A neuro-fuzzy-based system architecture for the intelligent control of multi-finger robot hands

In this paper, a new system architecture for the intelligent control of multi-finger robot hands that can operate in changing environments is presented. The conception of the control system is based on the combination of a neural network approach for the adaptation of grasp parameters and a fuzzy logic approach for the correction of parameter values given to a conventional controller. Typical tasks of dexterous hands are fine manipulation and exploration, what demands task planning and motion as well as force control capabilities. Therefore, a planning component determines initial manipulation parameters whereas a neuro-system level performs continual computation of suboptimal grasp forces and online learning of inference rules used on a fuzzy system level for parameter adjusting