ABSTRACT
In Chapter 4 were given several techniques for the design of neural network (NN) controllers for rigid-link robotic systems. In Chapter 5 were given NN controllers for complex practical robotic systems including force control, flexible-link robots, flexible-joint systems, and systems with high-frequency actuator dynamics. The NN controllers relied on a filtered-error approximation-based approach, and the weight tuning algorithms included a simple backpropagation-type method that works in an ideal case, and modified tuning algorithms that work in general cases that include disturbances. Two sorts of NN were considered— functional-link NN (FLNN) that are linear in the tunable weights, and two-layer NN that are nonlinear in the firstlayer weights V. It was shown how to overcome the linear-in-the-parameters (LIP) restriction of standard adaptive control approaches.
