Abstract: Influenced by turbulence, the practical odor plume in ventilated indoor environment is fluctuant and intermittent. Bigger eddies can easily lead to longtime local concentration maxima in some corners. In addition, the obstacles can also change the plume distribution. These make the mobile robot based odor source search quite complicated. A probability-fitness-function based particle swarm optimization (P-PSO) algorithm is proposed and used for multi-robot based odor source search in ventilated indoor environment with obstacles. The P-PSO algorithm uses probability instead of a definite number to express the value of fitness function. For the odor source search problem, the fitness function is expressed by the odor source probability estimated by Bayesian inference combined with variable-universe fuzzy inference. To validate the proposed search strategy, different odor plumes corresponding to the real boundary conditions of an indoor environment are set up. Simulation results demonstrate the feasibility of the P-PSO algorithm for solving the odor source localization problem.