unexpected results from laser sensors

asked 2013-03-19 06:32:09 -0600

SImone
33 ●6 ●7 ●10

updated 2013-03-26 09:50:15 -0600

Hi

I've configured my robot with two inclined laser scanner, then I've wrote some plugin to record laser data and ground truth pose of the robot. I've tried a very simple environment with only the z=0.

I notice that taking the data acquired by sensor, converting them from polar to cartesian and composing with the robot pose, the reconstructed plane is slightly different from the z=0 plane: the minimum height of the point cloud is about 0.002m (~2mm), while the maximum height is about 0.015mm (~1.5cm).

The curious thing is that height is maximum with laser beam acquired at small angles, while decrease with lateral angles. I was expecting that, since there is no noise addition, all the points should be exactly at z=0.

Why this does not happens? Is it related to the ray tracing process performed in the simulation? It depends on the surface of the objects?

edit retag flag offensive close merge delete

Comments

Can you please post the SDF used to create the laser and robot?

nkoenig ( 2013-03-19 11:55:59 -0600 )edit

here https://www.box.com/s/1r97diyyes6ye2tty06w there is the model, sorry for the late in the answer. I've commented out all my plugins invocation, they do not perform anything more than logging data.

SImone ( 2013-03-25 10:39:18 -0600 )edit

I've created a very simple sdf model, with only a laser scanner and some wall as references. I've packed in this file both the model, the world and the results of the scans plotted with an octave script: https://www.box.com/s/bgclvkgn6rtbuzyd9uua. The problem is still present: ranges does not correspond to a line, but have a bell shaped curve. In the packed file you can find a readme.txt file that explain all the details.

SImone ( 2013-03-28 09:38:34 -0600 )edit

any news on this topic?

SImone ( 2013-04-03 10:19:32 -0600 )edit

what version of gazebo are you running? There was a known bug that got resolved by pull request #351 https://bitbucket.org/osrf/gazebo/pull-request/351/fix-pose-computation-in-raysensor/diff in the default branch.

hsu ( 2013-04-03 11:30:42 -0600 )edit

no, this is not my the case for two reasons: 1) the bug you refer was noticed by me 2) the errors are in the ray length, as clearly demostrated with the second experiment I've posted, where measurement are expressed in laser reference frame and not in world reference frame

SImone ( 2013-04-04 02:22:05 -0600 )edit

add a comment

0

answered 2013-05-08 02:14:21 -0600

SImone
33 ●6 ●7 ●10

This was solved since version 1.6.3 (see change log http://gazebosim.org/wiki/Change_log)

edit flag offensive delete link

add a comment

0

answered 2013-04-08 15:42:38 -0600

nkoenig
7561 ●8 ●43 ●70 http://gazebosim.org/

Are you taking into account the minimum range value of 0.08m when processing your data?

edit flag offensive delete link

Comments

This is not the case of the simple test I've conducted: walls are quite far from the sensor, the minimum measure is 2.91 m. However, I've tried to change both resolution and minimum range parameters without any difference. This seems to be a problem related to ray tracing, because measured values change with beam angle.

SImone ( 2013-04-09 05:23:21 -0600 )edit

I am wondering if this could also lead to the problem that I have with the https://bitbucket.org/osrf/simulator_gazebo/src/cfbb9c28a8c0d27979889b7ca6439c23a1d85bbe/gazebo_plugins/src/gazebo_ros_block_laser.cpp?at=default. Generated point cloud namely results in severe curves the largest the pitch and yaw angles become from 0: http://picpaste.com/block_laser2-aaKDsmnO.png?

dejanpan ( 2013-04-12 02:17:49 -0600 )edit

Link to the screenshot again: http://picpaste.com/block_laser2-aaKDsmnO.png

dejanpan ( 2013-04-12 02:18:36 -0600 )edit

It seems exactly the same problem!

SImone ( 2013-04-12 06:52:49 -0600 )edit

@SImone: have you meanwhile been able to get more intuition where the bug could be in? I went into ODEMultiRayShape.cc ODERayShape.cc and printed out the contact points and the directions and they kinda made sense. But I guess that we probably have some rounding or type casting error somewhere..

dejanpan ( 2013-04-12 12:21:38 -0600 )edit

Are you using the latest version of Gazebo?

nkoenig ( 2013-04-24 16:59:06 -0600 )edit

I was not using the latest version, now I've upgraded to the repo version and it seems to be fixed. @dejanpan: the problem seems to be fixed in version 1.6.3 (http://gazebosim.org/wiki/Change_log)

SImone ( 2013-05-07 03:12:53 -0600 )edit

add a comment

unexpected results from laser sensors

Comments

2 Answers

Comments

Question Tools

Stats

unexpected results from laser sensors edit

Comments

2 Answers

Comments

Question Tools

Stats

unexpected results from laser sensors