Design of Multimodal Neural Network Control System for Mechanically Driven Reconfigurable Robot

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Reconfigurable robots offer adaptability but pose modeling and control challenges.
• Modular design principles are crucial for managing complexity in robotic systems.

Purpose of the Study:

• To develop a modular modeling method for reconfigurable robots.
• To establish a unified kinematic expression for automated model generation.
• To apply a neural network-based multimodal fusion model for robot control.

Main Methods:

• Classification and library creation of robot modules (rotary, swing, mobile).
• Modular modeling to establish kinematic models with a unified expression.
• Development of a long- and short-term memory neural network for multimodal information fusion and control.
• Hardware design of joint controllers and drivers using Modbus protocol for distributed control.

Main Results:

• A unified expression for modular kinematics was established, enabling analysis of reconfigurable robot kinematics.
• A specific long- and short-term memory neural network model was designed and applied to robot control.
• Successful implementation of distributed control using RS485 communication and Modbus protocol.
• Experimental validation of point and continuous path control confirmed system correctness and reliability.

Conclusions:

• The modular approach facilitates rapid identification and kinematic modeling of reconfigurable robots.
• The developed neural network model effectively addresses control challenges in mechanically driven reconfigurable robots.
• The distributed control system architecture ensures reliable operation and accurate motion control.