We compare the effectiveness of two widely used filters for nonlinear systems, i.e., the Extended Kalman Filter (EKF) and the Unscented Kalman Filter (UKF), in reconstructing the unknown environment where a mobile robot moves. The reconstruction is obtained by a novel cells-covering algorithm that only uses the distance measurements taken from the robot's on-board sonar sensors. We show that, despite the superior theoretical properties of the UKF, both filters perform comparably well, and that the proposed algorithm provides good localization performance and a reliable environment reconstruction.