Conventional robots typically utilize rigid materials and structures, employing rigid links to connect rigid joints in either serial or parallel kinematic chains. These rigid designs offer advantages such as high precision, improved load capacity, and enhanced stability during motion. Compared to robots with serial structures, like industrial manipulators or collaborative robots, those with parallel structures, such as the Stewart platform and Delta robot, further improve rigidity and stiffness. In more complex robotic systems, such as humanoid robots, hybrid designs combining both serial and parallel structures are also employed, leveraging the strengths of both.
However, rigid materials and structures come with inherent limitations, including reduced adaptability to dynamic environments and challenges in ensuring safe interaction in uncertain settings. For example, rigid robots pose risks of injury during unintended collisions. To overcome these limitations, researchers have sought to optimize traditional serial and parallel robots, as well as their actuators, by introducing flexible materials and structures.
In serial robots, medical and exploratory robots were among the first to adopt flexible materials and structures, leading to the development of continuum robots that can deform at any point along their bodies. The compliance of these robots allows for safer interactions in confined environments. More recently, research has extended to combining parallel structures with elastic components, giving rise to parallel continuum robots. These robots effectively combine the precision of parallel structures with the safety provided by elasticity under low loads. Additionally, the introduction of elasticity in robot joints has led to the development of various elastic actuators, which offer advantages such as improved shock absorption, enhanced safety in human-robot interaction, and energy efficiency through the storage and release of energy for smoother movements. Incorporating elastic structures and materials into robots introduces added complexity in control. These non-rigid components require individual modeling at varying levels of fidelity, which must then be integrated into the overall system for unified modeling and control. While this poses challenges, advancements in computational power and learning-based methods have led to significant progress in the modeling and real-time control of robots with elastic structures.
This dissertation focuses on robots utilizing elastic structures, including a tendon driven catheter continuum robot, a multi-tendon continuum robot, a parallel continuum robot, and a collaborative robot featuring four types of series elastic actuators. Contributions were made across various stages of design, system development, modeling, and control for these robots. To enhance system control, a dynamic Gaussian Mixture Model (DGMM)-based control framework was implemented. This framework learns the nonlinear behavior of robots with elastic structures through demonstrations performed by robots controlled using an analytical model, either in simulation or in real experimental setups, enabling more precise control based on the learned model.
In nature, organisms, including humans, consist of both rigid structures (such as bones) and elastic components (such as tendons and muscles). The incorporation of these rigid and elastic structures enhances adaptability to the environment and allows for more complex interactions between individuals and their surroundings. With ongoing advances in materials, modeling, and control approaches, robotics is expected to increasingly incorporate elastic structures and materials, leading to more adaptable, capable, and interactive robots.