Pawlowski, Ben, authorZhao, Jianguo, advisorPuttlitz, Christian, committee memberAnderson, Charles, committee member2020-01-132020-01-132019https://hdl.handle.net/10217/199791Soft robots are a new type of robot with deformable bodies and muscle-like actuations, which are fundamentally different from traditional robots with rigid links and motor-based actuators. Owing to their elasticity, soft robots outperform rigid ones in safety, maneuverability, and adaptability. With their advantages, many soft robots have been developed for manipulation and locomotion in recent years. However, the current state of soft robotics has significant design and development work, but lags behind in modeling and control due to the complex dynamic behavior of the soft bodies. This complexity prevents a unified dynamics model that captures the dynamic behavior, computationally-efficient algorithms to simulate the dynamics in real-time, and closed-loop control algorithms to accomplish desired dynamic responses. In this thesis, we address the three problems of modeling, simulation, and control of soft robots. For the modeling, we establish a general modeling framework for the dynamics by integrating Cosserat theory with Hamilton's principle. Such a framework can accommodate different actuation methods (e.g., pneumatic, cable-driven, artificial muscles, etc.). To simulate the proposed models, we develop efficient numerical algorithms and implement them in C++ to simulate the dynamics of soft robots in real-time. These algorithms consider qualities of the dynamics that are typically neglected (e.g., numerical damping, group structure). Using the developed numerical algorithms, we investigate the control of soft robots with the goal of achieving real-time and closed-loop control policies. Several control approaches are tested (e.g., model predictive control, reinforcement learning) for a few key tasks: reaching various points in a soft manipulator's workspace and tracking a given trajectory. The results show that model predictive control is possible but is computationally demanding, while reinforcement learning techniques are more computationally effective but require a substantial number of training samples. The modeling, simulation, and control framework developed in this thesis will lay a solid foundation to unleash the potential of soft robots for various applications, such as manipulation and locomotion.born digitalmasters thesesengCopyright and other restrictions may apply. User is responsible for compliance with all applicable laws. For information about copyright law, please see https://libguides.colostate.edu/copyright.reinforcement learningsymplectic integrationsoft robotsmodel predictive controlModeling, simulation, and control of soft robotsText