Whilst industrial robots draw their efficiency from specialising on highly structured working environments, the technological challenge behind realising service robots suited for everyday use lies in the intelligent, physical support of humans in as broad a spectrum as possible and in low-structured everyday environments.

Long-term sustainable and acceptable developments must be consistently aligned to the basic conditions created by humans when designing working and everyday environments. Robots must therefore reflect human features both in terms of their hardware (physical build, head, hands) and their software (sensors, language, cognition and ability to learn). On the one hand, this ensures high acceptance from the users and, on the other, the ability of the robot to come to terms with the natural environment of humans and to handle the tools designed for humans. Ultimately, this is the basis for successful, cooperative activity between humans and robots in everyday environments.

The key to reaching this ambitious goal lies in developing an ability to learn that is consistently geared towards the features robustness, adaptability and ease of passing on instructions. One of the issues guiding the research work is two-handed manipulation through an anthropomorphic robot. This ability combines high application potential and the gaining of knowledge alike.

The innovative, methodical basis is the development of action-based representations for manipulation processes and their connection with biologically inspired, high-performance learning and exploration methods aimed at attaining robustness and an ability to adapt to changing tasks. In order to design the interaction with the service robot naturally (and, in doing so, to ensure high acceptance among users), the robot must be in a position to recognise natural language instructions and gestures and to interpret them in the right situation. The provision of a visuo-linguistic front end is therefore the prerequisite on the one hand for learning - on the basis of imitation - new partial abilities that have not been pre-programmed and, on the other, for the targeted intervention of a human tutor whenever error situations occur.

• elementary, bimanual manipulation (e.g. opening screw tops) in a variety of situations
• flexible, bimanual interaction and learning with a tutor in new situations
• integration of action sequence detection with grasping actions
• integration of dialogue and action sequence recognition with grasping actions