Toaster (or any robotics folks)

[Paul Victorey]

To put this simply, I have a theory. It's a theory regarding a possible type of control system for an artificial hand. If successful, I think it would offer a vast improvement over traditional control mechanisms which have changed surprisingly little since the 1980s.

Now, the current artificial hands on the market are very unlike human hands in their structure. This is largely due to control issues. It also means that current artificial hand prosthetics are useless to this research.

What I need to do research on my own theory is an artificial hand that mimics the motions of a human hand. It would have to have all 14 (moving) joints of the human hand, and each joint would have to be independantly controlled and have a range of motion at least as good as the human hand.

For initial testing a hand with only two fingers could work, but the final experiments would need to be run on a hand that replicated the human hand as precisely as possible.

Also, if the hand provided an output of what the specific angle about each joint was, it would save time adding that functionality (which would be needed for tests).

I'm not a mechanical engineer (I'm biomedical/electrical). I have no experience with building anything like this on my own.

So, the questions:

1) Can I *buy* a robotic hand that would work for my tests?
2) If not, could I give CAD drawings to a company and make them manufacture one for me?

I would probably apply for grants to study this, so I should have a budget. Anyone have thoughts or ideas?

[M. A. Dockter]

If you need R-O-M of a human hand, get our a protractor and do some mesurements on your own hand and a few others...that's about the only thing I can help you with.

If you could mechanically imitate human muscles with some mechanical system, I do see it as a great improvement over current technology

[Paul Victorey]

The real problem is not that we can't mimic the hand's mechanical properties, it is that current on the market prosthetics use control systems that are essentially unchanged since the 1960s (there are slight improvements but no huge leaps). Current control systems can't mimic the hand's motions, so building hands as human-like as possible isn't done.

I came up with a design for a control system that looks very promising on paper as a replacement for the current mechanisms. Whether this can hold up to rigorous real-world tests remains to be seen. I may use this project as a Master's project for reasearch; I will be going on to grad school in the fall of 2002.

[Toaster]

Hello Paul and others,
Paul, I ran into problems with a design I tried. While the design may/might work, my biggest problem was "how" to grasp an object and rotate and the like.
Try the following with an assistant and you might get a grasp of what I`m saying:

1. Sit down with a large area that is clear of obsticles for at least 3 feet around you.
2. Have an assistant place an object in this area.
3. The assistant should not define the object in any form.
4. Now search for the object and pick the object up.

Now the fun part:
Ask your assistant that he/she at anytime of thier choosing, choose an object that either will spill or break with ease.
Try multiple objects and pick each up, one at a time.

Ok...thats the A.I of it, now the mechanical part:
This is where things get small, precise and a very close order of movement.
In designing a replica of the human hand, how it works is the hard part because you dont "think" about grabbing something anymore, you just grab it.
Now, tell all joints and other movable parts "how" to move.
To make a "hand" that wouldn`t be overly large and remain useful requires a great deal of flexibility which MUST include "skin" to work correctly.

The only suggestion I could come up with is the following in the actual design stage:
I made something similar out of "popsicle" sticks.
With these, I used wooden spools and any other wooden parts.
The mechanicals are the tricky part. The actual "work doers" such as solenoids, air cylinders, hydrolic cylinders. Working with electromagnetic devices was a blind alley for me due to size and control element sizes.

Is that helpful?

[Paul Victorey]

Hmm... yes.

However, this is not a robotics project per say -- the end product would be a prosthetic hand, to replace an amputated human hand. I'm not as concerned with how to make an AI grasp an object, I have a theory on a new type of control system that would allow an amputee to control the hand AS IF it were an organic hand. And it does it without any implants necessary, unlike neuroelectric control schemes.

I really need to know if it's possible, from ONLY a mechanical standpoint, to build a hand, which moved like the human hand, in which I could control all the joints of the hand independantly.

The control hardware here would not be trying to identify/grasp objects, it would be to identify how the user is trying to move their "hand", and respond in the same way that an organic hand would respond to the same stimuli.

Of course translating one set of control signals to another would be difficult; this control element is the weakest link in current designs, but my theory could bypass or eliminate all the major drawbacks, if it works in the real world.

[Toaster]

Hello again Paul,
Sure its possible, anything is possible.
It sounds as though you wish to make a "mock" up apperati to iron out details. This is where a "sketch" of sorts out of "popsicle" sticks worked for me.
I approach the feasabiliy aspect with a "larger" model to make things more then "paper good". What looks great on paper takes on a whole different life in the real world.
I assemble/partially assemble the project with common materials to examine movement(s) and possible control elements. If this proves useful, I try do downsize to the scale that the project is intended for. At this point, I have mechanical drawings to back up a "mock" build.
Now comes the problem/expense. With the mechanical drawings and your "mock" (you may or may not do the mock part but it helps bigtyme!) to a fabricator that can/will work at the scale you require with materials you stipulate.
If you catch the fabricator during off times, they will be reasonable in cost. In thier "on" times, they will be prohibative.

When I built "Herbert" (my annoying mechanical apperati), I went in steps. First was motivation, next was control elements and finally came "intelligence" (poor word as its not really applicable)
After now 7 years in the making, "Herbert" can only recently navigate "steps". (slowly I might add) In the past 2 years were the "leaps" of voice navigation and verbal annotation of "his" movements. (babble, babble, babble!)
He can open a door with a standard door knob or handle but the process takes about 3 min. for the "logic".
As for your human hand project, please continue but remember that it takes time and should not be rushed.
While I probably took a differing route that yours, I had many of the pitfalls that you WILL incur.
Your device may need more "active elements" (bones and such) then that of a human hand due to complexity of the muscular assist. A trial "big scale" build will enlighten and maybe shed light on other ways to achieve the same results.
I dont know how much help I could be using E-Mail on such a project but feel free to write.


Let me know if i`m failing to get the jest of your questions or project.

[Paul Victorey]

I can't email at all at the moment -- stupid university outbound mail server has been "blacklisted" as a spamming relay, and most of my mail is forbidden if I try to email non-campus addresses. Hope it gets fixed soon.

Actually, I just realized that, at the moment, a computer simulated hand would probably be ideal for testing. Being as the control mechanism is only loosely tied in with the actual mechanical interface, and being as the mechanical interface could be tweaked afterwards, I think I will begin building a simulation program. The real trick is going to be getting inputs from a multitude of sensors into the computer, but I'm an ECE type so I think it's very doable.

The simulation also would make it easier to test, as it would simplify the needed logic. Yet at the same time a simulation would be an excellent first test of the theory, as it tests all of the uncertain elements of the design.

Once I get a control system that is successful I could probably get the aid of our local robotics organization to help with the mechanical design.

[Toaster]

As so far as "stimuli" from the human host, I have no knowlege of this.
In this, I am not the one to ask. As far as a computer simulation, I advise caution as this often leads one down blind alleys.
However, small prototype control elements are something the company I work for is investigating albut to a minimum degree.
I have seen pressure sensors that were based on a barometric pressure system that were incredibly accurate. This is loosely what I used as a low frequency transducer I use for live recording.
Other control elements ive seen are temp variant dependant units, thermo "hase" units that act loosely as infa-red units and low pressure sensors that work in the milli-kpa range.
My old favorite for motion information was the old "sonar" camera device.
I have found that excessive information for a robotics project has many down falls. I tend to use the KISS principle.

I`ll have to see if I can locate any info on the above items, its been a while.

[Paul Victorey]

Well, for my application, a simulation would work equally well for a first run of the theory.

The goal is actually to test to see if I can interpret readings from a new type of sensor, and, based on these readings, determine how the person is trying to move their hand.

So, if the control hardware says "bend finger 2 by 42 degrees at the first joint" -- regardless of whether it 'says' this to a computer simulation or actual hardware, the control hardware is a success. Really, the artificial hand is almost an afterthought; the test themselves would be tests solely focusing on a new type of sensor and a new type of control hardware.

So, my plan was to treat the had as a "black box" that would simply receive the signals from the control hardware, and then use a computer as the "hand". Once the control hardware theory was found to be good or bad, THEN a real hand could be constructed to use the interface.

Now, the question becomes, to capture data necessary to doing control hardware I need also to know the exact angles of all of the joints in my hand at all times. I can easily rig up sensors for the 4 fingers, but I am at a loss as to how to do the thumb...

[Toaster]

Hello Paul,
Actually, I tend to think that the "thumb" would be the easiest part.
Think about it, the thumb does a "grasp" type function and to "steady" or "guide" the hand in may ways.
Me for example, I use my thumb as a "limiter" to guide my fingers close to the object in question and to set this figure. When soldering close work, my thumb sets a depth reference or do a simple "grasp" functon.
The problem I saw with "fingers" is the rapidity of movement and the differing angles of each finger while performing any particular task.
Take the task of writing for example:
Some folks utilize all fingers and thumb and some let one or 2 fingers idle.
This is where my "popsicle" make-up excelled because I could now see this device in all 3 dimensions and move the parts in an almost infinite mannor.