Localisation using IR beacons or LiDAR?

Projects General Guidance
1

Hey everyone, first post here, sorry if it's wordy,

I've been trying to find a cheap solution to a localisation problem I'm having with a grid-maze-solving robot. I need to know its XY position as using dead reckoning will give me serious errors.

My initial thoughts were to use active radio beacons, but this was quickly put to bed after I realised how much work and money would have to be invested into something that probably wouldn't even work. RFID is out of the question too as apparently beacons are hard to tell apart and also give inaccurate distances (i.e. signal strengths).

My next idea was to use LiDAR on a rotating axis to scan its environment and map it, then using some method of SLAM to navigate the maze, however this still leaves the problem of actually knowing the robot's position in order to correct any odometry errors.

I'm now at the point where I think combining a 2D LiDAR style of mapping with passive beacons located around the outside of the maze could potentially give accurate triangulation. My thinking is that if I could use IR diodes as beacons and a directional IR receiver on my robot I'd be able to triangulate my robot's position well. Another route is a LiDAR and large diameter beacons which would give 'spikes'. I'd be using a stepper motor in any case to know the rotational angle up to 1.8deg (at an average distance of 2m from beacons I'd imagine this is acceptable).

Here I've found another set of problems, however, as I can't find any IR receivers that are focussed enough; the best I can find is the Sharp GP2D series, giving a 60mm detection diameter at 80cm, not ideal. I can always make my own lens or house the receiver in a tube, however this still doesn't seem ideal. One more product that could be used is a CCD camera like IR Tracking Camera, however I'm not sure if an Arduino can handle this as it feels as though image processing is required?

Anyways, I was hoping that anyone who's done something similar like this or has experience in anything that I've mentioned could give me a couple pointers in the right direction. I'd happily do some testing myself however money's tight and I'd like to hold off on ordering a bunch of stuff until I know for certain my method can work.

Thank in advance!

EDIT: I realise that this maze problem can be solved using other methods (e.g. right hand technique, then algebraic simplification), however I plan on implementing a neural network to solve the maze once mapped so localisation is imperative.

2

Active radio beacons work well, and do exactly what you want. See https://www.pozyx.io/

Dead reckoning using wheel encoders also works very well on a 2WD robot, providing the wheels don"t slip.

3

Hey jremington and thanks for your reply!

jremington:
Active radio beacons work well, and do exactly what you want. See https://www.pozyx.io/

Unfortunately their price is very discouraging (€599 for 1 tag and 4 anchors). I know there are many other projects such as Localino that offer similar solutions but they're also pricey. Implementing my own active radio beacon is way too time consuming and beyond my abilities, although the DecaWave board looks promising for anyone interested (they've got a very active github with a large section on localisation). My budget is <=£100.

jremington:
Dead reckoning using wheel encoders also works very well on a 2WD robot, providing the wheels don"t slip.

I'm sure that is true, and I plan on implementing this as well, however that still introduces an unacceptable amount of error considering how many twists and turns the robot will be making, so I still need a way to update its actual position.

4

however that still introduces an unacceptable amount of error considering how many twists and turns the robot will be making,

Since I know otherwise, I would be interested in whatever evidence you have to support this statement. Perhaps our definitions of "unacceptable amount of error" are very different.

5

jremington:
Since I know otherwise, I would be interested in whatever evidence you have to support this statement. Perhaps our definitions of "unacceptable amount of error" are very different.

Have you done a similar robotics project? How much error did you realistically get? My thoughts are purely speculative at the moment, so I'd love to hear your experience. Just if I want to map the maze accurately I'd rather have an absolute position updated every now and then to correct any drift. I'd like to use the beacon method either way though as it ties in well with my project's context.

6

How regular is this maze? You mention a grid, so I'd expect standardized dimensions a la micromouse. If so, you can use distance to walls sensing in combination with your dead reckoning to navigate.

7

Yes, I build autonomous robots. At this point, your project seems to be entirely speculative and not even vaguely defined, so there are no points for comparison.

Check out the Micromouse contest, which has been running since the 1970s. Many of those mice determine their position using wheel encoders to millimeter accuracy, with corrections derived from wall detection events.