Abstract: The virtual system for teleoperating Stewart platform based on VR (virtual reality) technical was designed. The principle of the drive arithmetic for the virtual model was instructed. In this arithmetic, firstly proper coordinate system was established, and then a series of coordinate transform was adopted, so the movement correspond problem was solved. The neural network based on L-M training algorithm was used to solve the forward kinematics problem of the platform, and proved that the neural network had a very good generalization performance by experiment. Based on bounding volume hierarchies, the axis-aligned bounding boxes was selected to detect the collision in the virtual simulation system. A sort of simple control strategy was used to overcome the stability of the system because of the time-delay of the communicate link. In the local robot control system, the discrete form of tracking-differentiator was selected to ensure that the robot has good stability and well dynamic performance. The virtual system was validated to be effective by experiment finally.