0
$\begingroup$

Rosanswers logo

I'm using AMCL with a Velodyne VLP-16. I'm using hector_slam to build the map, saving the map with map_server.

I'm using an IMU and wheel encoder velocities (twist messages) to fuse with robot_localization for odometry and the odom->base_link (base_footprint in my case) transform.

I want to use AMCL for global localization, but I'm having problems. AMCL is not aligning the laser scan with the prebuilt map, leading to incorrect global localization. I've attached a screenshot (if you rotate the laserscan by ~90 degrees clockwise, you get the about the correct alignment.

image description

The only potential problem might be that the laserscan is flat in the velodyne's frame, which is ~0.35m above the ground. Could that be an issue for AMCL?

I'm not very sure where to start debugging this, so any tips/hints/suggestions/requests for more info would be appreciated. I didn't want to dump my entire config in the first post, but I'm more than willing to.

And my AMCL config is pretty standard:

<launch>
<node pkg="amcl" type="amcl" name="amcl">
  <!-- Publish scans from best pose at a max of 10 Hz -->
  <param name="odom_model_type" value="omni"/>
  <param name="odom_frame_id" value="odom" />
  <param name="base_frame_id" value="base_footprint" />
  <param name="global_frame_id" value="map" />
  <param name="use_map_topic" value="true" />
  <param name="odom_alpha5" value="0.1"/>
  <param name="transform_tolerance" value="0.2" />
  <param name="gui_publish_rate" value="10.0"/>
  <param name="laser_max_beams" value="30"/>
  <param name="min_particles" value="500"/>
  <param name="max_particles" value="5000"/>
  <param name="kld_err" value="0.05"/>
  <param name="kld_z" value="0.99"/>
  <param name="odom_alpha1" value="0.2"/>
  <param name="odom_alpha2" value="0.2"/>
  <!-- translation std dev, m -->
  <param name="odom_alpha3" value="0.8"/>
  <param name="odom_alpha4" value="0.2"/>
  <param name="laser_z_hit" value="0.5"/>
  <param name="laser_z_short" value="0.05"/>
  <param name="laser_z_max" value="0.05"/>
  <param name="laser_z_rand" value="0.5"/>
  <param name="laser_sigma_hit" value="0.2"/>
  <param name="laser_lambda_short" value="0.1"/>
  <param name="laser_lambda_short" value="0.1"/>
  <param name="laser_model_type" value="likelihood_field"/>
  <!-- <param name="laser_model_type" value="beam"/> -->
  <param name="laser_likelihood_max_dist" value="2.0"/>
  <param name="update_min_d" value="0.2"/>
  <param name="update_min_a" value="0.5"/>
  <param name="resample_interval" value="1"/>
  <param name="transform_tolerance" value="0.5"/>
  <param name="recovery_alpha_slow" value="0.0"/>
  <param name="recovery_alpha_fast" value="0.0"/>
</node>
</launch>

Originally posted by mjedmonds on ROS Answers with karma: 58 on 2017-05-23

Post score: 2


Original comments

Comment by Humpelstilzchen on 2017-05-24:
How far did you travel without localization?

Comment by mjedmonds on 2017-05-24:
Less than 2 meters. Should I travel farther to get a global alignment?

Comment by Humpelstilzchen on 2017-05-24:
I only observed small updates with amcl, try to drive a few times in circle

Comment by Humpelstilzchen on 2017-05-24:
Also was your 2d pose estimate as accurate as you could?

Comment by mjedmonds on 2017-05-26:
If I properly set the initial pose, it works, but this seems counterintuitive to me. The point of global localization is to align the sensor measurement to a prebuilt map. It seems amcl is really just keep the map roughly in place, but is incapable of a large adjust. Why is this the case? Thanks!

Comment by Humpelstilzchen on 2017-05-27:
Does this answer help you?

Comment by mjedmonds on 2017-05-29:
Yes, thank you. Makes sense.

$\endgroup$

1 Answer 1

0
$\begingroup$

Rosanswers logo

are you setting your initialpose with amcl?

EDIT: about your comment

The point of global localization is to align the sensor measurement to a prebuilt map. It seems amcl is really just keep the map roughly in place, but is incapable of a large adjust.

AMCL generates hypothesis of where the robot could be and how well the measurements of that hypothesis match the current measurement. When you call rosservice call move_base/global_localization it spreads the hypothesis all over the map and, as the robot moves, it reevaluate its hypothesis and prunes the unlikely ones, eventually converging to a reduced set of particles concentrated on the actual position.

But it is not magical, depending on the map you have and the measurements you are using, it may converge to a local minima and then it is hard to escape from there.

Localizing in a featureless map is really hard for AMCL, imagine your map is a square, and the robot is facing a corner. In which corner of the square are you? The /initialpose gives a hint to your filter and it will more often than not, converge to the true position.


Originally posted by Procópio with karma: 4402 on 2017-05-25

This answer was ACCEPTED on the original site

Post score: 2


Original comments

Comment by mjedmonds on 2017-05-29:
/global_localization is what I was looking for. Thanks! Also this answer from Humpelstilzchen explains why

$\endgroup$

Your Answer

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