This paper discusses the related problems existing in the costmap, and proposes a new algorithm for the generation of indoor map. Firstly, for the processing method of dynamic obstacles by radar, a new real-time updating method of dynamic obstacles is studied, which realizes the accurate positioning of obstacles, and improves the matching of costmap and indoor environment. Then, the algorithm realizes the classification and separate processing of obstacles by preprocessing static obstacles, greatly reducing the cost in the process of map updating, the calculation speed of costmap is increased in a disguised way, and on this basis, a complete multi-layer costmap is constructed using the equidistant expansion method. Finally, the algorithm is tested on the autonomous robot platform. The experimental results show that the proposed algorithm has a significant improvement in matching compared with the classical costmap, and an average increase of 77.6% in map update speed.