Abstract
Traditional robots are mostly made of rigid components and they are optimised for a specific industrial repetitive task, such as manipulation, welding or pick and place. Their performance in a controlled environment without human interaction is typically very reliable, however, they are unable to adapt to unstructured environments. Historically, it has been challenging to apply the same technology from factories and controlled environments in everyday settings. In order to bring robots to the human environment, new types of robots have emerged that mimic biological organisms, such as animals or humans, and they promise to operate safer, better and faster among humans. Over the past decade, a wide range of soft sensors and actuators have been developed. However, the control mechanisms behind these devices have attracted relatively less attention. This thesis explores new strategies for implementing intelligence in soft robotic devices both in hardware and software.The structure of the thesis presents a transition from brain-dominated intelligence to body- dominated intelligence. First, traditional machine learning algorithms are used to understand and to model the shape of soft robots. The highly nonlinear deformation of these robots are notoriously difficult to measure because of the high number of degrees of freedom. Ideally, the robot’s body should contain many tiny sensors, however, this is challenging and expensive. We show that it is possible to combine cheap, off-the-shelf sensors and machine learning techniques in order to reproduce the high resolution deformation only using the time series of low-dimensional embedded sensors. We also present how these algorithms can be used to create scalable soft tactile sensors. Then, we make a transition from the brain-dominated algorithms and present new soft robotic mechanisms that combine algorithms on silicon chips and smart materials in order to understand the physical world. More specifically, we present a soft robotic fingertip that is able to measure both temperature and strain by combining a CCD camera and a smart skin that changes its colour when the temperature of the environment changes. Finally, we present an entirely new computational paradigm in soft robotics and show that computational structures can be made using only soft materials. Our Soft Matter Computer is able to control electronic actuators and sensors using a fluidic tape. This fluidic-electric system is then used to create novel analog and digital computational building blocks as well as intelligent soft robots.
The findings presented in this thesis have potential applications in healthcare, virtual reality, entertainment, nuclear industry and manufacturing.
| Date of Award | 23 Jun 2020 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Jonathan M Rossiter (Supervisor), Andrew T Conn (Supervisor) & Helmut Hauser (Supervisor) |