Robust control of dynamic systems using neuromorphic controllers: a CMAC approach

The good performance of trainable controllers based on neuronlike elements hinges on the ability of the neural network to generate a `good´ control law even when the input does not belong to the training set, and it has been shown that neural nets do not necessarily generalize well. The author addresses this problem by proposing a feedback controller based on the use of a CMAC (cerebellar model articulation controller) neural net. It is shown that the proposed controller has good generalization properties. Moreover, by proper choice of the training set the resulting closed-loop system is guaranteed to be robustly stable with respect to model uncertainty