Title:SMAP: Simultaneous Mapping and Planning on Occupancy Grids

Abstract:Occupancy grids are the most common framework when it comes to creating a map of the environment using a robot. This paper studies occupancy grids from the motion planning perspective and proposes a mapping method that provides richer data (map) for the purpose of planning and collision avoidance. Typically, in occupancy grid mapping, each cell contains a single number representing the probability of cell being occupied. This leads to conflicts in the map, and more importantly inconsistency between the map error and reported confidence values. Such inconsistencies pose challenges for the planner that relies on the generated map for planning motions. In this work, we store a richer data at each voxel including an accurate estimate of the variance of occupancy. We show that in addition to achieving maps that are often more accurate than tradition methods, the proposed filtering scheme demonstrates a much higher level of consistency between its error and its reported confidence. This allows the planner to reason about acquisition of the future sensory information. Such planning can lead to active perception maneuvers that while guiding the robot toward the goal aims at increasing the confidence in parts of the map that are relevant to accomplishing the task.