I want to model self driving cars. since GPS is not accurate, i was wondering if you know of any tutorials to build an accurate triangulation setup or kits i could buy for less than $100? IT should be accurate within a few centimeters

 

 

I'm not sure there is any technology that meets your requirements.

 

https://anyplace.cs.ucy.ac.cy/

 

https://anyplace.cs.ucy.ac.cy/

And you're representing that a hobbyist can build this for $100.00 or less. What a crock!

 

And you're representing that a hobbyist can build this for $100.00 or less. What a crock!

No. It claims "World-class accuracy of 1.96 meters!" So how is a hobbyist going to better that (by a factor of 50?) for $100?

 

Ok, forget the price limit, does anyone know of anything else?

 

To locate a single car, or multiple?

You might be able to use ultrasonic rangefinding on the car, with multiple axes, to pinpoint where it is relative to the walls.

 

What is the size of the model?

 

Maybe a couple of spinning/scanning light beams and a retro reflector on the model?

 

From a project that I never managed to implement, my conclusion (I could be dead wrong here) was: if range is maybe several tens of meters, you cannot ask for one or two centimeters resolution and the other way round, what ensures mm resolutions negates the many meters option.

There were two moving bodies, starting some 20 m away to finally enter in full contact in a well defined relative position to each other.

Kind of a short bedsheet dilemma to me...

 

I was also thinking about that to create a robot pathfinder, the only decent idea I got is to use (assuming a square-y house) four high power RF transceiver to calculate the input attenuation and by that assume a theoretical position, using four of them you can then quadrangulate the actual position and using a different frrquency transfer that position to the robot itself, or send it to a microcontroller for advanced (TM) privacy destroying.

It's only theoretical and it's probably not a good idea anyway.

 

Don't think it exists. Here is an old MIT project that did this with IR and radio beacons: http://cricket.csail.mit.edu/
At one point I tried to do this with IR, but could not get enough directional information out of them. Pololu has an IR beacon https://www.pololu.com/product/701 that suggests it is possible. It uses 4 IR receivers. The receiver with the strongest signal would be the direction of the other beacon.

Probably worth rethinking the whole idea. For example, in your location, put several IR beacons which are unique (like each broadcasts a unique id that identifies the beacon). For the receiver mounted on your bot, use an IR camera with a 360 degree lens attached to the raspberry pi.

Locations could be derived geometrically or the beacons could serve as your landmarks for GraphSLAM.

The beacons could be Arduino based and could be made for under $25 each. Probably want at least 6 in a room. An IR Pi receiver with camera and 360 degree lens might be $200.

There are a lot of alternatives. Like instead of using IR beacons, place colored circles on your wall and identify them with OpenCV. Or, place the receiver on the ceiling and a beacon on the bot.

 

Or identifiable IR beacons (flashed at different speeds by a '555 would be simple and cheap) and a rotating IR receiver on the bot, like a radar dish.

 

Or identifiable IR beacons (flashed at different speeds by a '555 would be simple and cheap) and a rotating IR receiver on the bot, like a radar dish.

Yes, that would work as a beacon. Probably better to have multiple receivers than rotate one.

Would beacon headings be sufficient to fix location with/without knowing beacon location (i.e. with/without a map)? Is a rangefinder like lidar needed?

 

With the "radar dish" you can get the angle of maximum amplitude. From the angles to the beacons you can triangulate the location relative to the beacons. Then from the map of beacon locations you can fix the bot location on the map.

 

With RF it will be difficult unless you can work with Ghz+ frequencies with single-cycle resolution (though there are techniques that can get around this)

With audio (or ultrasonic) it's a lot more tractable since 33 kHz has a wavelength of about 1 cm. So if "a few centimeters" is good enough, this is not a huge hurdle.

You can set up beacons that transmit a coded stream of bits and have the car lock onto the stream with sub-bit resolution. You do that for several streams and, knowing where the beacons are located, figure out where the receiver has to be. This is actually quite similar to how GPS works.

 

Not sure I follow. Can this be made to work with only ultrasonic beacons?

 

Not sure I follow. Can this be made to work with only ultrasonic beacons?

Provided the beacon is of the right type, it should be able to work. You want a beacon that you can transmit a pseudorandom bit stream of data from.

 

33 kHz was a little longer last time I looked.