To meet the requirements of indoor service robot working under unknown dynamic environments, an incremental sampling path planning based on local environments is proposed in this paper. At first, the estimation of collision risk in current environment is built by a probabilistic model. Then, during the searching tree expansion process, a novel cost function using the Euclidean distance and estimation of collision risk is constructed. Thus, the collision checking for new vertex and potential extensible edges in each iteration can be reduced, and then the algorithm efficiency can be increased. Meanwhile, the best extension in current structure of searching tree can be obtained by referred the rapidlyexploring random graph algorithm. In addition, the performance analysis is provided. Finally, the simulations and experimental results show that the proposed algorithm owns good planning performances and efficiency (less calculating time and iteration times) respectively, which satisfies the needs of real time path planning for indoor service robot.