To my understanding appearance-based SLAM uses only information coming from sensors for mapping and localization. It completely discards control information from actuators. I'm I correct?
This depends on the localization approach on your SLAM (Simultaneous localization and mapping), i.e. AMCL localization algorithm using the laser (sensor) + wheels encoders (actuators sensors).
This algorithm based on adaptive Monte Carlo + EKF (extended Kalman filter)
Appearance-based SLAM methods model the set of discrete locations in a space and the existence of paths between these locations.
I suggest you first read this paper Appearance based slam for mobile robot . Try to know the basic idea behind the appearance based slam.