Improved Occupancy Grids for Map Building

Occupancy grids are a probabilistic method for fusing multiplesensor readings into surface maps of the environment. Although theunderlying theory has been understood for many years, the intricacies ofapplying it to realtime sensor interpretation have been neglected. Thispaper analyzes how refined sensor models (including specularity models) andassumptions about independence are crucial issues for occupancy gridinterpretation. Using this analysis, the MURIEL method for occupancy gridupdate is developed. Experiments show how it can dramatically improve thefidelity of occupancy grid map-making in specular and realtimeenvironments.