Revision history [back]

Looking at the AMCL wiki page, you´ll notice that AMCL only subscribes to tf and gets the odometry pose from there. This means that covariance information is not used for AMCL (simply because it is not available via tf). The (probabilistic) motion model used within AMCL is determined by the AMCL node parameters.

One could certainly argue that the "proper" way of doing things would be subscribing to odometry messages and using the covariance information provided within, but this would also make setting up a real system harder.