Adaptive Robust Control of Uncertain Euler-Lagrange Systems Using Gaussian Processes.

Area of Science:

• Robotics
• Control Systems Engineering
• Machine Learning

Background:

• Euler-Lagrange systems are complex and susceptible to external disturbances.
• Accurate modeling of system uncertainties is crucial for high-precision control.
• Existing adaptive control methods may struggle with data efficiency and rapid learning.

Purpose of the Study:

• To develop a novel adaptive robust control strategy for uncertain Euler-Lagrange systems.
• To leverage Gaussian processes (GPs) for data-based uncertainty modeling and adaptive control.
• To improve tracking precision and uncertainty learning efficiency in dynamic environments.

Main Methods:

• Utilizing Gaussian Process Regression (GPR) to create a nonparametric uncertainty model with confidence intervals.
• Integrating GPR posterior means for dynamic compensation and posterior variances for feedback gain adjustment within an Adaptive Sliding Mode Control (ASMC) framework.
• Introducing a new adaptive law for hyperparameter updates based on tracking error feedback.

Main Results:

• The proposed control strategy demonstrates robustness against significant system uncertainty using low feedback gains.
• The novel adaptive law enables data-efficient and faster uncertainty learning compared to traditional methods.
• Simulations on an underwater robot model confirm enhanced tracking control and uncertainty learning performance.

Conclusions:

• Gaussian processes combined with an adaptive law offer a powerful approach for robust control of uncertain systems.
• The method ensures semiglobal asymptotic convergence of tracking error with a defined probability.
• This research advances adaptive control by integrating machine learning for improved performance and learning efficiency.