Revision history [back]

The (undocumented) "map_with_known_poses" parameter can be used together with the "use_tf_pose_start_estimate" parameter set to true to perform mapping with known poses. That is, there is no localization performed, instead a map is created based on the provided tf data. So that´s probably not what you want.

Featureless corridors are indeed tricky and motion is not observable in them (solely from scans). There is no provision to improve the system for this use case currently. We played around with estimating a covariance for scanmatching and one for odometry and fusing those, but discretization artifacts due to the grid map representation backend make this non-trivial. If you have odometry information available, I´d recommend trying gmapping. If not, you certainly try changing hecter_mapping code.