Configuration identification of on-demand variable stiffness strain-limiting layers in zig-zag soft pneumatic...

Area of Science:

• Robotics
• Materials Science
• Artificial Intelligence

Background:

• Soft pneumatic actuators (SPAs) typically have fixed trajectories, requiring redesign for different motions.
• Achieving multiple trajectories necessitates complex re-fabrication of SPAs.
• Passive modular variable stiffness SPAs offer adaptability but pose challenges in inverse problem modeling.

Purpose of the Study:

• To develop a method for predicting strain-limiting layer (SLL) configurations for SPAs to achieve specific tip trajectories.
• To enable functional and application-specific SPA deployment without structural redesign.
• To address the challenge of solving the inverse problem for SPAs with modular SLLs.

Main Methods:

• A hybrid methodology combining feed-forward neural networks and convolutional neural networks was employed.
• The approach predicts the required SLL configuration for a desired SPA tip trajectory in Cartesian space.
• The methodology is designed to be generic for various SPA configurations.

Main Results:

• The proposed method accurately predicted SPA configurations for diverse applications, including endoscopes, grippers, and human finger mimics.
• An average prediction error of 1.65% was achieved across tested applications.
• The study demonstrated the feasibility of generating specific trajectories by varying SLL configurations.

Conclusions:

• A versatile methodology is presented for creating function and application-specific SPAs.
• Strategic combination of SLLs and machine learning enables trajectory generation without re-fabrication.
• This approach enhances the adaptability and utility of soft robotic manipulators.