Data-Driven Twisted String Actuation for Lightweight and Compliant Anthropomorphic Dexterous Hands.

Area of Science:

• Robotics
• Mechanical Engineering
• Artificial Intelligence

Background:

• Traditional robotic hand actuation systems are heavy, complex, and lack compliance.
• Twisted String Actuators (TSAs) offer lightweight, compliant alternatives but struggle with nonlinear control under variable loads.

Purpose of the Study:

• To develop a high-precision control method for TSAs under variable loads.
• To design and validate a lightweight, anthropomorphic dexterous robotic hand using innovative mechanisms.

Main Methods:

• A dual hidden-layer Back Propagation Neural Network (BPNN) was used for data-driven TSA displacement prediction.
• A biomimetic, underactuated five-finger dexterous hand with a tendon-spring transmission was engineered.
• An experimental platform integrated BPNN-kinematic modeling for performance validation.

Main Results:

• Accurate prediction of TSA displacement under loads ranging from 0.1-4.2 kg.
• Achieved an ultra-lightweight dexterous hand with precise bending angle prediction.
• Demonstrated versatile gesture replication and robust grasping with a maximum fingertip force of 7.4 N.

Conclusions:

• The integrated data-driven modeling and biomimetic design advance TSA control for variable-load applications.
• This work presents a new paradigm for creating high-performance, lightweight dexterous robotic hands.