Landmark Based Outdoor Robot Localization

[mreda]

Hi,
I am working on a project where I will tell my robot to follow a path drawn out with a line. I want it to memorize this path by any means possible so it can replay it without the line there. I have found that landmark localization might be the cheapest way to accomplish this but I am not sure if there is an established system out there or any more details on how to do it other than just the idea. If anyone could shoot me some ideas that would be great.
Thanks,
Matt

[EricDrost]

Hi mreda,

If I understand correctly, you're trying to find your location based off other reference-points with known locations. This is the way GPS technology works. Using the known location of three or more satellites and your distance from them, you can calculate your own location with a linear system of equations.

I have a patent on a similar, reference-point-based, 3D location algorithm. Feel free to PM me if this is helpful and you have any more questions.

[RyanCahoon]

Hi Matt,

Can you explain what you're looking for that's different from what was discussed in your last thread?

[mreda]

Just sort of narrowing down the search. Figured the last thread might be too general/old. But same project.

[Jalerre]

Do you just need the robot to follow a pre-made path? With some sensors such as encoders and gyroscopes you should be able to do this pretty simply through programming using motion profiling or PID loops.

Or do you mean that you want to draw out a path on the ground, have to robot learn the path, remove the line and then have it follow the path that it just learned, or do you mean something completely different?

[mreda]

Originally Posted by Jalerre

Do you just need the robot to follow a pre-made path? With some sensors such as encoders and gyroscopes you should be able to do this pretty simply through programming using motion profiling or PID loops.

Or do you mean that you want to draw out a path on the ground, have to robot learn the path, remove the line and then have it follow the path that it just learned, or do you mean something completely different?

Exactly the second one:

draw path
follow drawn path
learn drawn path
erase drawn path
follow the path based on its memory

[gblake]

Originally Posted by mreda

Hi,
I am working on a project where I will tell my robot to follow a path drawn out with a line. I want it to memorize this path by any means possible so it can replay it without the line there. I have found that landmark localization might be the cheapest way to accomplish this but I am not sure if there is an established system out there or any more details on how to do it other than just the idea. If anyone could shoot me some ideas that would be great.
Thanks,
Matt

There doesn't seem to be enough information about the problem yet in your description.

Answers could currently vary from dropping a track as you follow (and "learn") the path, to setting up transponders for triangulation, to using inertial navigation devices, to putting encoders on a wheeled vehicle, to ...

You also didn't specify the accuracy, precision, or repeatability you need; and/or whether you always start from a known location.

You also implied that you care about cost, but didn't do more than that.

The project sounds like fun, but with the info you supplied so far, your "solution space" is mighty big. Potential helpers will have a hard time offering practical advice without more info.

It's Is this a homework project, or a just-for-fun, on-your-own project?

Blake

[gblake]

Any means possible .... ?

Deploy an aerial minibot carryipng a camera able to see the entire path.
Stabilize the xyz location (and orientation) of the minibot (an exercise left up to the reader).
Record (video) the path-follower as it follows the path, and store the result.
Once the path is removed, have the minibot tell the path-follower where to go to have it retrace the recorded path.

PS: The minibot might be a camera on a tall stick at the beginning of the path, if that camera can see the entire path.

PPS: The minibot might be a camera on a tall stick that is atop the path-following vehicle. Does the terrain close to the path contain discernable landmarks/features?

[mreda]

Originally Posted by gblake

Any means possible .... ?

Deploy an aerial minibot carryipng a camera able to see the entire path.
Stabilize the xyz location (and orientation) of the minibot (an exercise left up to the reader).
Record (video) the path-follower as it follows the path, and store the result.
Once the path is removed, have the minibot tell the path-follower where to go to have it retrace the recorded path.

PS: The minibot might be a camera on a tall stick at the beginning of the path, if that camera can see the entire path.

PPS: The minibot might be a camera on a tall stick that is atop the path-following vehicle. Does the terrain close to the path contain discernable landmarks/features?

I like the idea of the minibot which can watch it from a height. Are their any systems for this that might already exist. For this you are saying that a camera watching over the whole work space can trace its path and then guide it over that path in the future? The camera will obviously have to communicate to the robot wirelessly to make that functional. Is there any more information on this. It sounds like it might be able to work.

[gblake]

Originally Posted by mreda

I like the idea of the minibot which can watch it from a height. Are their any systems for this that might already exist. For this you are saying that a camera watching over the whole work space can trace its path and then guide it over that path in the future? The camera will obviously have to communicate to the robot wirelessly to make that functional. Is there any more information on this. It sounds like it might be able to work.

I know it can work - Making it work is another matter.

Whose project is this? Yours or mine? I brainstormed four or five methods. You get to do the rest.

Good luck doing the research into whatever method(s) you choose, and then iterating your implementation(s).

[mreda]

Originally Posted by gblake

I know it can work - Making it work is another matter.

Whose project is this? Yours or mine? I brainstormed four or five methods. You get to do the rest.

Good luck doing the research into whatever method(s) you choose, and then iterating your implementation(s).

This is part of my project. I had some ideas I was simply trying to find some extra resources and to see if there was something that was already done that I might be able to benefit from. I do appreciate you taking the time and helping me out. I apologize if I appear to be asking for too much.

Thanks again,
Matt

[gblake]

Originally Posted by mreda

This is part of my project. I had some ideas I was simply trying to find some extra resources and to see if there was something that was already done that I might be able to benefit from. I do appreciate you taking the time and helping me out. I apologize if I appear to be asking for too much.

Thanks again,
Matt

I doubt you are asking for too much - I just don't want to give too much (and rob you of the fun of learning ).

Also, I'm no expert in this field, I only know enough to dive in where angels fear to tread.

Have fun,
Blake

[DonRotolo]

Ideas:
First, imagine you wanted to do this in a gym. Perfect environment, flat floor, no slippage, etc. In this case, odometry (measure wheel roaions accurately) and a gyroscope (for turns) might be enough.

Move it outdoors, with uneven and slippery terrain.

Same odometry & gyro, but you can only be assured to get to 'about' the same place (I am ssuming a series of straight lines, connected by turns). So what you need is a "waypoint": a known location that, if your robot gets near enough, it can make a correction to be in exactly the right place at each 'turn'. (for a long straight, there could be one or more in the middle too).

A waypoint can be many things: Two laser beams that cross. A buried RFID tag. A QR-Code-Like target*. A pole that lines up with another pole. A seies of flashing lights, all flashing a different code, and a camera that scans 360 degrees to find its location. You can think of others I'm sure. If the location and path are always the same, it is a lot easier than when it changes all the time.

Hope this gets you thinking down a viable path.

*Look at a modern camera-based automotive wheel-alignment machine.

[Jalerre]

One simple (but not most precise/accurate) way of going about this would be to have to robot follow the line you draw out using infrared light sensors. While it is following the line, have the robot record data from encoders and a gyroscope. Then have the robot use this data to repeat the path it just followed using just the encoders and gyroscope and without the infrared sensors.

I have never tried this so I don't know how well it would work but it theoretically makes sense.

[gblake]

I stumbled on this article today. GA Tech Auton Dirt Track racing

It's about driving an autonomous vehicle as fast as possible around a dirt track.

The work being done and the solutions they are using might connect directly or indirectly to this project's needs.