Robot localization with AMCL and EKF

asked 2016-04-13 07:37:53 -0500

murdock
75 ●12 ●17 ●20

updated 2016-04-13 08:52:39 -0500

Hello,

I have been following an example here with jackal. My goal so far was to understand how AMCL works. However, in his example he uses ekf_localization and AMCL is just a tool to output cloud with the current robot's position. Here is a rqt_graph when I ran the localization example (which is running gazebo, amcl and rviz) . I thought that AMCL was a separate algorithm for localization and it did not need anything else. What am I missing? Why they are using both EKF and AMCL?

Screenshot of rqt_graph: http://tinyurl.com/z73k3nj

answered 2016-04-13 12:16:41 -0500

al-dev
883 ●7 ●14 ●20

AMCL is a global localization algorithm in the sense that it fuses LIDAR scan matching with a source of odometry to provide an estimate of the robot's pose w.r.t a global map reference frame. It is common to use an EKF/UKF such as those implemented in the robot_localization package to fuse wheel odometry with an IMU (or other sensors) and create an improved odometry estimate (local pose estimation) for AMCL. I am not sure where the example you mention uses both AMCL and an EKF, but it is probably something similar.

Another way to use an EKF together with AMCL is to fuse two global estimates, e.g to fuse the pose provided by AMCL with the pose provided by another global localization method (e.g beacon-based triangulation..)

If you haven't done it already, I suggest you have a look at the robot_localization wiki page and this talk from the 2015 ROSCON