Oscillators and crank turning: exploiting natural dynamics with a humanoid robot arm.

This paper presents an approach to robot-arm control that exploits the natural dynamics of the arm. This is in contrast to traditional approaches, which either ignore or cancel out arm dynamics. While the traditional approaches are more general, they often result in systems and robot designs that are not robust. The alternative approach gives systems that are computationally simple, robust to variation in system parameters, robust to changes in the dynamics themselves, and versatile. The approach is examined using the example of a compliant robot arm, controlled by independent neural oscillators, in a crank-turning task. A model is constructed, and the robot behaviour compared with the model. These data show that the arm-oscillator system is exploiting the natural dynamics by finding and exciting the resonant mode of the underlying mechanical system. Since this is a natural behaviour of the system, the robot behaviour is robust. The paper concludes by discussing the opportunities and limitations of this approach.