Inverse-free zeroing neural network for time-variant nonlinear optimization with manipulator applications.

Area of Science:

• Robotics
• Neural Networks
• Optimization

Background:

• Time-variant optimization with nonlinear constraints is challenging.
• Existing neural network methods like zeroing neural network (ZNN) and gradient neural network (GNN) have limitations.
• Traditional ZNN requires computationally expensive matrix inversion, while GNN lacks sufficient accuracy.

Purpose of the Study:

• To propose a novel inverse-free zeroing neural network (IFZNN) algorithm.
• To address the computational complexity and accuracy issues of existing methods.
• To ensure robust performance for time-variant optimization problems.

Main Methods:

• Developed an inverse-free zeroing neural network (IFZNN) algorithm.
• The algorithm avoids matrix inverse and multiplication, reducing computational load.
• Theoretical analysis of convergence performance was conducted.

Main Results:

• The proposed IFZNN algorithm demonstrates superior accuracy and efficiency compared to traditional ZNN and GNN.
• Numerical simulations and comparative experiments validate the algorithm's performance.
• Path tracking for Universal Robot 5, Franka Emika Panda, and Kinova JACO2 manipulators confirmed practical applicability.

Conclusions:

• The novel IFZNN algorithm effectively solves time-variant optimization problems with nonlinear constraints.
• It offers significant advantages in reduced computational complexity and enhanced accuracy.
• The algorithm is suitable for practical applications, including robotic manipulator path tracking.