How to get the position of a manipulator's end

What will you measure? Where will your reference be?

Jerry

There are any number of choices, depending on how precise you need to be, how much money you have to spend, what sort of an environment you're working in -- the possibilities are almost endless.

I assume that by "calculate the formulas" you mean that you feel that measuring the angle of each joint and calculating the end-affector's position isn't sufficient. Is this correct?

So you're looking at a way to figure out the end-affector's position in an independent, global way? One method that I've seen mooted about is a sort of "sonic LORAN", where a number of transmitters send out ultrasonic pulses, and a receiver detects their times of arrival to figure out where it is. You could augment this with some accelerometers and gyros for some inertial nav effect, but by the time you'd increased your accuracy by any significant degree you may have spent enough money to just hire a full professor & equip him with a tape measure and a level (I don't know about this -- you may be able to get significant accuracy quite cheaply this way).

At worst, you could always use three string pots. This wouldn't be at all good for anything resembling a harsh environment, but then you haven't told us anything at all about yours.

A professor...with a tape measure and a level?

That is sooooo not real time

The OP didn't specify any time requirements, real or otherwise.

Tim Wescott wrote: > The OP didn't specify any time requirements, real or otherwise.

The lag man...the lag...U'll kill us all!!!!!!

How about three non-coaxial cameras? If telecentric lenses with fields as large as the desired motions are unavailable, perspective correction will be needed. The camera angles needn't be mutually orthogonal. Other computations can resolve the positions along orthogonal internal axes. Those would need to be recursed with the perspective correction until a stable solution is found. Modern computers can probably manage that within one scan time of a standard CCTV camera. Simple, really.

Tell me again what's wrong with calculating position from measured joint angles?

Jerry

thanks, but I am worry about the heaviness of the equipment added, cameras may work but I think sometimes the end of the manipulator may be overlapped, so the cameras can't see it. And of course, the computing time must be considered.

Can you explain more about the ultrasonic sender and receiver? I am afraid the end of manipulator may turn to any directions,so where to put the receiver is a big problem.

Before we dig too deeply into possible solutions, consider the recent thread "Quality qualificatif" and give us a hint about why calculating the position from limb lengths and joint angles is inadequate.

I feel that a sonic approach will not be robust in the presence of inevitable reflections. If you try that way anyway, put the sender on the moving member and use three or more fixed receivers.

Jerry

I have never, ever actually used the sonic approach -- I've just seen it mentioned.

I would also like the OP to give us some idea of why he can't measure joint deflections. I could see needing a _really_ precise measurement that wouldn't fly by using joint deflections, but if you need to get that tight you have to expect to sweat bullets for weeks or months to solve your problems.

Is there a robotics newsgroup? This seems like a problem that should have been solved by now...

There must be a robotics newsgroup, but I don't know it.

The position of the end of the arm is seldom of interest in itself. The position *and orientation* of some point on what the arm is holding is more usually needed. Fitting tab A into slot A requires either a learned sequence or a calculation even if the location of a point on the arm is known precisely.

With heavy loads, members can flex and joints be displaced. A learned sequence accounts for that directly if the loading is consistent. Strain gauges can measure the deflections and allow software to account for them. There's often more than is first imagined.

Jerry

My manipulators are designed to do experiments and not intended to a certain industry process, and it is a flexible one, so we can't calculate precisely where the position of the end,only based on the joints' information.

Since the trajectory is constantly changed for several experiments, I have to install a more general detection system for it.

How flexible? How do you flex it? How can you tell the orientation of the end? Even the position of a snake is determined by the sum of the angles of its vertebrae.

Jerry

On Thu, 04 Oct 2007 19:04:43 -0700, workaholic proclaimed to the world:

What about designing a system that uses three or more sonic sources at known locations at the limits of the manipulator work area. Drive synchronized pulse trains from these sound sources and receive these pulses with a receiver mounted on the manipulator arm. You should be able to derive location from the phase shifts that result from the variable distances from the source signals.

I have also looked at some coordinate measuring systems that were able to resolve a tip location to within microns. I am not for sure what they used for sensing, but the arm itself did not support any load, you simply touched to tip to the material being measured.

It would be better to have one sonic source at a reference point on the arm, and three sensors positioned around the device. The sensors should measure flight time of the sound. My guess is that you will want the transducer to send out some patterned sound, so that the sensors can get a better idea of what time each wave left.

Bear in mind that sound velocity may change ever so slightly if one side is warmer than the other, or dustier, or whatever. Also, sound tends to be blocked. I don't know the size of your equipment, so I don't know if it is likely that someone will walk around. Also, if it lifts any material then that will act to deflect sound from one direction or another.

Michael

I second the doubt about using sound. I have never heard of finding an end effector's position by sound before. However, I know people that have done this with cameras. I believe they used two cameras and a lot of math. Maybe even a model or two ( ooooh ). The purpose of the system was to grab or connect to something in the ocean. It had to be able to keep up with the wave action.

I can provide the name of the system integrator that did this and the name of the person on the project if interested.

Peter Nachtwey

Doesn't such phase measurement require an anechoic field?

Jerry

In spite of sonar having been my suggestion, this sounds better.

OP, have you done a web search? Surely this is a solved problem?

My question is the static tip position of the manipulator is easy to obtain, but when it is moving, it oscillates a lot, and my task is to control its very trajectory in an acceptable error, so how to test the actual real-time trajectory is very important to form a closed loop.

And until now I can not find anyone doing this.

Cameras do work in detecting the static tip position, but it doesn't fit in this question.

Sure does. Read this.

"We want to build a closed loop to control the trajectory of the end of a manipulator, how to detect the precise position of it rather than just calculate the formulas? Which kind of sensor should I choose? thanks! "

Peter Nachwey