Robot mapping preprint reports smaller maps and lower processing costs

Researchers report a robot mapping method that produced maps 13-15x smaller and used 6-8x less integration CPU time than uniform fine-resolution mapping. The preprint describes a feedback loop in which the robot's planned route determines where incoming LiDAR measurements receive fine detail, while the rest of the map stays coarse. Tests replayed two real-world teleoperated datasets offline against an independent full-resolution map. Planned paths had similar cost per meter to those from fine-resolution maps. Uniform coarse mapping achieved the same size savings but increased cost per meter by 210% in the harder scenario.