Biomimetic Hybrid Feedback Feedforward Neural-Network Learning Control.

Area of Science:

• Control Systems Engineering
• Biomimetic Systems
• Artificial Intelligence

Background:

• Uncertain nonlinear systems pose significant challenges for traditional control methods.
• Existing neural-network learning control (NNLC) often requires complex state feedback.
• Human motor control offers insights into robust and adaptive learning mechanisms.

Purpose of the Study:

• To develop a novel biomimetic hybrid feedback-feedforward neural-network learning control (NNLC) strategy.
• To address limitations in existing NNLC methods for uncertain nonlinear systems.
• To simplify the analysis and synthesis of NNLC by utilizing recurrent reference signals.

Main Methods:

• A hybrid control structure combining a proportional-derivative (PD) controller and a radial-basis-function (RBF) neural network (NN).
• The RBF NN acts as a feedforward predictive machine, learning from recurrent reference signals.
• Analysis based on semiglobal practical exponential stability under specific control parameter constraints.

Main Results:

• The proposed NNLC strategy guarantees semiglobal practical exponential stability for the closed-loop system.
• Accurate neural network approximation is achieved within a local region along recurrent reference trajectories.
• The method simplifies NNLC by avoiding the need for recurrent plant states as NN inputs.

Conclusions:

• The biomimetic hybrid NNLC strategy effectively controls uncertain nonlinear systems.
• Utilizing reference signals as NN inputs significantly simplifies control system design and analysis.
• The approach demonstrates a promising direction for advanced adaptive control systems inspired by biological learning.