Control of soft robots with inertial dynamics.

Area of Science:

• Robotics
• Control Theory
• Machine Learning

Background:

• Soft robots offer enhanced safety and capabilities in human-centric and complex environments.
• Modeling and controlling soft robots is challenging due to infinite degrees of freedom and nonlinear dynamics.
• Existing methods often limit soft robot control to quasi-static motions or quasi-linear deflections.

Purpose of the Study:

• To advance the modeling and control of soft robots into the inertial and nonlinear dynamic regime.
• To develop a rapid, data-driven approach for accurate soft robot modeling and control.
• To enable high-speed and high-deflection movements in soft robotic systems.

Main Methods:

• Leveraged Koopman operator theory for a data-driven modeling approach.
• Introduced the static Koopman operator as a pregain term in optimal control.
• Trained and controlled two morphologically different soft robots.

Main Results:

• Achieved control of soft robot motions with velocities 10x greater and accelerations 40x greater than prior work.
• Successfully controlled high-deflection shapes exceeding 110° of curvature.
• Demonstrated rapid model training (<5 min) and low computational cost for model building (0.5 s).

Conclusions:

• The developed approach enables rapid modeling and control of soft robots in the inertial, nonlinear regime.
• This work overcomes limitations of quasi-static and quasi-linear models for soft robots.
• Paves the way for next-generation compliant, highly dynamic soft robots.