Observing features/landmarks in occupancy grid map generated by GMapping particle filter SLAM
There are feature-based SLAM methods such as EKF SLAM and Graph SLAM which represent the state vector estimate containing both the robot pose and features of interest, i.e. these feature-based SLAM layers can provide a Gaussian prior on the robot pose and on the locations of landmarks.
However, Gmapping SLAM ROS package runs a Rao-Blackwellized particle filter to learn grid maps from laser range data. So the map representation here is an Occupancy Grid Map (OGM) instead of features/landmarks. I am trying to understand how we can obtain landmarks (with position, covariance) from the OGM generated by gmapping SLAM, or obtaining a state consisting of robot pose, OGM, and the landmark locations (landmark pose and covariance) in a ROS implementation.
What are some ways to do this? Should the landmarks' pose and covariance be observed from laser range data directly, or should there be some form of processing on the OGM such that we can get landmark pose and covariance?
