Simultaneous localization and map building of skid-steered robots

In this article, an alternative scheme is proposed for the SLAM problem, where the state vector holds only the robot pose and the map is feature-based in the form of line segments. It is a lighter representation in comparison to occupancy grids and point maps, something necessary as large scale environments are addressed. The basic idea is to obtain one estimate of the robot's pose from an innovative dead-reckoning scheme and one from a laser scan matching algorithm. The fusion is done through a covariance intersection filter, avoiding the assumption that holds in Kalman filter that the cross correlations are zero. In this way, the uncertainty of pose estimate while the cross correlations are unknown and non zero are minimized. In the following section, the general framework of the SLAM algorithm is described. Then a short description of the dead-reckoning scheme, feature extraction scheme and the map matching algorithm are discussed. The loop with the covariance intersection fusion of the robot pose estimates are concluded. The description of the global map update algorithm and the results from representative experiments in real world experimental conditions are presented