3D laser SLAM in outdoor large-scale scenes has been a challenging problem in the current field of robot navigation and environment sensing. Due to the instability of GPS signals in some environments and the error accumulation property of laser SLAM, traditional algorithms perform poorly in large-scale scenes. In order to solve the problems of error accumulation of 3D laser SLAM (simultaneous localization and mapping) in outdoor large-scale scenes, this paper proposed a GPS-Laser-IMU fusion mapping algorithm based on iterated Kalman filter, which improves the mapping accuracy and robustness. The algorithm used IMU to predict the robot state, while laser and GPS data were used as observations to update the robot state. The measurement equations and the related Jacobian matrix were derived. Integrating the absolute position information of GPS data in odometer can solve the problem of error accumulation in long-time operation. In the environment with sparse features, the algorithm may collapse due to insufficient constraints. The introduction of GPS can improve the robustness of the system. Additionally, gravity plays a crucial role in the prediction of robot state. Although gravity is a three-dimensional vector, the module length of gravity remains constant when the region does not change, so gravity was treated as a two-degree-of-freedom vector. Transforming gravity optimization into optimization on SO(3) successfully avoids over-parameterization issues, thereby improving precision. Its performance was compared with that of other algorithms in outdoor environments and the robustness and accuracy were verified in large scale scenes. The results show that the root mean square error of this algorithm is 0.089 m, which is 54% lower than that of other algorithms.