Advanced controller design using neural networks for nonlinear dynamic systems with application to micro/nano robotics

The dissertation focuses on neural network (NN) control designs for nonlinear systems with application to micro/nano robotic. Critical problems in nano scale including thermal drift are also addressed. This dissertation is given in the form of several papers.; To start with, a suite of novel controllers is developed in the first paper for the manipulation of microscale objects in a micro-electromechanical system (MEMS). The proposed robust and the adaptive neural network controllers overcome the unknown contact dynamics and ensure their performance in the presence of actuator constraints.; Next, in the second paper, thermal drift, as the major source of uncertainty in nano scale, is discussed and compensated by using block based phase-correlation method. This consideration is needed to realize a truly automatic manipulation of nano objects.; Subsequently, the third paper uses the drift compensator from the second paper to develop a NN-based adaptive force design for nanomanipulation to accommodate the unknown dynamics, while maintaining a constant force applied on the nano sample.; In order to address the optimality in terms of a standard quadratic cost function, the fourth paper introduces a reinforcement learning-based controller for the nanoscale manipulation by considering the Bellman equation. This controller consists of an action network and a critic network. Both of the networks are trained in an online fashion with the updating algorithms derived from dynamic programming (DP).; To make our scheme applicable to a more general class of affine systems with immeasurable states, an output feedback design with an extra NN observer is introduced in the final paper while relaxing the separation principle. By using the Lyapunov approach, the stability of the above mentioned controller designs are demonstrated.