I am excited about our 3DV paper on adaptive-resolution octree-based volumetric SLAM [1] (Emanuele Vespa, Nils Funk, Paul H. J. Kelly, and Stefan Leutenegger): in both the tracking step (ICP) as well as the map fusion, we can choose the voxel resolution adaptively to match the pixel size, whereby the representations at different scales are kept consistent using a lazy-evaluation scheme in the underlying octree data structure. This way, we achieve speed-ups of up to 6 times compared to the single-resolution baseline (supereight), and we much better preserve details where observations from different distances are fused.
