Neuro-Fuzzy Inverse Forward Models

Internal cognitive models are useful methods for the implementation of motor control [1]. The approach can be applied more generally to any intelligent adaptive control problem where the dynamics of the system (plant) is unknown and/or changing. In particular, paired inverse-forward models have been shown successfully to control complex motor tasks using robotic manipulators [1]. One way this is accomplished is by training a pair of multi-layered neural networks to learn both the forward model of the plant dynamics and an inverse controller model simultaneously using an extension of the backpropagation algorithm. This paper explores a variation of the traditional multi-layered network used for teaching an inverse-forward model pair (IFMP). We investigate the use of a simple fuzzy-neural system for implementing both models.