Evaluation of a neural network for fault-tolerant, real-time, adaptive control

A number of simulations of the cerebellar-model-arithmetic-computer- (CMAC-) based control scheme have been conducted to evaluate its ability to control a nonlinear dynamic system, compensate for erroneous input information, and respond appropriately to previously unlearned inputs. The simulations were carried out on a VAX station II using the C programming language. The goal was to evaluate the neural network´s potential as a controller for artificial limbs or limb-assist devices. The simulation results demonstrate the capabilities of this type of controller in a situation where inputs are possibly erroneous or incomplete or have a wide range of variability, thus confirming its suitability for the intended application