Adaptive Predefined-Time Tracking Control for Robotic Manipulator Based on Actor-Critic Reinforcement Learning.

Area of Science:

• Robotics
• Control Systems
• Artificial Intelligence

Background:

• Manipulator systems often face uncertainties in dynamics, complicating precise control.
• Traditional control methods may struggle with fast convergence and guaranteed settling times for complex systems.

Purpose of the Study:

• To develop a novel predefined-time adaptive neural tracking control method for uncertain manipulator systems.
• To integrate predefined-time stability theory with reinforcement learning for enhanced control performance.
• To achieve fast convergence with explicitly preset settling time bounds.

Main Methods:

• Utilizing an Actor-Critic reinforcement learning framework with neural networks.
• An Actor network approximates unknown dynamics and generates control signals.
• A Critic network optimizes the learning process by evaluating the cost-to-go function.
• Incorporating specific terms in control laws and weight updates for predefined-time convergence.
• Employing Lyapunov stability theory for rigorous analysis.

Main Results:

• Demonstrated predefined-time convergence of tracking errors within preset bounds.
• Ensured boundedness of all closed-loop signals.
• The settling time is adjustable via a single design parameter, independent of initial conditions.
• Simulation results show superior performance compared to conventional PID control.

Conclusions:

• The proposed predefined-time adaptive neural control method is effective for uncertain manipulator systems.
• The Actor-Critic reinforcement learning framework successfully achieves fast and stable tracking control.
• This approach offers a significant improvement over existing control strategies for robotic manipulators.