In recent years, with the increasingly complex robot application environments, the research of robotic autonomous obstacle avoidance motion planning is one of the key technologies to improve its intelligence. In this paper, we present an improved algorithm based on Rapidly-exploring Random Tree (RRT) algorithm, which is called Probability Adaptive RRT (PARRT). We analyze the defects of the original algorithm in detail from the perspective of Voronoi diagram. For some complicated occasions, especially when the robot is surrounded by obstacles, the efficiency of the original algorithm will be greatly reduced. For this problem, we proposed an expansion probability adaptive method. By dynamically adjusting the extending probability of nodes near obstacles, the invalid node expansion is reduced effectively and the planning speed is greatly improved. Finally, we experiment in both two-dimensional space and the joint space of manipulators. The planning results show the improved effect of the proposed algorithm.