Decentralized variable structure control of robotic manipulators: neural computational algorithms

The authors describe a decentralized variable structure control system where a multilayer backpropagation neural network is used to generate the required control signals given the deviation from the sliding line and the state of the system on the phase plane. An implementation of the neural network controller to accommodate adaptive selection of sliding manifold parameters together with control gains is proposed. Results of some simulation experiments performed to illustrate the performance improvements due to these adaptive implementations are given for regulation tasks. A plot of the error profiles for the entire trajectory in the cases considered demonstrates the superior performance features of this algorithm.