Neural network based predictive control for time-delayed teleoperation

The goal of teleoperation control architectures is to maintain stability while achieving a desired performance, known as transparency. The existence of communication channel time delay and environment dynamic uncertainties create a significant challenge in the design of stable transparent teleoperation controllers. This thesis is concerned with the design of novel predictive control strategies that compensate for the effect of delay for improved stability and transparency.;An early control methodology for time delayed plants is the Smith predictor, in which the plant model is utilized to predict the non-delayed output of the plant and move the delay out of the control loop. Recent Smith predictor based teleoperation control architectures have used linear or fixed-parameter dynamic approximations of the slave/environment at the master for environment contact prediction. This thesis proposes a novel pseudo two-channel nonlinear predictive controller and its variations that uses neural networks to online estimate the dynamics of the slave and environment, allowing replication of the environment contact force at the master using a similar local network. (Abstract shortened by UMI.)...