Aimed at the problems of low precision, large amount of calculation and severe sample degeneracy of simultaneous localization and mapping(SLAM), this paper presented a distributed unscented marginalized particle filter(DUMPF) algorithm based on the combination of the distributed unscented particle filter(DUPF) with the marginalized particle filter(MPF). In the proposed method, the SLAM system was divided into several subsystems according to the distribution algorithms. The unscented particle filter(UPF) was used in each subsystem to estimate a part of the states. The marginal distribution of the UPF was optimized to reduce the computational complexity. The estimated results of the subsystems were transmitted to the master filter to obtain the final result. The simulation results showed that the improved DUMPF could prevent the particle degeneration problem, and had a higher precision and a smaller computational complexity.