Date of Award

2025

Degree Type

Thesis

Degree Name

Master of Science in Mechanical Engineering and Applied Mechanics

Department

Mechanical, Industrial and Systems Engineering

First Advisor

Chengzhi Yuan

Abstract

Soft robots, primarily composed of compliant materials, exhibit highly complex and nonlinear dynamics, making precise control a significant challenge. Traditional control methods often struggle with these complexities, requiring more advanced and data-driven methods. In our previous work, “Automatic Control of a Soft Trunk Robot Actuated by Strings” [1], a proportional controller was developed to drive the steady-state tracking error to zero. While this method was effective, it required time-consuming manual tuning and was not easily adaptable to physical modifications of the robot.

To address these limitations, this research implements gain-scheduled feedback control with state-space models calculated through a multiple linear regression (MLR) algorithm, for a more advanced and dynamic soft trunk robot. A total of 8 plant models are generated using two distinct training methods across 16 training iterations, and are combined to create a set of 64 discrete-time models. Each model is simplified to retain only the most essential system behaviors while maintaining accuracy for effective control. For each configuration, optimal feedback control gains are computed using linear quadratic regulator (LQR) methods. These gains are scheduled based on the system’s configuration, providing control tuned for each operating region.

To enhance performance and ensure smooth transitions between scheduled gains, several key control features are added, including gain smoothing, rotation fighting protection, and origin protection. These improvements reduce instability during transitions and improve overall system settling time. The final system is operated through a custom graphical user interface (GUI), which allows for real-time monitoring, switching control modes, saving data, and plotting system responses.

Experimental validation shows that the system successfully tracks step reference inputs across its entire workspace, maintaining a steady-state error of less than 1 mm in position and less than 1 deg in orientation, with a settling time of around 3 seconds. These results reflect the system’s resolution limits and demonstrate the controller's accuracy and robustness. However, tracking sinusoidal reference inputs reveals limitations in the simplified linear modeling, as the steady-state error does not converge to zero and remains as a small-amplitude sinusoid. This reveals opportunities for future improvements through modeling improvements or machine learning-based approaches.

Overall, this work demonstrates that gain-scheduled LQR control combined with multiple linear regression modeling provides a practical and effective framework for controlling soft trunk robots. It offers a scalable solution for addressing the challenges of nonlinear dynamics, and its design allows for future extensions and improvements in both modeling and control strategies.

Trunk_New_StateSpace_Control_SupFile.py (22 kB)
Appendix D.1 Python Code

Trunk_New_Feedback_Control.ino (2 kB)
Appendix D.2 Arduino Code

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.