The use of an acoustic speech recognition in the industry for controlling/handling a robot entails risks. Commands can be misinterpreted from the system in case of high backgroundnoise. In this master thesis a visual speech recognition system is developed. The lips of people are read by image processing. Short commands are recognised by an assistant robot. The visual speech recognition system can be used to support an acoustic speech recognition system. In the evaluation, it is shown that the system can reliably recognise short commands.