# Get a Quality Score from AMCL

I'm trying to figure out a way of creating a quality score for the amcl_pose.

The most naive solution I could think of is to see "how small/big" the particle cloud gets, and normalize both sizes as 0 and 1. Then map the current cloud size to that scale.

Has anyone done something similar before? Or does anyone know where I can get the particle cloud size from the AMCL package or the code directly?

The point is that when the robot gets really lost (the quality score drops below an acceptable area), I'll ignore it.

Also open to other ideas.

edit retag close merge delete

1

Not sure if I fully understood your question, but what about using the covariance of the pose topic?

( 2018-03-06 03:14:45 -0600 )edit

@Humpelstilzchen from what I understand about covariance, it's just a degree of correlation between different states. Not sure how to extract an accuracy score out of that. Unless you know otherwise? The only way I can see accuracy is the size of the point cloud. Thanks for the input.

( 2018-03-06 17:32:21 -0600 )edit

I'm a bit on unstable terrain here, but in the diagonal of the cov matrix you have the variance of each pose variable, right? And the less confident amcl is about its position the higher its values should be.

( 2018-03-07 00:11:23 -0600 )edit

@Humpelstilzchen I took some sample covariance data. You can see it here: https://docs.google.com/spreadsheets/...

Are you saying the less confident the higher the sum will be in the cells going from top left to bottom right?

( 2018-03-15 17:13:44 -0600 )edit

Yes -- enter at least 7 more characters

( 2018-03-16 01:46:13 -0600 )edit

Thanks for the help. Could you clarify what you mean by enter at least 7 more characters.

( 2018-03-16 10:47:26 -0600 )edit

lol, that was just the error I got when I tried to reply "yes"

( 2018-03-16 11:02:05 -0600 )edit