Abstract

The field of robotics has undergone a significant transformation, extending its scope well beyond its traditional role in manufacturing automation. It has now found applications in various domains such as healthcare, field exploration, and collaborative human-robot interactions. Nevertheless, a main concern across these diverse applications remains the safety of interactions involving humans. Traditional robots, comprised of rigid links and joints, inherently carry risks when operating close to human beings. This risk is exacerbated by the absence of compliance in their actuation mechanisms. In contrast, soft robots are constructed from inherently soft or extensible materials, affording them the ability to deform and absorb energy during collisions. This distinctive characteristic endows them with a continuously deformable structure and muscle-like actuation, closely resembling biological systems and offering a greater number of degrees of freedom. Consequently, soft robots hold the potential to exhibit extraordinary levels of adaptability, sensitivity, and agility.

The emergence of soft robots marks a new frontier at the intersection of multiple disciplines, including engineering, materials science, mechanics, physics, chemistry, biology, and robotics. This interdisciplinary confluence catalyzes innovation, pushing the boundaries of robotic capabilities and unlocking fresh avenues for exploration and practical applications.

Among the materials ideally suited for soft robotics, Dielectric Elastomer (DE) is one of the promising candidates due to its exceptional performance attributes. However, the intricate nonlinear characteristics inherent to Dielectric Elastomer Actuators (DEAs), including phenomena such as hysteresis, stress relaxation, and various dependencies, pose great challenges in modeling and control.

This dissertation is dedicated to advancing the modeling and control strategies for Dielectric Elastomer Actuators (DEAs) with the primary objective of integrating them into soft robot applications.

The research endeavors commence with a solid foundation in the form of extensive experimental tests. These tests investigate the input-output characteristics of DEAs, systematically exploring their responses under varying input amplitudes, frequencies, and mechanical loads. The experimental results unveil intricate and multifaceted behaviors influenced by factors such as input frequencies, amplitudes, and external mechanical loads.

This study focuses specifically on conical and planar Dielectric Elastomer Actuators (DEAs) and introduces two distinct models based on fundamental physical principles. These proposed models are inspired by the concept of free energy within viscoelastic materials, allowing them to comprehensively capture the intricate behaviors exhibited by DEAs while considering their complex dependencies. Particularly, these models can describe the intricate influences of multiple factors that shape DEA behaviors. The precision and effectiveness of these models are rigorously validated through meticulous comparisons with experimental data.

Due to the necessity for actuator-specific details in physics-based models, an innovative approach is presented, namely a data-in-loop model. This groundbreaking model adopts nonlinear elements, encompassing phenomena such as creep and hysteresis, thus avoiding the need for geometry-specific information and effectively representing the intricate behaviors of DEAs.

The presence of nonlinear effects in DEAs can lead to harmful consequences, including inaccuracies, oscillations, and instability. To effectively counter these effects, a controller design approach is proposed, adopting feedforward inverse compensation methods for controller design. In this framework, a model based on Prandtl-Ishlinskii (PI) hysteresis blocks is adopted to account for the nonlinearities within DEAs. The direct inverse compensation technique is employed with the inverse of the PI model. Building upon this foundation, a robust adaptive controller is then developed. This comprehensive methodology is designed to mitigate the adverse impacts of nonlinearities in DEAs, ultimately enhancing their control performance and addressing the formidable challenges posed by dynamic behaviors.