Robot position sensing

I like a rotating IR sensor stalk with at least 2 beacons in the room, each modulated uniquely, more if there are obstructions or hallways. With their known positions and the angle the sensor was at when it 'saw' each one you can calculate your location in the room(s) easily.

Optical mouse technology drifts.

I like using a radio direction finder.

A 4 antenna receiving device gives the bearings of two or more transmitters as the robot moves around.

Ramsey kits makes a nice one its a TDOA for about $150

Guy Mac> How creative are you?

A single rotating ultrasound ranger with a geared motor turning it or a couple of hall sensors on the stalk so it stays calibrated. As it turns, it generates a map of the room based on how far walls are from it. If you can control the room (eg, it's your living room, put an infrared diode on one wall (and an infrared receiver on the ultrasound stalk) so it always knows which wall is which. I suppose a compass might work too...

Same would work using a laser rangefinder, of course, or radar.

Let's see... let's come up with something a little more rube goldberg'esque. How about a rotating antenna wave guide that tracks RF interference from your electrical equipment and maps its position based on how much interference it's getting?

Or..... You could use the optical mice technology and periodically calibrate for drift by firing a suction cup in all directions and winding in the thread until you get tension (so you can measure the string). :)

Now THAT'S what I call "creative"!!!

The mental picture of the suction cups firing...

That RF idea is really quite elegant. Would it lose it's way when you turned off your PC?

Do you mean the DF-1 Foxhound?

looks like it's not trivial to get running.

Here's one with RS-232 output.

price for an assembled tested unit.

(I'm interested in something much more limited - two antennas and a tethered transmitter.)

--kyler

You could use this mapping paper as floor material:

Mitch Berkson

How about accelerometers/gyros to get relative motion, with absolute positioning updated by GPS? OK, so this is for a room the size of the AstroDome...

In conjunction with images of the walls and dimensions of the room in the robot, use cameras for orientation and ultrasonic range finders to triangulate.

Bill

Lots of interesting ideas so far!

Here's one:

Send an omnidirectional ultrasonic pulse out from the robot. Pick it up with an unidirectional ultrasonic microphone (probably the same transducer) you will see a pattern of weak echoes from the walls, from the two sided (wall meets wall, wall meets ceiling) corners and from the two sided (wall and wall and ceiling meet) corners. Process them and try to figure out the geometry of the room. That tells you where you are, but not where you are pointed. For that, add a cheap electronic compass. or simply measure- move-measure.

-------------------------------------------------------------

Are we *sure* that optical mouse technology slips? Yes, the mice you buy slip, but they have a 16x16 pixel CCD with two possible output levels (bright/dark). I suspect that a 64x64 matrix with

256 levels would be able to track the pattern going under it without slipping - especially if you kept in the robot's memory a virtual 1024x1024 image with the present "mouse" image centered in the virtual image. (The areas that the system hasn't seen yet will be blank, of course...) I suspect that such a system could keep track of it's position without slipping.

-- Guy Macon, Electronics Engineer & Project Manager for hire. Remember Doc Brown from the _Back to the Future_ movies? Do you have an "impossible" engineering project that only someone like Doc Brown can solve? My resume is at

Info on optical mouse technology:

Could it be that the slippage is all in the internal image to quadrature pulse algorithm, and that we can get an absolute position by processing the image externally?

If all you need is a complete very very exhaustive list check out "Sensors for Mobile Robots" by H. R. Everett, AK Peters, Ltd. 1995 I even think you can obtain the book for free (as a PDF) online somewhere.

Keep in mind that it covers all the different concepts and most of the variations/combinations of some really expensive sensors but it might give you ideas.

Later,

Guy Mac> How creative are you?

Guy:

You might want to read Borenstein's "Where am I?" paper. Here's a URL that contains some interesting references:

By the way, the references come from some work I did with laser triangulation based on some work done by Jim Ubersetzig.

Hope this helps,

-Wayne

Thanks! I will order the book.

As for what I really want, I want an interesting discussion in the comp.robotics.misc newsgroup, and maybe a few ideas I hadn't thought of before. I am getting plenty of both.

Here is another idea: Make a paint that is clear in the visible range but when hit with a red light fluoresces in the infrared range. Scan for it with a red laser that has an IR detector pointed in the same direction as the laser. Paint a few invisible marks that only the robot can see on the ceiling.

Overview:

Long PDF (282 pages): [ ]

This is really good information. Thanks!

The optical mouse, ( assuming the agilent products, ) has a 16 by 16 array at 6 bits per pixel. A digital camera, calibrated to read surface features, could correleate image shift to distance traversed. a sophisticated algo could even account for rotation.

The optical mice are pretty slow for direct image transfers, so mapping them into memory is pretty much out of the question, for the amount of data that you get.

My tests have shown a certain amount of drift between the quad output and real position. Look forward to the April or May issue of Nuts N Volts, where I cover the mouse hack in detail.

Doc Brown from the _Back to the Future_ movies? Do you have an "impossible" engineering project that only someone like Doc Brown can solve? My resume is at

I heard of a commecial mobile robot that used a system consisting of a checkerboard pattern of black/white tiles on the floor. A sensor on the robot simply noted when it went over a tile transition. Probably need to combine this with a direction sensor like a compass, though.

Also, has anyone mentioned using active ultrasonic beacons. A series of ultrasonic transmitters in known locations around the room. They would each transmit a pulse in sequence, and transmit an RF signal to the robot telling exactly when the ultrasonic pulse was transmitted and from which transmitter. The robot would receive the ultrasonic pulses and triangulate using the measured time of flight. Not good for a cluttered environment though.

Dave

A commonly used method for airplane navigation utilizes a ground based transmitter that emits an omni-directional broadcast in combination with a rotating directional beam. It works by first sending out the omnidirectional pulse. Then the rotating beam rotates at a known rate

360 degrees, at which time another omni-direction pulse is sent out. Thus, the receiver times the difference between the two pulses and can thus determine direction to the transmitter. With two of these receivers, each tuned to a different transmitter, a precise position can be determined using triangulation.

A similar technique could be used using IR or other transmissions instead of radio where it may be hard to make a small directional broadcasting antenna, not to mention playing with actual radio beacons may require a license.

-Brian

The most accurate would be to have a human pick up the robot and put it over a little dot on the floor. The robot would know for sure where it is.

There's also the separate overhead camera that looks down, and communicates this with the robot. Robot soccer uses this. Easier to implement than a patterned floor, and these days very cheap. Localizing can be a problem under desks and such (same with most other sensor systems), but a fusion of other techniques like odometers could provide intermediate positioning information untilt he robot is back into the clear.

But I have a reverse question for you: if you're in a room, do you know where you are absolutely? Or by looking around do you know where you are relatively? If you were to navigate through a room, especially one that is new to you, do you really need absolute position? Wouldn't you be looking for features that indicate such things as doorways and pathways, rather than distances from walls?

In other words, does a robot really need to know it's 2.457 feet from the northeast corner of the room, or does it need to know there's an old sock in the middle of the floor that it must steer around or roll over?

-- Gordon Author: Constructing Robot Bases, Robot Builder's Sourcebook, Robot Builder's Bonanza

Guy Mac>

Let's assume that it's a sock-picker-upper robot... :)

Knowing where you are relative to the walls, door, etc is indeed sufficient to navigate. I was being imprecise in my use of the phrase "relative positioning" when I was referring to dead reckoning

- knowing where you are relative to where you were before. The problem. of course, is that dead reckoning errors accumulate.