Neuron variable structure controller

The problem of nonlinear adaptive joint controller design using a proportional feedback structure is addressed. The design is considered in the context of the decentralized control of manipulators where ease of implementation and reliability are desirable characteristics. The design incorporates a nonlinear sigmoidal loop gain characteristic within a conventional proportional feedback structure. Proper selection of sigmoid parameters can significantly improve tracking performance in comparison with conventional designs using a constant loop gain. A methodology for selecting and adaptively updating sigmoid parameters is presented. The adaptive updating scheme is implemented with shunting short-term memory neurons. The implementation exploits the nonlinear response characteristics of these neurons in a learning process which updates the sigmoid parameters. The design is illustrated for a direct-drive joint control