A Neural Network Based Approach to Inverse Kinematics Problem for General Six-Axis Robots

Area of Science:

• Robotics
• Artificial Intelligence
• Computational Mechanics

Background:

• Inverse Kinematics Problems (IKP) are critical for robot control but are challenging due to high non-linearity and complexity in multi-axis manipulators.
• Existing methods often struggle with precision and efficiency when solving IKP for general robotic systems.
• Six-axis robotic manipulators present significant computational hurdles in achieving accurate real-time control.

Purpose of the Study:

• To propose a novel neural network (NN) based approach for precise and efficient solution of Inverse Kinematics Problems (IKP).
• To address the inherent complexity and non-linearity challenges in solving IKP for six-axis robotic manipulators.
• To enhance the accuracy and reduce the computational cost of IKP solutions in robotics.

Main Methods:

• A joint space segmentation strategy simplifies IKP complexity, with data generated via forward kinematics.
• Multilayer Perception (MLP) networks are trained piecewise to learn the IKP solution space.
• Classification models are employed to reduce inference computational cost by selecting appropriate MLPs.
• Numerical error minimization refines the initial NN-predicted solution for improved accuracy.

Main Results:

• The proposed NN approach with joint space segmentation and error minimization demonstrates feasibility for IKP.
• Simulations on a 6-DOF manipulator (Xarm6) validate the method's effectiveness and high-precision capability.
• The algorithm shows superior efficiency and accuracy compared to existing NN-based IKP methods in the literature.

Conclusions:

• Neural networks offer a viable and effective solution for complex Inverse Kinematics Problems in robotics.
• The proposed methodology provides a significant advancement in solving IKP for general robotic manipulators with high precision requirements.
• This approach enhances both the computational efficiency and the solution accuracy for robotic control applications.