Multi-Robot Mapping

Several robots can map a building faster than one. But each robot builds its map in its own frame, starting from wherever it switched on, so their maps do not line up. To merge them, the robots must find the transform between their frames, for example by seeing each other or by recognizing the same place.

Show both maps overlaid in robot A's frame

Merging two maps

Robot B's map is in its own frame. If robot A sees robot B at a relative pose $T_{AB}$, every cell of B's map can be moved into A's frame:

$$\mathbf{p}_A = T_{AB}\,\mathbf{p}_B$$

Without a meeting, the transform can also come from matching the overlapping parts of the two maps, for example with scan matching, or from recognizing a shared place.

What to notice

Before the robots meet, B's map is drawn as if its frame matched A's, so it lands rotated and shifted, and the overlay makes no sense. Once they meet and measure each other, B's map snaps into place and the two halves form one building. In practice, errors in the meeting measurement and in each robot's own drift must be spread out by optimizing a joint pose graph, much like a loop closure between robots.