A new multi-robot system framework and a novel kind of robot control structure are introduced to solve the problem of multi-robot map building under entirely unknown environment Firstly. In order to deal with a mass of data, distributed fusion framework is adopted in this multi-robot system. Compared with the classical architectures, this multi-robot system enhances overall system performance, especially real-time performance. Then a few general basic belief assignment functions (gbbaf) and a hybrid DSm model of sonar are constructed to deal with the uncertain and imprecise, sometimes even high conflicting information, which is obtained by sonar sensors with the application of new information fusion method Dezert-Smarandache theory (DSmT). DSmT is extended from Bayesian theory and Demster-Shafer theory (DST) in the system and consideration of characteristics of sonar sensors. Finally, Pioneer II mobile robots are used to carry out experiments of map building for both single-robot system and multi-robot system. And a 3D general basic belief assignment map is structured by openGL. The comparison of created ichnography with the real map testifies the validity of hybrid DSm model and efficiency of multi-robot system for fusing imprecise information and map building proposed in this research.