Abstract: The real-time terrain modeling problem of mobile robot with a laser scanner in outdoor unstructured 3D environments is studied. The underlying uncertainties from multiple sources during modeling are taken into account and modeled as zero-mean Gaussian noises, and subsequently the multi-level coordinate transformation matrixes are created to convert the measurements from laser scanner into probabilistic elevation estimations in the global coordinate systems, which will be associated with several terrain cells according to the confidence interval of the estimation. The elevation estimations assigned to each cell can be fused through a probabilistic approach to update the map locally. In addition, a local measurement window is defined to detect the occlusions, and the 3D localization of the mobile robot in outdoor environment is solved simultaneously. Experimental results demonstrate the real-time performance and effectiveness of the proposed method.