Primitive skill-based supervisory robot control applied to a combined visual-acoustic localization task

The rising complexity of robot tasks in the public, private or industrial environment requires the introduction of a new generation of "humanoid" robots provided with the capability to interact with its environment. Therefore a flexible discrete-continuous supervisory control concept in combination with an intelligent multi-sensor fusion is needed. The functionality of the robot control architecture proposed in this paper captures both the hierarchy required for representing complex tasks using the notion of Primitive Skill as well as the mechanisms for detecting changes in the environment thanks to a multi-sensor supervision. The task can be thus adapted in order to face the detected unexpected situations. As case study an algorithm for the combined visual and acoustic localization of fallen objects based on the estimation of time-difference of arrival is presented and the delivered information is used to dynamically adapt the discrete plan of two robot arms performing a pick and place task.