Growing multi-experts network based controller design for realistic dynamic robot via simulation

In this paper, a nonlinear variable gain PPI cascade controller that exhibits self-constructing and self-adaptation capabilities is presented. The conventional independent joint PPI cascade controller is augmented with a nonlinear variable PI gain control signal using a dynamic structural network. The proposed approach enhances the adaptability of conventional PPI cascade controller. The other attractive feature is that using dynamic network structure in neural control design results in a much smaller network, effective utilization of basis function with reduced storage and computational requirements. A commercial realistic dynamic simulator EasyRob is used as the experimental testbed