For the teleoperation of a robot over an IP network, we must deal with the variable network delays as well as with events that can occur during the teleoperation. In this paper both control and architectural aspects are exposed. We first present a stable remote position/force control scheme for manipulator robot, based on a delay regulator and Smith's predictor principle. Some experiments show the improvement obtained by this method. However to face very important network delay variations and possible connection rupture, we designed a supervised delay regulator based architecture. This architecture, embedded on a mobile robot, is presented and results of experiments are also given and explained. All these experiments, performed on long distance (Mexico-France: ∼8000 km), show the improvement obtained on both themes, that conjointly enhances and secures the teleoperation through the Internet.