In order to solve the problems of traditional Rao-Blackwellized particle filter (RBPF) inspection robot mapping method in the unstructured environment, such as low calculation accuracy and large calculation amount, an improved RBPF substation inspection robot mapping method based on point cloud matching is proposed in this paper. Firstly, based on the RBPF method, the subgraph construction strategy is designed, and the Adaboost learning algorithm is introduced to identify adjacent subgraphs. Secondly, in view of the low splicing accuracy of traditional point cloud matching methods when splicing only partially overlapping sub-images, the NDT and ICP algorithms are combined to design a matching algorithm to solve the relative poses between sub-images. Thirdly, according to the relative poses between the sub-graphs, the global pose of the sub-graphs is optimized through the general graph optimization (G2O) algorithm to obtain a complete substation map. Finally, the Gazebo software is used for simulation to compare the mapping effects of different methods. The results show that the improved RBPF method can reduce costs of hardware, improve the accuracy of mapping in the complex environment of substations, and can provide a reference for the design of unattended substations.