, for which multiarticulated "bionic" prosthetic hands are typical examples, have more going on in configuration space than in control space which means:
- configuration space contains too much bulk,
- that bulk potentially moves and that is a real risk.
So, extra bulk also spells out as painful shoulder and abraded stump after a bit of typing or wearing such a device for a bit, and, and extra motions that no one controls means trashed shit, and extra wiggle with extra weight in combination then is to cause blisters on the arm. So in a nutshell, with such a prosthetic arm, you hurt a lot more, and you trash a lot more shit - now, hell, crikey to write it down from actuation 1:1 vs underactuation - but that is how we normally describe average myoelectric arm function, also in social media. And I am not kidding you. First time I wore a myoelectric arm to work, stuff fell out of the hand left and right and from the Munster socket my stump was red and blue from bruising, I posted there saying WTF - and all else where, why complain, that is entirely normal, why do you believe we gave up wearing that shit long time ago?
Once we sit down and provide a serious analysis of the DOF of the myoelectric interface, we will see that our worlds collide:
- The input signal to the myoelectric arm actually is a non-binary signal to begin with. The muscles that I use to actuate the open or close command also are actuated for any other reason, such as lifting my elbow or extending the shoulder and such. If I want to avoid this, using a myoelectric arm is far more of a body-powered exercise - keeping elbow and shoulder stiff, etc. - than ever is advertised or made explicit to users. So we start with an input vector whose percentage of true garbage (with regard to the specific goal of opening or closing a prosthetic gripper or hand to be entirely wilful) is significant.
- The transmission of the signal from muscle over skin to electrode introduces further, undeclared DOF: the influence of sweat (to stop the fun, and, to cause electrochemical burns in conjunction with a loose fit), the fit itself (as it may be a bit loose), the dryness of the skin and other aspects will add extra control input dimensions that are outside the direct control of the user but that add to the fun to be had;
- The cables and motor themselves provide a system that acts as an antenna and can (and will, eventually) pick up signals from such items such as tramlines or trainlines;
- The hand device itself has a few degrees of freedoms extra which are partly constrained, partly switchable and in part uncontrollable.
- The combination makes for a really wild error matrix and explains how (a) over 40 years, actually, myoelectric control error rates declined in laboratory or workbench ideal environment tests, and how (b) real world myoelectric error rates (10-30%) are vastly worse than the already unacceptable 1-2% error rates (you drop 1-2 out of 100 cups and tell me what that costs you over 1 year) of myoelectric prosthetic arms.
So, unless these guys get their perception of basic underactuation right, they will never understand what "actuated JUST RIGHT" means and what role that plays in a heavily constrained device. Because, if you understand the actuation and its connection to the constrained use space, only then can you build stuff that works. Not that you ever wanted that - I KNOW - but please pretend you would. This whole website builds on the myth that academic and industrial R&D one day provide a better prosthetic arm than the stuff we have today. Maybe it doesn't but please help me pretend it does. Wait.