Abstract: In order to locate the underwater robot in the pools of the nuclear power plant, the scan sonar is used. First, the signal characteristics of the scan sonar are analyzed, and the preprocessing method of sonar is used to reduce signal interference and eliminate redundant data, and the computational efficiency is improved by preprocessing of the threshold denoising, distance limitation and reduction of sampling. Then, the probabilistic iterative correspondence algorithm is proposed based on the measurement noise of the sonar. The nearest matching points between sonar image and the map of the pools of the nuclear power plant are computed by the Mahalanobis distance. Meanwhile, the degree of confidence is used to improve the matching accuracy, and the absolute position and orientation of underwater robots is estimated by the optimization iterations. The algorithm is compared with the traditional iterative closest point algorithm and the results show that the proposed algorithm improves the estimation accuracy of underwater robots. Finally, experiments carried out in the pool verify the effectiveness of the proposed algorithm.