To address the time-varying observation noise and particle position distribution on simultaneous localization and mapping (SLAM) accuracy in particle filter SLAM (PF-SLAM) for simultaneous localization and mapping of mobile robots, this article proposes an adaptive PF-SLAM algorithm based on variational Bayes, which adopts a Gaussian mixture model to formulate the time-varying observation noise and iteratively estimates the unknown parameters in the mixture model by using a variational Bayesian method. Meanwhile, the particles are divided into fixed particles and optimized particles according to the particle weights, and the particle positions are adjusted by the topological position distribution relationship between two particles, which handle the time-varying observation noise and optimize the particle position distribution. In this way, the optimized particle set could represent the robot position probability distribution and realize the adaptive observation noise and particle position distribution. Compared with the traditional PFSLAM algorithm, simulation results show that the positioning and map building error of this algorithm is reduced by 76. 45% . Compared with the traditional PF-SLAM algorithm, the actual experiments show that the environmental contour error of this algorithm is reduced by 61. 87% . It effectively improves the state estimation accuracy of mobile robot and provides a new reference for mobile robot real-time positioning and map construction.