ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question
0

gmapping slam map shrunk

asked 2011-08-25 00:25:20 -0500

davo gravatar image

updated 2011-08-25 00:47:52 -0500

dornhege gravatar image

https://docs.google.com/drawings/d/1e4BW-T9eKJRz9DEIlpouFsAMtXd5bdWstPRDinUFEM0/edit?hl=en_US

My maps are shrunk.....:( I've been working on this for a while > week

This is fairly typical of what is happening, except this time the whole map is shrunk. It starts off ok, the first few metres look great then there's a scan a match and a portion usually the top north west end of the corridoor shrinks from almost 2.70 to less 2.0 and the whole map starts to change.

Been playing with slam options not much joy.

I posted a screen grab on google docs as not only is my flat shrinking I have no karma either.....what a day!

I'm using cturtle, on arm lucid and a base PC under maverick

I have a small issue with magnetic influences as I'm using a compass, but as you can see from the scans fairly straight lines along the corridoor by hook or crook but the compass error is less +- 5 degrees

odom looks ok, error is about 10cm over 8.3 M doing the trig on the broadcast odom

The laser ( ummm sorry, a PML sharp 20-150 cm ir range finder on a 180 degree 1 degree step servo) The laser data looks reasonable to me. I'm scanning every 40 to 50 cm from a halted position

I think there may be some issues with the laser going into and out of shadows which I can work it seems to cause a sweep at the longer range of a wall at an acute angle.

The sharp it's fairly accurate about +-1cm < 100 1.0 -1.5 about +- 2.5 but often better, of course depends on the surface angles etc....that's with filtering and curve fitting not the typical linearisation technique

I have a small issue with magnetic influences as I'm using a compass, but as you can see from the scans fairly straight lines along the corridoor

The features are ok in the undecayed scan points

I've tried lots of slam options, not sure about how to guestimate stt str srt srr or actual what they are.

output of gmapping slam

 -maxUrange 1.5 -maxUrange 1.5 -sigma     0.05 -kernelSize 1 -lstep 0.05 -lobsGain 3 -astep 0.05
 -srr 0.1 -srt 0.2 -str 0.1 -stt 0.2
 -linearUpdate 0.02 -angularUpdate 0.5 -resampleThreshold 0.5
 -xmin -25 -xmax 25 -ymin -100 -ymax 25 -delta 0.05 -particles 120
[ INFO] [1314266203.037209866, 1314264197.070439983]: Initialization complete
update frame 0
update ld=0 ad=0
Laser Pose= -0.00781462 0.0184101 1.97222
m_count 0
Registering First Scan
update frame 1
update ld=0.539888 ad=0.174533
Laser Pose= -0.118015 0.546932 1.79769
m_count 1
Average Scan Matching Score=89.0036
neff= 118.987
Registering Scans:Done
update frame 2
update ld=0.519974 ad=0.0174533
Laser Pose= -0.248544 1.05026 1.81514
m_count 2
Average Scan ...
(more)
edit retag flag offensive close merge delete

4 Answers

Sort by ยป oldest newest most voted
0

answered 2011-09-04 01:05:02 -0500

davo gravatar image

I got side tracked realising that an ir distance sensor is not man enough for the job, so I'm looking at stereo web cams and pointcloud to laserscan conversions to overcome the lack of sensor data, and that also unties from the arduino for laserscan messages.

The coreslam does seem to be better with the pml, and it's highlighted the areas of magnetic variations. I am hoping that with a camera setup to be able to fuse the odometry a bit better either via improved odometery or some form of feature recognition even if I have to coloured label on the wall. The areas of magnetic variation are 10 degrees in 1 cm so it needs to be precise.

Dave

edit flag offensive delete link more
0

answered 2011-08-25 18:38:31 -0500

Mac gravatar image

I think the every half-meter or so is your problem; such long steps between range data can give gmapping's scan matcher a fit. Over a small distance, try taking a scan every 5cm or 10cm, and see if that improves things.

edit flag offensive delete link more
-1

answered 2011-08-25 23:51:04 -0500

davo gravatar image

thanks....

I'll give that a go, I've got a stereo cam on order but it would be nice to get the "pml" mapping just as a win.

Dave

edit flag offensive delete link more
-1

answered 2011-08-26 06:20:19 -0500

davo gravatar image

i upgraded a system to diamondback and been testing coreslam which somewhere there was a hint it may be more suitable for the "PML", The first half of the map is looking good, after tweaking/reducing the theta and sigma. I'm using bag files.

The error in the second half may also be due to the sweeps which may be magnetic variance so I'll try to adjust for that in the odom as well.

Tomorrow when I get myself together I'll run the robot again at short intervals between sweeps.

Thanks

Dave

edit flag offensive delete link more

Comments

In general, the right way to respond is to either (a) edit your original question or (b) comment on my answer. Yes, that's different than the way things usually work, but it's totally better.
Mac gravatar image Mac  ( 2011-08-26 09:24:01 -0500 )edit

Question Tools

Stats

Asked: 2011-08-25 00:25:20 -0500

Seen: 1,001 times

Last updated: Sep 04 '11