In order to improve the safety and manipulation performance for telerobotic system, a kinesthesis teaching control strategy for telerobotics was presented. Utilizing the characteristics of human’s sensitivity to force information as well as the specific feature of vision’s forecast and guidance, the control force of the operator with the virtual forces was merged, including the attraction force from objectives, and the rejection force from obstacles, which were generated from the recognition results of working fields based on vision techniques. Furthermore, it combined the advantages of machine intelligence with human’s high-level decision, which could improve effectively the working efficiency as well as local autonomy of tele-robot, and reduce the proficiency requirements of operator. Especially, in a teleoperation system with large time delay, the method could avoid the blinded manipulation of robot caused by lagging transmission of image. The system structure, extraction of the operating environment spatial information and control methods were introduced in detail. The superiority of the proposed control strategy in improving the operating efficiency and security of the telerobotic system were verified by relative experiments.