The paper is devoted to the issues of teleoperation of ground mobile robots. Problems of moving a robot in an un-structured environment by commands of a human operator are considered. A significant problem that reduces a quality of control and often leads to loss of stability is time delays that occur in information channels of the complex. To partially compensate for the negative impact of the time delays, an approach based on the prediction of the local goal of movement, the real position of the robot at the time of commands formation and the model of an operator is proposed. To test the approach, a computer simulation of the robot control process was performed on the basis of a training complex based on the Unity engine. The task was consisted in controlling the movement of the robot along the reference trajectory displayed on the screen. The task execution time and the similarity of the recorded trajectory with the reference one were evaluated. The experimental results confirmed the positive effect of the proposed compensation method on the efficiency of the control system.