Can map merging from multi-agent SLAM work in HTM?

I’ve come across a couple methods in multi-robot SLAM for sharing maps. In one, the robots would see each other and use their respective orientations to re-orient their own maps, then overlay them and combine particles with similar enough locations [1]. Another looked at the relative orientations of groups of three points that were nearby each other, and compared many sets of three points in both maps, then expanded outward until a large enough set of matching points was found [2]. The second one used a sparse information filter, and to maintain sparsity, a method called Bayesian filtering was used [3][4].

Would any of this be possible with nupic? After assuming similar inputs, it seems like bayesian filtering could work with both the spatial and temporal pooler to maintain dendrite sparsity. Once input similarity and dendrite sparsity is assured, it seems like just adding the links on top of each other and then sparsifying would satisfy the map merging problem. (Assuming this was on an HTM system working on SLAM.)

[1] https://www.researchgate.net/profile/Ergin_Ozkucur/publication/220797575_Cooperative_Multi-robot_Map_Merging_Using_Fast-SLAM/links/0f31753b183070a1ed000000/Cooperative-Multi-robot-Map-Merging-Using-Fast-SLAM.pdf
[2] http://robot.cc/papers/Thrun03e.pdf
[3] http://robots.stanford.edu/papers/thrun.tr-seif02.pdf
[4] http://www.ttic.edu/ripl/publications/walter08a.pdf

1 Like