Abstract: To address the challenges of insufficient positioning accuracy and weak environmental adaptability in current wall-climbing robot localization, an external positioning system based on pan-tilt laser tracking is proposed. This system utilizes a zoom camera to track AprilTags for controlling the motion of a two-axis pan-tilt mechanism and fuses data from a laser rangefinder to calculate the three-dimensional coordinates of the robot. To address the issue of non-Gaussian outlier noise caused by the temporary deviation of the laser spot from the marker during dynamic tracking, a smooth robust extended Kalman filter (SREKF) algorithm based on the Sigmoid function is proposed. This algorithm constructs a Sigmoid weighting factor using the Mahalanobis distance of the observation innovation to adaptively adjust the observation noise covariance matrix, thereby smoothly suppressing outlier interference. Experimental results demonstrate that the system achieves stable localization on a 6m×2.5m steel structure surface. Compared with the standard extended Kalman filter (EKF) and Huber robust extended Kalman filter (HREKF), the positioning accuracy of the SREKF algorithm in complex motion modes is significantly improved, with the root mean square error (RMSE) reduced by 1.40 cm and 0.38 cm, respectively. The overall RMSE is better than 0.02 m, and the maximum positioning error is less than 0.08 m. The results indicate that this method effectively suppresses the interference of outlier noise on localization accuracy and improves system precision, providing a highly reliable position reference for robot operations in large-scale steel structure environments.