Simultaneous Localization and Mapping (SLAM) technology is one of the research hotspots when robots perform tasks in unknown unstructured environments. SLAM can be performed by a single robot in small-scale and simple scenarios. With the emergence of complex application scenarios, it is necessary for multiple robots to obtain more comprehensive and real environmental data through collaborative perception of the environment, and then build real-time maps. Multi-robot cooperative SLAM can fuse multi-source heterogeneous sensor data of different robot configurations by means of graph optimization, so as to obtain its own pose and surrounding environment information. Compared with single-robot SLAM, it has stronger fault tolerance, flexibility and robustness. In this paper, based on the graph optimization theory, a collaborative SLAM method based on factor graph is designed to solve the multi-robot SLAM problem, and the global consistent map is generated according to the factor graph optimization algorithm with Kalman filter. Finally, in the environment built by Gazebo, the simulation experiments of single robot and multi-robot are compared to verify the feasibility and effectiveness of the designed algorithm.