Hybrid feedback feedforward: An efficient design of adaptive neural network control

Area of Science:

• Robotics and Control Systems
• Adaptive Control Theory
• Neural Network Applications

Background:

• Euler-Lagrange systems are widely used in robotics and control.
• Adaptive approximation-based control (AAC) offers robust performance for uncertain systems.
• Traditional feedback AAC (FB-AAC) requires complex tuning and extensive system knowledge.

Purpose of the Study:

• To develop an efficient hybrid feedback feedforward adaptive approximation-based control (HFF-AAC) strategy.
• To simplify the control structure and reduce implementation costs for uncertain Euler-Lagrange systems.
• To enhance control performance in the presence of discontinuous friction.

Main Methods:

• Implemented a hybrid control structure combining proportional-derivative (PD) feedback and radial-basis-function (RBF) neural network feedforward.
• Utilized a sigmoid-jump-function neural network to handle discontinuous friction.
• Modified the RBF neural network input to only require desired outputs, simplifying the design.

Main Results:

• Achieved semiglobal practical asymptotic stability with simpler control schemes.
• Reduced the number of RBF neural network inputs and the requirement for prior knowledge of plant uncertainty bounds.
• Demonstrated comparable or superior performance to traditional FB-AAC with lower computational cost and simpler synthesis.

Conclusions:

• The proposed HFF-AAC strategy offers a cost-effective and efficient solution for controlling uncertain Euler-Lagrange systems.
• The simplified design and reduced knowledge requirements make it practical for real-world applications.
• This approach minimizes hardware selection, algorithm realization, and system debugging efforts.