Abstract: SLAM (Simultaneous Localization and Mapping) and path planning are two important research directions in the field of robotics. How to explore an entirely unknown dynamic environment efficiently is a difficult problem for intelligent mobile robots. In this study, a new method of information fusion i.e. DSmT (Dezert-Smarandache Theory) which is an extension of DST (Dempster-Shafer Theory) is introduced to deal with high conflicting and uncertain information and then multi-agent robot system with GREM (Generalized Evidence Reasoning Machine) based on DSmT is presented for mobile robot’s SLAM and efficiently planning smooth paths in unknown dynamic environment. The single robot is treated as a multi-agent system and the corresponding architecture combined with cooperative control is constructed. Considering the characteristics of sonar sensor, the grid map method is adopted and a sonar sensor mathematical model is constructed based on DSmT. Meanwhile a few of gbbaf (general basic belief assignment functions) are constructed for fusion. In order to make the A* algorithm which is the classical method for the global path planning suitable for local path planning, safety guard district search method and an optimizing approach for searched paths are proposed. Finally, SLAM and path planning experiments are carried out with Pioneer 2-DXe mobile robot. The experimental results testify the validity of hybrid DSm (Dezert-Smarandache) model under DSmT framework for fusing imprecise information during map building and also reveal the validity and superiority of the multi-agent system for path planning in unknown dynamic environment.