In this article, robust topology optimization under geometric uncertainty is proposed. The design domain is discretized by an extended finite element method. A bi-directional evolutionary structural optimization carries out the optimization process. The performance of the proposed method is compared with the Monte Carlo, solid isotropic material with penalization, perturbation and non-intrusive polynomial chaos expansion methods. The novelty of the present method lies in the following three aspects: (1) this article is among the first to use the extended finite element method in studying the topology optimization under uncertainty; (2) by adopting the extended finite element method for boundary elements in the finite element framework, there is no need for any remeshing techniques; and (3) the numerical results show that the present method has a smoother boundary region and minimum value of the mean and standard deviation of compliance than the other methods, in particular mesh size.