Memory-Efficient Gridmaps in Rao-Blackwellized Particle Filters for SLAM using Sonar Range Sensors

Simultaneous Localization And Mapping (SLAM) has been an important field of research in the robotics community in recent years. A successful class of SLAM algorithms are Rao-Blackwellized Particle Filters (RBPF), where the particles approximate the pose belief distribution, while each particle contains a separate map. So far, RBPF with landmark based environment representations as well as gridmaps have been shown to work. Existing gridmap approaches typically used laser range scanners, because the high accuracy of that sensor keeps the state uncertainty low and allows for efficient solutions. In this paper, we present a combination of our previous work on map-matching with RBPF, which enable us to solve the SLAM problem also with low-resolution sonar range sensors. Furthermore, we introduce a simple and fast but very efficient shared representation of gridmaps which reduces the memory cost overhead caused by inherent redundancy between the particles. An experimental comparison to a plain gridmap implementation shows the effective limitation of memory in loop-wise exploration.