An Efficient Rao-Blackwellized Genetic Algorithmic Filter for SLAM

A Rao-Blackwellized particle filter approach is an effective means to estimate the full SLAM posterior. The approach provides for the use of raw sensor measurements directly in SLAM, thus obviating the need to extract landmarks using complex feature extraction methods and data association. In this paper a solution framework based on Rao-Blackwellized particle filters (RB) and genetic algorithms (GA) is proposed for recovering the full SLAM posterior using raw exteroceptive sensor measurements, i.e. without landmarks. The resultant Rao-Blackwellized genetic algorithmic filter (RBGAF) permits the uses of any arbitrary measurement model unlike FastSLAM with scan matching. Since the proposed method represents the environmental map state for each robot trajectory using a population of chromosomes as opposed to grids, RBGAF is much more memory efficient than DP-SLAM. Memory efficiency is further enhanced through the exploitation of dynamic data structures for representing the maps and the robot trajectories. This makes the proposed RBGAF very suitable for large scale SLAM in 3D environments. Further, the proposed method's provision for adaptation of chromosome lifetime/group sizes and its ability to incorporate alternative map representations makes it adaptable to varied environments and different sensors. Simulation and experimental results obtained in an outdoor environment using a laser measurement system are presented to demonstrate the method's effectiveness.