Reinforcement Learning-Based Boundary-Optimized Control of Flexible Manipulators Under Jointly Connected Switching...

Area of Science:

• Robotics
• Control Systems Engineering
• Applied Mathematics

Background:

• Flexible manipulators present challenges in precise control due to vibrations and external disturbances.
• Existing control strategies often lack robustness in dynamic environments with intermittent communication.
• Fault tolerance is crucial for reliable operation in complex robotic systems.

Purpose of the Study:

• To develop boundary-optimized fault-tolerant tracking control for flexible manipulators.
• To address challenges posed by switching digraphs and heterogeneous linear leaders.
• To minimize tracking errors, vibration deflections, and control energy.

Main Methods:

• Design of a distributed observer for leader information in switching graphs with interrupted communication.
• Development of a partial differential equation (PDE)-based fault observer (FO) using boundary states.
• Formulation of a novel long-term integral cost function for optimization.
• Approximation of boundary optimal control laws using actor-critic neural networks (NNs) and reinforcement learning (RL).

Main Results:

• The proposed controllers ensure that the flexible manipulator's error states are uniformly ultimately bounded (UUB).
• Demonstrated effectiveness through numerical simulations, validating the control approach.
• Successful estimation of unknown faults using limited boundary state information.

Conclusions:

• The developed boundary-optimized fault-tolerant control is effective for flexible manipulators in complex communication environments.
• The integration of PDE-based fault observers and RL-based control offers a robust solution.
• The proposed method enhances tracking performance while ensuring system stability and fault tolerance.