View article

In corpus-based robotics, the primitive functions built into a robot are determined by the functional content of human utterances spoken to the robot. In the example of route instructions treated in this paper, the 15 primitives found include functions such as turn (), cross (), landmark_is_located (), etc. These are natural primitives of human behaviour but complex robot functions, some of which are not normally thought of by roboticists. Primitives must cope robustly with a variety of environmental conditions and require autonomous navigation capabilities based for instance on visual landmark recognition and localisation, navigable space mapping and path planning. Thus, the requirement of human-robot interaction creates specific and demanding functional targets for robot designers. A major obstacle to human-robot communication lies probably in current robots’ perception capabilities. Furthermore, for human-robot communication to match the performance of human-human communication, the robot must also be provided with capabilities of re-interpreting and correcting instructions at execution time. Such a large autonomy raises wider issues of safety and control.