Society of Robots - Robot Forum

Software => Software => Topic started by: marcus on August 24, 2009, 07:18:15 PM

Title: Wall recognition algorithm
Post by: marcus on August 24, 2009, 07:18:15 PM
Hi,

I am interested in doing some SLAM with my 3Pi Pololu robot.
I have been looking around at the various methods and have decided that i will probably use a sharp ir range finder.
I think i understand the plotting part of it...
But what i cant get my head round is how the robot plots itself on the map its made, i.e. how does it process the ir distance measurements and recognise what reflections are coming from walls and then match that up with its stored map to find its current coordinates.
Im guessing there is some clever algorithm I haven't come across yet..

Any links / descriptions would really be appreciated
Thanks, Marcus
Title: Re: Wall recognition algorithm
Post by: Admin on August 25, 2009, 08:27:05 AM
Quote
doing some SLAM with my 3Pi Pololu robot.
Won't happen! You'll need a fast processor with a lot of memory. Microcontrollers can't do SLAM in any realistic situation (at least not yet).

SLAM works by scanning a map, then the robot moves, and then scans another different map with X% overlap. Using pattern matching of 10's of thousands of scanned datapoints, and estimated change in position, it determines a 'best fit' through statistical analysis. I'm sure if you google around, you'll find free code out there somewhere. You'll need a laptop on your bot to do it, though . . . and a slowly scanning sharp IR (or SICK laser, or lots of sonar ;D).
Title: Re: Wall recognition algorithm
Post by: marcus on August 25, 2009, 09:11:36 AM
ah ok, thanks for breaking it to me    :)
Is there any hope in a less complicated approach where the robot would start from a known location and maybe follow walls and find the intersection of two walls, therefor knowing its location? ive attached an explanation (I think)
Title: Re: Wall recognition algorithm
Post by: Admin on August 25, 2009, 09:59:25 AM
wavefront algorithm?
Title: Re: Wall recognition algorithm
Post by: marcus on August 25, 2009, 10:01:43 AM
Thats what i was looking at ( this site is so helpful )
So do you think it could work? id have to have a think about how to deal with objects because that could throw it off completely..
Thanks, Marcus
Title: Re: Wall recognition algorithm
Post by: Admin on August 25, 2009, 01:06:45 PM
Well, the advantage to SLAM is that your map won't drift.

With wavefront, over time, your map will drift off course no matter how accurate your sensors are. Global sensors, like GPS or a compass, can prevent drifting.
Title: Re: Wall recognition algorithm
Post by: marcus on August 26, 2009, 04:46:48 AM
hmm, ok.
ive had a look for indoor compasses but cant find anything, do they exist? Because ive tried with a normal compass and the electrics through it off..
Thanks, marcus
Title: Re: Wall recognition algorithm
Post by: SmAsH on August 26, 2009, 05:12:51 AM
normal compasses can be thrown off my "noisy" circuits if near them...
if you keep them away from that and anything magnetic it should be fine...