Map spread factor based confidence weighted average technique for adaptive SLAM with unknown sensor model and noise covariance

Abstract

This investigation presents an adaptive simultaneous localization and mapping (SLAM) system using sensor fusion based on confidence weighted average technique. The confidence weights for the sensor data are adapted based on instantaneous sensor accuracy evaluated during robot navigation. To this extent, a performance metric called the map spread factor is formulated which is based on the mismatch between the past and present map retranslated using sensor measurements on robot location. As this metric evaluates the sensor performance without any prior knowledge of its characteristics based on the maps acquired from the scanner the method is independent of the type of sensor employed. Our experiments demonstrate the accuracy of the proposed approach over traditional extended Kalman filter based SLAM.