Dealing with uncertain measurements in virtual representations for robot guidance

Abstract

When measurements are collected with real sensors, it is known that they are always subject to some level of uncertainty due to the imperfection of practical systems. Often, it might be acceptable to neglect the uncertainty of measurements in order to simplify the development. But such a simplification can also result in critical situations under some specific circumstances such as the building of virtual maps for robot guidance. Erroneous data can easily lead to collisions between the robot and its environment which might result in important damage both to the robotic system and its surroundings. This paper presents an analysis of how uncertainty originating from multiple sources including uncertain spatial relationships in a robotic workcell can be propagated up to the scene representation. Various ways are studied to propagate the uncertainty over a chain of geometrical transformations in 3-D space and an approach is proposed to merge the propagated uncertainty into an occupancy grid that models a cluttered environment.