Discrete-time neural inverse optimal control for nonlinear systems via passivation.

Area of Science:

• Control Engineering
• Artificial Intelligence
• Robotics

Background:

• Optimal control of nonlinear systems often requires solving the complex Hamilton-Jacobi-Bellman equation.
• Accurate system models are crucial for effective control, but often unavailable for real-world nonlinear systems.
• Passivity theory provides a robust framework for analyzing and designing controllers for complex systems.

Purpose of the Study:

• To present a discrete-time inverse optimal neural controller that bypasses the need to solve the Hamilton-Jacobi-Bellman equation.
• To develop an on-line neural identification method for modeling unknown nonlinear systems.
• To demonstrate the controller's effectiveness on challenging systems like unstable nonlinear systems and robotic manipulators.

Main Methods:

• Utilizing inverse optimal control principles to design controllers without explicit Hamilton-Jacobi-Bellman solution.
• Employing a recurrent neural network trained with an extended Kalman filter for on-line system identification.
• Integrating passivity theory into the controller design for enhanced stability and robustness.

Main Results:

• Successfully designed a discrete-time inverse optimal neural controller for nonlinear systems.
• Developed an effective on-line neural identification technique for unknown system dynamics.
• Validated the controller's performance through simulations on an unstable nonlinear system and a planar robot model.

Conclusions:

• The proposed discrete-time inverse optimal neural controller effectively manages unknown nonlinear systems.
• The combination of inverse optimal control and neural identification offers a practical alternative to traditional optimal control methods.
• The approach shows promise for applications in robotics and control of complex unstable systems.