Position probability grids for mobile robots obtained by convolution

The paper presents an approach to use relative sensor information for position estimation in an absolute position probability grid. Here relatively measuring sensors are the odometry and nine narrow beam infrared sensors with nonlinear characteristics mounted on a mobile robot. An inaccurate indoor GPS sensor is available for absolute position data. However, for the best position estimate all these sensors have to be considered. The data fusion can only be done with comparable data. Therefore, the relative sensor information is transformed into absolute position information by convolution and represented as individual position probability grids. To determine the resulting position of one robot these grids are combined according to Bayes theorem.