Reflex Research Group for Adaptive Systems SET

Most successful state-of-the-art robo-tic manipulators have the characteristic of producing precise, fast, smooth and reproducible movements. Their drawback is that they tend to have a limited repertoire which can only be extended by costly inverse kinematics calculations or direct teach-in sessions. The goal of our project is to develop a exible open world robot , which adapts to its task and environment progressively and in an iterative manner. It begins carefully and slowly, with small jagged movements , but, after repetition, reaches an acceptable level of smoothness and execution speed. It can be placed in new surroundings with new task deenitions, requiring only a further practice time before becoming expert. Similarities from other sub-tasks are recognized and may be transferred to the new domains. The basic constituents of the adaptive robot are a simple iterative inverse kinematics and driver programs. In this paper the idea of driver programs for robot arm control is introduced and their properties investigated.