Adaptive Observation Covariance for EKF-SLAM in Indoor Environments using Laser Data

In this paper we describe an approach to concurrently localize a robot and to build a feature based map using laser sensor. Stochastic simultaneous localization and mapping (SLAM) is performed by storing the robot pose and map landmarks in a single state vector, and estimating this state vector via a recursive process of prediction and updating. In our case, these estimates are updated using an extended Kalman filter (EKF). The main novelty of this proposal is the development and test of an adaptive measurement covariance matrix that permits to include close and distant features in the updating stage of the EKF-SLAM algorithm, providing more precision to closer detected features