Abstract: We propose an adaptive robust loop closure algorithm for the Graph SLAM to address the problem where the back-end for conventional Graph SLAM obviates the influence of false loops efficiently. The influence of indefinite parameters in the cost function to the optimization procedure is analyzed. The parameters are renewed by the latest information obtained from iterations to speed the convergence rate. The algorithm is adaptive to different datasets. The experiment is performed for the proposed algorithm with public datasets. The comparison results show that the proposed algorithm is adaptive to different datasets with different types and numbers of outliers and the convergence rate is higher, which verifies the efficiency of the algorithm.