0
$\begingroup$

Rosanswers logo

I noticed in the gmapping documentation, it says "This version is optimized for long-range laser scanners ... short range lasers ... will not work that well with the standard parameter settings.".

I was wondering what parameters people have been successfully using for Kinect-based navigation? (My results so far have been less than perfect)

Thanks, -Jon

Here is what the map looks like. My method was: drive into an empty space, and rotate in a circle, then move on to a new space:

image description

--- Edit ---

Here's what the map looks like when I make sure to always have a corner in the field of view when moving to a new part of the map. It's much, much better:

image description


Originally posted by Jon Stephan on ROS Answers with karma: 837 on 2013-01-29

Post score: 2


Original comments

Comment by dornhege on 2013-01-30:
Actually I think if you are doing indoors stuff the short range of the kinect isn't really that much of a problem in comparison to the very limited field of view (55 deg vs. 180-270 deg of common lasers).

Comment by Jon Stephan on 2013-01-31:
Ok, that makes sense. It seems to do a good job building a map when it sees two walls in the same field of view, but not when it has to rotate from one wall to the other.

Comment by Jon Stephan on 2013-02-05:
Yep, tried again making sure I always had a corner in view, and it does much better. Unfortunately, I think that means it probably won't be able to autonomously build a map, without some heavy-duty programming.

Comment by Jon Stephan on 2013-02-05:
It would really help if I could save the map half-way through and use that as the initial map when restarting because it takes so long to map the house and it often screws up half-way through. I think I'll open a new question.

Comment by Jon Stephan on 2013-02-05:
@domhege, if you post the answer "Use the default parameters, and make sure you always have a corner in the field of view", I'll mark that as the correct answer.

Comment by Jon Stephan on 2013-03-08:
By the way - I just upgraded the laptop on my robot (from a single core pentium to a dual core core i3), and it seems to do the mapping much better. I can be a little more careless about my driving and still get a pretty good map.

$\endgroup$

2 Answers 2

0
$\begingroup$

Rosanswers logo

If you can manage to always see at least one corner it should be way better. Aligning on one wall only is quite bad.


Originally posted by dornhege with karma: 31395 on 2013-01-31

This answer was ACCEPTED on the original site

Post score: 2


Original comments

Comment by Jon Stephan on 2013-02-06:
Yes, the default parameters work fine when keeping one corner in the field of view.

$\endgroup$
0
$\begingroup$

Rosanswers logo

I've used the Kinect with the default parameters in gmapping to successfully map out a large area. I did have to tweak the map by hand at the end to get it to be perfect. Could you post an image of your results so that we can see what "less than perfect" means?

In a couple of cases where map generation failed, it was more of a problem of bad odometry or doing crazy maneuvers on the robot.


Originally posted by piyushk with karma: 2871 on 2013-01-30

This answer was NOT ACCEPTED on the original site

Post score: 1


Original comments

Comment by Jon Stephan on 2013-01-31:
What did you use to tweak the map?

I'm traveling, so can't post a map right now. My odom seems pretty good - the laser scan with persistence seems like a better map than what gmapping comes up with.

Comment by Jon Stephan on 2013-01-31:
I think the problem is that it seems like it only uses the odom as a starting point for the scan matching, and a fallback if that fails. I think it would come up with a better map if it used the odom as the scan's position if that gives a reasonable match.

Comment by Jon Stephan on 2013-01-31:
In other words, put more weight on the odom, and less weight on the scan matching.

Comment by piyushk on 2013-01-31:
@Jon Stephan: You reasoning seems right. It is a bit odd that gmapping seems to be performing poorer that simply using odom. I can't comment on how the parameters should change. You might have to go through the original paper to figure it out. I tweaked the map in GIMP.

Comment by Jon Stephan on 2013-02-01:
I couldn't really make sense of that paper (and I'm an engineer by trade, so I would think I should be able to).

Comment by miguel on 2015-09-29:
hey guys, how would you put more weight on odom and less on scan matching with the gmapping params?

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.