Abstract: For large-scale unknown environments, a method of topological node matching based on visual feature is presented, and a local scan matching strategy is integrated to realize map merging for multi-robot system under RTM (robot technology middleware) framework. A main-auxiliary structure model of multiple robots is developed, and an improved SP²ATM algorithm is adopted to incrementally constructing topological map in unknown environments. Based on this, the hierarchical topology structure including SIFT (scale-invariant feature transform) feature information is presented, which is combined with ICP (iterative closest point) algorithm to realize map merging of multi-robot systems. RTM is taken as communication platform to improve the realtime performance, flexibility and robustness of the system. Simulation on USARSim and experimental results in actual environments verify the effectiveness of the proposed method.