Teleoperation is widely used for unmanned ground vehicle (UGV) navigation in military and civilian fields. However, the human operator has to limit speed to ensure the handling stability because of the low resolution of video, limited field of view and time delay in the control loop. In this paper, we propose a novel guidance point generation method that is well suited for human–machine cooperative UGV teleoperation in unstructured environments without a predefined goal position. The key novelty of this method is that the guidance points used for navigation can be generated with only the local perception information of the UGV. Firstly, the locally occupied grid map (OGM) was generated utilizing a probabilistic grid state description method, and converted into binary image to constructed the convex hull of obstacle area. Secondly, we proposed an improved thinning algorithm to extract skeletons of navigable regions from binary images, and find out the target skeleton related to the position of the UGV utilizing the k-nearest neighbor (kNN) algorithm. The target skeleton was reconstructed at the midline position of the navigable region using the decreasing gradient algorithm in order to obtain the appropriate skeleton end points for use as candidate guidance points. For visually presenting the driving trend of the UGV and convenient touch screen operation, we transformed guidance point selection into trajectory selection by generating the predicted trajectory correlative to candidate guidance points based on the differential equation of motion. Experimental results show that the proposed method significantly increases the speed of teleoperated UGV.