Date of Award


Degree Type


Degree Name

Doctor of Philosophy in Mechanical Engineering and Applied Mechanics


Mechanical, Industrial and Systems Engineering

First Advisor

Chengzhi Yuan


Inspired from biological systems, researchers are designing soft robots to replace conventional rigid body robots in many applications, including human-machine interaction, manipulation, medical instrumentation and wearable devices.

Soft robots, due to their flexible bodies capable of achieving complex movements, are suitable for motion in unstructured complex environments. Moreover, soft robots can provide a more pleasant interaction experience for humans, or grip and manipulate fragile objects. However, because of the potential ability of the soft robot, controlling the soft robot still remains as a difficult task, as the soft robot contains high complexity and non-linearity. Currently, machine learning is involved in many types of research to solve those hard problems. So here, a special type of machine learning algorithm called deterministic learning is being studied and tested to see if we can solve soft robot problems.

In this dissertation, Two types of soft robots will be discussed include a cable-driven switching-legged inchworm-inspired soft robot, and an air-pressured trunk-type soft robot. Detailed discussion about their design ideas, fabrication process, and locomotion are introduced. Then an adaptive neural network learning algorithm is considered and applied to those soft robots to achieve modeling and control. This kind of machine learning-based data-driven strategy can achieve accurate modeling or trajectory tracking without needing precise physical modeling or a complex sensor to measure the whole environment. This learning algorithm is developed based on Deterministic Learning (DL) with the radial basis function neural network (RBF NN), which is able to effectively handle the soft robot's complex nonlinear uncertain dynamics. As for the result, the soft robot can be accurately modeled, and precise tracking control is achievable with a short period of online neural network learning.

Available for download on Wednesday, May 08, 2024