In a complex and uncertain environment, using a single sensor on the robot localization is poor in accuracy and reliability, and susceptible to the interference. For this problem, the mobile robot SLAM algorithm based on the particle filter is improved for use of multi-sensor fusion algorithm. The new algorithm fuses observe information on the feature level to take advantage of redundant information collected by various sensors, and the fusion information of observations were used to estimate the posterior probability distribution of robot path and the environmental characteristics. The simulation results show that the improved algorithm in accuracy and reliability of SLAM has greatly improved, and demonstrated the feasibility of the methods.