Abstract: In order to resolve the withdrawal problem of small ground mobile robots in natural environment, the 2D laser radar is used for environmental perception and terminal precise navigation, and an automatic withdrawal system is designed and implemented. Firstly, an adaptive curvature filtering algorithm is proposed to process the radar data in advance. Then, the scattered data is clustered in the process of detecting the guiding target according to the density and the nearest neighbor measurement. The target detection and matching are carried out by using the geometric structure constraint between the guiding target and the auxiliary target. Finally, the real-time obstacle avoidance and the optimal forward direction selection are conducted based on the method of affected layer division and candidate direction estimation. Thus the robot is accurately guided to accomplish automatic withdrawal. The automatic withdrawal system is validated in natural environment. And some aspects of the automatic withdrawal system are analyzed experimentally. Experimental results show that the proposed method can effectively accomplish the automatic withdrawal of the small ground mobile robot in natural environment.