Composite Observer-Based Optimal Attitude-Tracking Control With Reinforcement Learning for Hypersonic Vehicles.

Area of Science:

• Aerospace Engineering
• Control Systems
• Artificial Intelligence

Background:

• Hypersonic vehicles face significant uncertainties and nonlinearities during reentry, making accurate attitude control challenging.
• Traditional control methods struggle with unknown dynamics and external disturbances common in the reentry phase.
• Precise attitude tracking is critical for mission success and vehicle stability.

Purpose of the Study:

• To develop an observer-based reinforcement learning (RL) control strategy for optimal attitude tracking of hypersonic vehicles in the reentry phase.
• To address the challenge of unknown system dynamics and external disturbances.
• To improve the convergence and stability of the RL control system.

Main Methods:

• A composite observer, integrating a neural-network (NN)-based Luenberger-type observer and a synchronous disturbance observer, was designed for state estimation.
• A reinforcement learning (RL) tracking controller was synthesized using information from the composite observer.
• Concurrent learning was employed to enhance critic network weight convergence, replacing traditional persistent excitation requirements.

Main Results:

• The proposed composite observer effectively estimates system states and identifies unknown nonlinear dynamics and disturbances simultaneously.
• The RL tracking controller, informed by the observer, achieves optimal attitude tracking for hypersonic vehicles.
• Concurrent learning improved convergence performance, and theoretical analysis confirmed bounded weight estimation errors under specific conditions.

Conclusions:

• The developed observer-based RL control approach demonstrates significant effectiveness and superiority for hypersonic vehicle attitude-tracking systems.
• The method provides a robust solution for controlling vehicles in complex reentry environments with uncertainties.
• This research contributes a novel framework for advanced control of autonomous aerospace systems.