How to get the position of a manipulator's end

You are asking for the solution to a very difficult problem. In addition to the problem of location, you want to servo out the vibration that inevitably accompanies a flimsy mechanical design.

Jerry

Tim Wescott wrote in news:Y9KdnezmHvko9pnanZ2dnUVZ snipped-for-privacy@web-ster.com:

Given that most robots allow you to define the end effector as a 'tool' very accuratly. (Relative to the joint angle precision, which is usually quite high.) I do not understand why this method will not work, unless you need micron precision 2 meters from the end of the robots last joint. In that case, there should still only need to be certain regions where that accuracy is required, not the entire work envelope. A tri-axis laser set up may work well, but you are still looking at some feedback loop lag. Cameras have a considerable amount of lag by the time it gets back to the robot.

workaholic wrote in news:1192271875.156817.81950 @v23g2000prn.googlegroups.com:

You could do it with a laser array, but it would need to be a dense array and it would be planar.

It may be better to redesign your end effector to have several displacement sensors, maybe by having 4 quadrant placed displacement sensors, attached to thin wires that run the length of the effector. Maybe from these 4 sensors you could determine the minute length changes as the system bends?

Just how precisely must the OP measure position? Are we talking inches, thousandths, tenths of a thousandth, millionths? And what time scale is acceptable for the measurement? How fast is the device moving? Is there something on the manipulator that will impart torque? What is wrong with simple modeling? How rigid is the device?

Somewhere along the line the OP will have to work through the marginal cost of decimal places in measurement. What is the acceptable cost of this project?

Michael

Who do you expect to implement this? This kind of project isn't trivial.

Even if you know the answers what could you do? I agree with Jerry. This project is probably too complicated for a newsgroup. Especially when the OP won't or more probably doesn't know the required data. This is much more than a student problem. Even if I had all the data I wouldn't do anything unless I was being paid. It cost money to pull teeth.

That is what I would do. I don't think that it will be simple though. Notice that the rest of the group members have not mentioned modeling. From previous threads and posts I have come to the conclusion that the group members don't believe in models or more precisely, don't believe their models.

Do you think the OP knows? I bet the model must be determined empirically. That will not be easy. This is much more difficult that just a simple second order system. This system may be an under damped second order upper arm attached to a under damped second order lower arm. I have yet to set a piece of equipment that comes with a transfer function. This system doesn't sound like one that was designed. No one designing the mechanics AND doing the controls would design a system that isn't rigid. If you can't tell by my posts the last year, I have very little respect for most mechanical engineers.

Do you think the OP knows?

Do you think the OP knows? It has been two weeks now. If the OP was serious at all then he would have asked me about the system integrator that can do this job. Now I wouldn't provide the name because it would just be a botther to the system integrator and I wouldn't want to waste their time. We are wasting our time. I know this borders on cynical but it is justified.

Peter Nachtwey

Thanks, but I don't know how much precision can I obtain by the displacement sensors, and time delay of them, if too high, it still can not be used. For my systems is a general purpose experiment equipments, the more precise, the better it will be, so it is not desirable to just add some constraints on it or add other physical adjustment.

...

You need more than the Cartesian coordinates of the tip of the arm. Without knowing the mechanics, its orientation in space remains unknown, and that is critically important if the arm is to be useful.* The arm's load will in any case obscure a sensor behind it in some orientations, and may cause unwanted reflections from a sonic or laser sensor. My imagination may not encompass what you actually propose, but you haven't told us enough to let me imagine more.

You still haven't provided a clue about how you achieve flexibility. Are there many joints, like the spine of a snake? Is there a central compression column surrounded by three or four tension members? It is futile to attempt to design instrumentation for an unknown machine.

Jerry _________________________________________

• Although my own arms are made of rigid parts, the several joints -- rotary and flexural -- make them effectively flexible. In order to sew or use a soldering iron, I need to know much more than the X-Y-Z coordinates of my thumbs.

On Sun, 14 Oct 2007 23:55:00 -0700, pnachtwey proclaimed to the world:

Actually, I assumed that a model was the first thing considered and discounted because of variables that are unknown and cause to many errors.

If you include me as one that does not "believe" in models, then you are mistaken. I have faith in the great model in the sky. I believe! Actually I believe models have their place and are a great asset when used properly and with the understanding that a model is not perfect. What I have seen here is not so much a discounting of modeling, but the discounting of intuitive processes, which are models really.

When I join in this group I assumed it was populated by practicing instrumentation and controls engineers, trading knowledge and discussing the finer nuances of the field, not a cheat site for students. I was somewhat puzzled as to why the members here were willing to take the time to try and find a design solution to some of the problems posed without thoughts of getting paid. It made more sense when I realized that the majority of the question were coming from students. I am willing to help those people, but what about the guy who is out of school and now working for someone who is paying them to solve problems like this. My tolerance goes down in these cases. Consulting fees start coming to mind then.

But seriously, what is your beef and why hang around if that beef gets you angry enough to make the kind of posts you do?

This nebulous system has to be defined a bit more in purpose and construction before you can get a meaningful answer. Anyone can specifiy a need for infinite precision, but very few people have any actual use for it. Even aerospace projects are rarely more precise than a tenth of a thousandth of an inch. In addition to the questions I asked previously, it would be a good idea to know how big this system is going to be in the first place. If it is supposed to have a work volume of a few cubic feet, then it is an entirely different scale of problem than one that must operate in a 50 cubic meter work volume, or one which is constrained only by the (adjustable) length of its arms.

The same goes for speed. Is this thing going to track bullets and carefully alter their trajectory enroute? Is it going to just pick up a box and put it on a shelf?

How about starting with the basics? Just what is this system? How is it designed? Is it a gantry, or a multijoint arm? How fast do you need the answer? Is the precise position during a trajectory needed, or just the endpoints? What is happening during travel?

What are the system requirements of accuracy and precision, and repeatability? In many cases, the extra expense of the extra decimal isn't worth it. If it costs an extra $10,000 to get the next decimal point, is it worth it? Remember, a toaster might be able to pop the toast up to within a thirty second of an inch each time, but all we really care about is that it is up and toasted. We could do with a tolerance of -0.5 +2 inches easily on that position. Making one more precise and accurate on position would be a waste. Instead of worrying about the precision you can obtain with the sensors, look at exactly what precision you need for your application.

It is possible to use optical measurements to detect the position of a gantry to within a few millionths of an inch if you really need to. That can be extended to 3 dimensions if you want. It is expensive, but it can be done.

Simple motor encoding and a good control board will get you a couple decimal places. Is that enough? Perhaps adding some models that define the stiffness of the device, masses, travel, etc will get you a little closer.

Michael

This groups lets too much trash pass for information.

Why do you think I am angry? I am just telling it like it is. The OP is appears to be clueless because of the lack of informatio0n provided.. The OP needs a system integrator that has done something like this before. The OP has a system that will be very hard to control because it flexes. It will cost a lot of money to control because the controller will need to be custom made. It will cost even more money if the system integrator has to spend two weeks or more to get the required information about the system. PaulM, what part do you want to dispute?

Peter Nachtwey

On Tue, 16 Oct 2007 01:42:10 -0700, pnachtwey proclaimed to the world:

On subjects other that models and PID?

I understand your anger with the OP, but you posted about more than just this thread. It was about the entire nature of the group. I want to know more about that. I don't know the credentials of many of the posters here. There regulars, yes, like you, Jerry and Tim. If someone posts trash info in other groups I am involved in and continue to do so, some of the regulars normally get together and post regular disclaimer info about those people, filer out their posts and ignore them. The nature of a open Usenet group is that they always have a fringe element, Kooks if you want to use popular vernacular.

I want very much for this group to be useful to me. After working in the field for 20+ years, due to some health problems, I have been absent from the field during a time where computers have revolutionized my profession...my passion. Keeping up with developments is crucial to me. I don't need to read bullshit, more so than you.

I don't dispute any of this. I think you are correct. It is you approach that I question. Since we know little about the application details, all we can do is throw out possible ways of approaching the OPs problem. I can think of a dozen ways that might work, but this as far as anyone can go with the info provided. This is what you stated (and much more) earlier. I just think you could approach this in a more constructive way. Being angry over how things are in itself is not constructive.

I am sorry, even now I don't know how much will the precison be enough, for it is a research program, and we want establish a experiment platform to do some experiments for flexible manipulator control, so the more precise, the better.

What the good of knowing where it is if you don't know which way it's pointing?

Jerry

I have seen the phrase "the more (of some desirable quality) the better" kill more promising projects and products than anything else. It's a cop out. Quality costs money, so what you're really saying is "I want it to cost more than any budget I can get".

Your grants are finite, and so is your time. Unless you're just working on a Master's thesis where you can get by with some impressive flogging that goes nowhere, you should actually do something productive with your time and money -- if for no other reason than getting stuff done is how you'll end up with more money and help.

I presume you're in an engineering department -- well, engineering is all about getting the best bang for your buck. If you can't decide on what's good enough, perhaps you should allocate a certain amount of money toward measuring your end-effector's position and do your best with that.

I can guarantee you, from personal experience, that trying for infinite performance on a finite budget just leaves you with a smoking hole in your resume.

On Wed, 17 Oct 2007 10:52:18 -0500, Tim Wescott proclaimed to the world:

Here is an example of this. The last job I accepted was to come in to a factory and get a fabricating machine running that had been plagued with problems. The company paid over a million US dollars for the one of a kind line that made a very simple part made from a soft metal stock strip bent into an oval with a tab spot welded on it. The part is used in a exhaust hanger brackets. The machine do the entire process, about 15 steps automatically. When running it produced one every two seconds. It had never run more than around 4 hrs before breaking down. I got some of the bugs out of it and charged for my time. The company was livid that they had to pay me the amount I charged. They were pouring money into getting this thing running.

Here is the rub. They used to make this part by hand on four stations. They had orders for enough of them a year that they had two people working full time at making them on a single shift. The new machine did the entire year supply in 16 hrs of operation, yet they could not get it to run that long. The payback for the cost of the high volume machine was never going to be returned. One of the investor/owners had gotten sold on buying this machine for no reason other than it was neat. Even if it worked flawlessly, it was a stupid management decision.

A million US dollars would have been cheap if the machine had included a flexible manipulator arm.

...

Jerry

On Wed, 17 Oct 2007 15:53:21 -0400, Jerry Avins proclaimed to the world:

Just model the thing as if it were a couple of big springs and be done with it! Avoid the temptation of getting mesmerized by the XYZ indicators that measure in microns.

ok, we are getting somewhere here. If you are looking at basic control techniques, you are not going to need precision to the gnat's eyelash. You can do very well with much less.

How about starting with some basic measurments. Just look at the various angles of each elbow and the length of each section of the arm. That will most closely coordinate with your controls and your actual position. Do your basic experimentation with this level of sensing. Perhaps you might use an lvdt on your arm to help it figure out how closely you are to a fixed object to determine repeatabilty, etc.

After you have learned all you can with running the manipulator with this level of measurement, then examine how much more precisely you need to go for your next step. You will find that most of the controls you want to develop will work very well at the lower level of precision, with much less cost and headache than the super high precision possibilities.

There is a lot to learn for not all that much expense. You may find that even with the simple physical model based positioning, that you are close enough for many applications. After you prove your techniques, then go out and get the money to do it better.

There is such a thing as too much precision. I made an instrument one time which could measure the in situ thickness of a material under 80 tons per square inch pressure to within a millionth of an inch. Unfortunately, it also was able to measure the vibrations of the building during the experiments.

Michael

A couple of springs? How? The OP claims that there are no measurable joints because that his arm is flexible. He hasn't addressed (or won't discuss) the issue of orientation as distinct from position. I think that the concept is too amorphous at this stage to make any in-depth analysis fruitful.

Jerry