This paper presents a real-time approach to generate the path of robot manipulator. In this approach, octree model is used to express environment and detect collision. The arms of robot in space are expressed by a set of points on the surface of them. Then the distance between manipulator and obstacle is estimated by the potential generated by octree modeling. Based on the distance estimated, graph search is executed to find a path. The effectiveness of the approach is proved by simulations and the experiments realized by a FANUC robot, LR Mate 200iB. At last, the calculation time of proposed approach is compared with the result of typical graph search approach. The result shows that our approach is faster. It can act as real-time path planner. Whereas the approach of typical graph search is slow for real-time path planning