Neural network-based model predictive tracking control of an uncertain robotic manipulator with input constraints

Area of Science:

• Robotics
• Control Systems Engineering
• Artificial Intelligence

Background:

• Robotic manipulators often face challenges with model uncertainty and input constraints, hindering precise control.
• Model Predictive Control (MPC) is a powerful control strategy, but its effectiveness can be limited by complex system dynamics and real-time computational demands.

Purpose of the Study:

• To propose a novel neural network-based Model Predictive Control (NN-based MPC) method for robotic manipulators.
• To address challenges posed by model uncertainty and input constraints in robotic manipulator control.
• To enhance the accuracy and stability of robotic manipulator control systems through adaptive learning strategies.

Main Methods:

• Utilized two groups of radial basis function neural networks (RBFNNs) for online model estimation and optimization.
• Implemented online learning strategies within RBFNNs to manage system uncertainty and improve model accuracy.
• Employed an actor-critic scheme with adaptive learning for balancing tracking performance and system stability.
• Incorporated a nonquadratic cost function to ensure adherence to input constraints.

Main Results:

• The proposed NN-based MPC method demonstrated effective handling of model uncertainty in robotic manipulators.
• Online learning strategies and RBFNNs significantly improved model estimation accuracy and predictive capabilities.
• The adaptive learning approach successfully balanced optimal tracking performance with predictive system stability.
• Input constraints were effectively guaranteed, and ultimately uniformly boundedness (UUB) of all variables was verified via Lyapunov analysis.

Conclusions:

• The developed NN-based MPC framework provides a robust solution for controlling robotic manipulators under uncertainty and constraints.
• The integration of RBFNNs and adaptive learning strategies offers a promising direction for advanced robotic control.
• Simulation results validate the effectiveness and stability of the proposed control method.