Data fusion design for skill transfer systems

Abstract

In modern robotics, the robot is seen not as a mechanical unit, but instead as an intelligent unit, with planning and cognitive structures that are capable of making intelligent decisions and of helping human beings in high-level and interactive tasks. A new generation of robots is ready to emerge: human-oriented robots. In the near future, they should be able to perform perfect human-robot symbiosis, such as helping disabled people in their basic day-to-day problems. A key issue to accomplish this goal is the development of robust skill transfer systems, in order to teach some basic tasks in a natural way. This paper describes the design of a data fusion module for skill transfer applications. It consists of two independent modules for optimal fusion and filtering. A different interpretation of the Kalman filter equations is made, in order to achieve a "model-free" equation that is capable of following arbitrary variables. The presented fusion algorithm is global, and could easily be extended to any arbitrary system. It was successfully tested in the Institute of Robotics and System Dynamics at the German Aerospace Centre (DLR).