PD Control Compensation Based on a Cascade Neural Network Applied to a Robot Manipulator

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Machine Learning

Background:

• Proportional Integral Derivative (PID) controllers face degradation over time, reducing stability and bandwidth in robot manipulators.
• Proportional Derivative (PD) controllers offer improvements but struggle with uncompensated gravity and system non-linearities.
• Existing adaptive controllers can be enhanced with more efficient learning methods to manage system uncertainties and disturbances.

Purpose of the Study:

• To develop an advanced adaptive control scheme for robot manipulators.
• To address limitations of conventional PID and PD controllers in handling system non-linearities and uncertainties.
• To improve the performance and stability of robot manipulator control systems.

Main Methods:

• Implementation of a nominal control law for sub-optimal performance.
• Integration of a cascade neural network for non-linear compensation.
• Utilizing radial basis function neural networks and a recompense function for weight updates and system identification.

Main Results:

• The proposed cascade neural network-based adaptive controller demonstrated improved performance compared to conventional PD control.
• The scheme effectively compensated for unknown dynamic non-linearities and external disturbances.
• Validation was performed on a two-degree-of-freedom robot manipulator.

Conclusions:

• Cascade neural networks offer a powerful approach for adaptive non-linear compensation in robot control.
• The developed weight update function and radial basis function networks enhance learning and identification capabilities.
• This adaptive control strategy provides a robust solution for improving robot manipulator performance and stability.