Ask Your Question
0

High Stdev issue using laser_height_estimation

asked 2012-03-09 22:15:32 -0500

alfa_80 gravatar image

updated 2012-03-10 01:07:07 -0500

I was trying to use laser_height_estimation, nevertheless, I obtained the following warning(It's an error, indeed):

[ WARN] [1331374177.626411632]: Stdev of height readings too big to determine height, skipping (stdev is 4.373286, max is 0.100000)
[ WARN] [1331374177.826413613]: Stdev of height readings too big to determine height, skipping (stdev is 4.359676, max is 0.100000)
[ WARN] [1331374178.026414126]: Stdev of height readings too big to determine height, skipping (stdev is 4.382531, max is 0.100000)

Is this a limitation of this package when we have a high standard deviation or any remedy for this one?

Thanks in advance.

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
2

answered 2012-03-10 03:00:32 -0500

Ben_S gravatar image

Looks like you are feeding your complete laserscan into the laser_height_estimation. Since this package does not automatically detect beams corresponding to ground-readings, you have to manually restrict your laserscan to contain only beams, that typically hit the ground. (5-10 degrees looking straight down on the floor)

You should try the laser_scan_splitter and configure it to publish three groups according to your laser scanner. Lets say your scanner is facing straight down (i.e. the middle of the scan hits the ground directly beneath the mav) and let n be the number of the beams in one scan, you should create groups with (n-10)/2, 10 and again (n-10)/2 beams and use the second group as input for the height estimator.

edit flag offensive delete link more

Comments

This warning I obtained when I restricted my laser scanner to have a field of view of 90 deg that pointing downwards. It is the same like what you've explained I think, or do I still need to group them into three?

alfa_80 gravatar image alfa_80  ( 2012-03-10 04:58:34 -0500 )edit

90 degrees is way too much. You should try something like 5 to 10 degrees like i said. You have to make sure, that your scan contains nothing else than ground readings. (And that your transforms are correct of course. The estimator needs to know, that the scanner is pointing downwards.)

Ben_S gravatar image Ben_S  ( 2012-03-10 05:59:52 -0500 )edit

If you can restrict it to that field of view in your driver, you dont need the scan_splitter. That was just a hint how you could easily get a scan with the desired direction and field of view without losing the ability to use the complete scan in another node.

Ben_S gravatar image Ben_S  ( 2012-03-10 06:02:25 -0500 )edit

I've tried both with 10 degree and 6 degree FOV, but then, it complaints because of missing transform. With 90 degree, there's no such missing transform. I agree, with 90 degree is too much, but how do I fix it? Any thought?

alfa_80 gravatar image alfa_80  ( 2012-03-10 06:50:27 -0500 )edit

And with 10 and 6 deg FOV, I still received the error "[ WARN] [1331404549.797437213]: Stdev of height readings too big to determine height, skipping (stdev is 1.741402, max is 0.100000) "

alfa_80 gravatar image alfa_80  ( 2012-03-10 07:00:52 -0500 )edit

How do you exactly limit the FOV of your scanner? This should have no influence on the transforms at all. Did you check your scanresult with rviz? All visible points of the scan have to be on (almost) the same z-level beneath the robot. This should show if there are problems with angles in the scan.

Ben_S gravatar image Ben_S  ( 2012-03-10 09:56:24 -0500 )edit

The way I did it, e.g. for 6 degree FOV, I just regard points that resides in the middle, that means considering points from 528 to 552 because I have 1080 points per scan line with 0.25 angular resolution(or angle increment). Looking at the rviz, it seems to be there with that much of FOV.

alfa_80 gravatar image alfa_80  ( 2012-03-10 23:18:12 -0500 )edit

and beneath the robot as you described.

alfa_80 gravatar image alfa_80  ( 2012-03-10 23:18:50 -0500 )edit

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools

Stats

Asked: 2012-03-09 22:15:32 -0500

Seen: 124 times

Last updated: Mar 10 '12