A study of neural network control of robot manipulators

Abstract

SummaryThe basic robot control technique is the model based computer-torque control which is known to suffer performance degradation due to model uncertainties. Adding a neural network (NN) controller in the control system is one effective way to compensate for the ill effects of these uncertainties. In this paper a systematic study of NN controller for a robot manipulator under a unified computed-torque control framework is presented. Both feedforward and feedback NN control schemes are studied and compared using a common back-propagation training algorithm. Effects on system performance for different choices of NN input types, hidden neurons, weight update rates, and initial weight values are also investigated. Extensive simulation studies for trajectory tracking are carried out and compared with other established robot control schemes.

Neural network inverse control techniques for PD controlled robot manipulator

“…Further evaluation of the gradients in (17) and (26) yields the following back-propagation algorithm to update the intenal weights…”

Neural network inverse control techniques for PD controlled robot manipulator

“…Further evaluation of the gradients in (17) and (26) yields the following back-propagation algorithm to update the intenal weights…”

Neural network impedance force control of robot manipulator

“…Figure 1 shows the torque-based impedance control structure. The torque-based impedance control law F is F =D U +ĥ + F e (10) whereD ;ĥ are estimates of D ; h and F e is the exerted force on the environment. And U is given by U = X r +M ?1 B( _…”

“…The control system is depicted in Figure 1. Combining (9), (10), and (11) yields the closed loop tracking error dynamic equation…”

On an effective design approach of cartesian space neural network control for robot manipulators

“…where K P and K D are n ϫ n symmetric positive definite gain matrices for the desired damping , and x d is the desired trajectory . Combining (9) , (10) , and (11) yields the closed loop tracking error dynamic equation…”

“…In this way , the same output dimensions of NN filter 1 and NN filter 2 are maintained . Combining (30) with (9) yields…”

Neural network inverse control techniques for PD controlled robot manipulator

“…Further evaluation of the gradients in (17) and (26) yields the following back-propagation algorithm to update the intenal weights…”

Neural network impedance force control of robot manipulator

“…Figure 1 shows the torque-based impedance control structure. The torque-based impedance control law F is F =D U +ĥ + F e (10) whereD ;ĥ are estimates of D ; h and F e is the exerted force on the environment. And U is given by U = X r +M ?1 B( _…”

“…The control system is depicted in Figure 1. Combining (9), (10), and (11) yields the closed loop tracking error dynamic equation…”

On an effective design approach of cartesian space neural network control for robot manipulators

“…where K P and K D are n ϫ n symmetric positive definite gain matrices for the desired damping , and x d is the desired trajectory . Combining (9) , (10) , and (11) yields the closed loop tracking error dynamic equation…”

“…In this way , the same output dimensions of NN filter 1 and NN filter 2 are maintained . Combining (30) with (9) yields…”

Neural network inverse control techniques for PD controlled robot manipulator

“…Further evaluation of the gradients in (17) and (26) yields the following back-propagation algorithm to update the intenal weights…”

Neural network impedance force control of robot manipulator

“…Figure 1 shows the torque-based impedance control structure. The torque-based impedance control law F is F =D U +ĥ + F e (10) whereD ;ĥ are estimates of D ; h and F e is the exerted force on the environment. And U is given by U = X r +M ?1 B( _…”

“…The control system is depicted in Figure 1. Combining (9), (10), and (11) yields the closed loop tracking error dynamic equation…”

On an effective design approach of cartesian space neural network control for robot manipulators

“…where K P and K D are n ϫ n symmetric positive definite gain matrices for the desired damping , and x d is the desired trajectory . Combining (9) , (10) , and (11) yields the closed loop tracking error dynamic equation…”

“…In this way , the same output dimensions of NN filter 1 and NN filter 2 are maintained . Combining (30) with (9) yields…”