Ask Your Question
3

ROS amcl package: Interpretation of the abbreviation "AMCL"

asked 2020-10-08 06:42:41 -0600

BoC gravatar image

updated 2020-10-08 06:45:51 -0600

Hello everyone, I am currently working on the theoretical background of Monte Carlo localization. In this context I have a question regarding the general interpretation of the abbreviation "AMCL". According to amcl ROS wiki: (http://wiki.ros.org/amcl)

“amcl is a probabilistic localization system for a robot moving in 2D. It implements the adaptive (or KLD-sampling) Monte Carlo localization approach (as described by Dieter Fox), which uses a particle filter to track the pose of a robot against a known map.”

In this context, the "A" can be understood as "adaptive" in the context of the adaptive change of the number of particles according to the complexity of the belief (adapting the size of particle sets). However, according to further information according to the ROS wiki page, the AMCL package also implements the "Augmented_MCL" algorithm. Augmented_MCL is designed to adapt the number of a randomly generated particles (keyword: recovery from global localization errors).

The AMCL package thus corresponds to a combination of KLD-Sampling-MCL and Augmendetd-MCL. The "A" in "AMCL" thus describes the adaptation of the total number of particles (KLD-Sampling) as well as the adaptation of the randomly generated particles (Augmented-MCL).

Are my previous assumptions correct? Thanks for your feedback!

edit retag flag offensive close merge delete

1 Answer

Sort by » oldest newest most voted
3

answered 2020-10-08 12:13:47 -0600

bob-ROS gravatar image

updated 2020-10-08 15:38:43 -0600

Ok realized some ambiguity. Dieter refers to non-KLD-Sampling as adaptive based on the random sampling. (see 8.3), note that this version not adapt the number of particles. I would call Augmented-MCL "Adaptive" MCL because of adding the random poses. Yes, he also refers to the set of particles as "adaptive" because of the KLD sampling seen in (8.4) which does not have the random addition of poses. Hence the AMCL package is Adaptive Monte Carlo Localization with KLD sampling.

(M is the number of particles). image description image description image description

edit flag offensive delete link more

Comments

Thanks for your reply. I like your interpretation of the "AMCL" package as an adaptive MCL with KLD sampling.I believe the general ambiguity is based on the fact that the abbreviation "AMCL" is interpreted differently in the literature. Some literature sources use "AMCL" in the context of KLD sampling while others use "AMCL" to describe the augmented MCL.

BoC gravatar image BoC  ( 2020-10-09 03:37:07 -0600 )edit

I also like the interpretation from the book "Programming Robots with ROS" (Morgan Quigley et al.): The amcl node implements a set of probabilistic localization algorithms, collectively known as Adaptive Monte Carlo Localization (AMCL).

BoC gravatar image BoC  ( 2020-10-09 04:50:49 -0600 )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

2 followers

Stats

Asked: 2020-10-08 06:42:41 -0600

Seen: 817 times

Last updated: Oct 08 '20