| This thesis summarizes an investigation of the application of neural network (NN) control to a commercial robot. It presents the results of simulations and experiments that illustrate the effect of the neural network learning rate on system behaviour.; The development of a real-time NN controller is described. The NN is implemented on a digital signal processor (DSP), and its signals are communicated to the controller of a CRS Robotics A465 robot in real-time. The NN is trained to generate a compensating signal that is added to the commanded PID torque for the first three joints of the robot, in order to reduce the trajectory tracking errors of the robot joint actuators.; A state-space model of the system is developed, which is subsequently utilized in a numerical simulation. The simulation is used to perform a sensitivity analysis, to determine the effect of variations in the NN learning rate on system behaviour.; Experiments are conducted to investigate the effect of the NN learning rate on the reduction in joint trajectory tracking error. (Abstract shortened by UMI.)... |
