Neural network-based adaptive controller design of robotic manipulators with an observer

Area of Science:

• Robotics
• Control Systems
• Artificial Intelligence

Background:

• Robotic manipulators often face challenges with unknown dynamics and nonlinearities, hindering precise trajectory tracking.
• Accurate estimation of joint velocities is crucial for effective control, especially when only position measurements are available.

Purpose of the Study:

• To propose a novel neural network (NN)-based adaptive controller with an observer for robotic manipulators.
• To address trajectory tracking issues in systems with unknown dynamics and limited sensor measurements (joint angle positions only).
• To guarantee uniform ultimate bounds for tracking errors, observer errors, and NN weights.

Main Methods:

• A linear observer is utilized to estimate robot joint angle velocities.
• Neural networks (NNs) approximate the modified robot dynamics function to enhance control performance.
• A conventional adaptive algorithm using linearity in parameters is developed for comparative analysis.

Main Results:

• The proposed NN-based adaptive controller ensures bounded tracking errors, observer errors, and NN weights.
• Simulation studies demonstrate the effectiveness of the NN approach compared to the conventional adaptive method.
• The controller successfully handles unknown dynamics and nonlinearities in robotic manipulator systems.

Conclusions:

• The NN-based adaptive controller with an observer offers a robust solution for trajectory tracking in robotic manipulators.
• The proposed method provides superior control performance over conventional adaptive techniques, particularly in the presence of uncertainties.
• This approach enhances the reliability and precision of robotic systems through advanced AI control strategies.