We present a preliminary project for a remote control in hostile environment. The main constraint is represented by the communication time, that we evaluated of two orders higher than the inner clock of the control system. We overcome this difficulty giving a short-term autonomy to our system. Namely, we configure the control system as a neural network on board of the robot that reacts to external stimuli. A copy of the network is in the remote control station. Here a workstation periodically identifies the system dynamics, the robot environment and the optimal control strategy. The strategy is translated in a proper selection of the free parameters of the neural network. The values of these parameters are transmitted as remote control action to the robot neural network. The feasibility of this project is discussed in concern of a very simple robot running in the Oil and Gas University of Ploiesti and usual training techniques operating in the University of Milan. Those universities will be the remote control terminals.