Abstract: Submap and adaptive covariance based 2D SLAM solution can not only achieve efficient loop-closure detection but also accurate localization．Firstly,the loop-closure is detected by efficiently matching 2D geometric features between local submaps．Unlike the previous methods which often use the number of the measure frames as the criteria of the division,we employed the number of features as the main criteria．To achieve accurate localization,we proposed an adaptive Kalman filter to estimate the final pose．Moreover,the prediction and observation covariance are adaptive and estimated by the scan-matching algorithm．Finally,if a loop-closure is detected,the optimized transformation and covariance from the backend can be fused directly in the Kalman filter．In the first experiment,the comparison between the two kinds of submap division mechanism verifies the validity of the proposed method．The second experiment shows that the proposed method can accurately localize the robot only using a single lidar.

Key words: 2D mobile robot,Submap division,Adaptive covariance,Loop closure detection,Localization