phil's profile - activity

You need to test. In principle you publish one robot's map on /world and remap /align to /map for the mapstitch process. Then whenever there are enough ORB feature matches in both maps a tf between both maps will be published, i.e. you localize one robot relative to the other one.

You might wanna give the mapstitch package (http:// www.ros.org/wiki/mapstitch) a try.

It uses opencv's estimateRigidTransform method to calculate rotation and translation between two maps. The estimation works best the larger the maps are, that's why mapstitch actually needs three topics of published maps: map, world and align. The map on the align topic will be rotated/translated to the world map and this transformation will be published whenever the map topic gets updated.

We aligned maps after they have been completely build (which was enough in our use case), while replaying the original dataset. It might however also work in a setup with incomplete maps.