Controlling artificial muscles in the context of soft robots

Abstract

Rigid robots are highly specialised and can perform tasks with incredible precision. In contrast, soft robots provide a promising solution for creating robotic systems that are inherently better suited for unstructured and dynamic environments. Furthermore, they should be cheaper and safer than existing models. Artificial muscles comprise one of the core components of soft robots. Dielectric elastomer actuators (DEAs) represent the technology that comes closest to the capabilities of a natural muscle. However, their viscoelastic effects may limit the applicability of DEAs and represent the main reason why suitable control methods are required. Thus, the objective of this thesis is to answer the following research question: How can a bioinspired spiking closed-loop control system be used to improve the performance of a silicone-based commercially available stacked DEA in the context of soft robots? By answering this question, the research reported in this thesis attempts to take a step towards creating true soft robots. Furthermore, the research findings could potentially impact the field of neuromorphic engineering and advance research on neuromorphic motor control by introducing a new actuator technology. A real-time neuromorphic test bench was developed in the first step. An architecture was presented that can successfully drive different silicone-based commercially available stacked DEAs and measure their reaction to a stimulus. It can also be used to capture current and voltage continuous as well as peak values. Based on a field programmable gate array (FPGA), the neuromorphic test bench also offers the possibility of exchanging data with a digital neuromorphic hardware platform, which is ideally suited for the development of a bioinspired spiking closed-loop control system. The results presented provide a common foundation for research on the bioinspired controllability of dielectric elastomer actuators. Secondly, a control-orientated model of a silicone-based commercially available stacked DEA was defined and identified. For static modelling, the application of the Hookean law is sufficient for the deformations of such an actuator. Concerning dynamic modelling, a black box identification with a process model and a conversion into a state space representation is the preferred approach. The combined model delivered very good results for all validation signals in the low frequency range, which correspond to the frequency range of other proposed models. This led to the main finding, namely that for the development of soft robots on the basis of a DEA, methods of system identification are sufficient and more effective than the viscoelastic models considered. Finally, the results of the two previous steps were used for the development of a bioinspired spiking closed-loop control system. A spiking neural network (SNN) was developed that comprised the main part of the controller and whose output was used as the control value. All information inside the controller was represented via spikes and the controller was implemented on neuromorphic hardware. During the validation, a general functionality was proven and a frequency-dependent tracking performance was revealed. Nevertheless, with regard to the R2 value, a slightly better performance than a proportional integral (PI) controller with similar resolution and data preprocessing was observed. Furthermore, the findings show that the integration of a local online learning rule is essential to realise an advantage over classic control methods. With the results presented in this work, the development of a bioinspired spiking closed-loop control system is not completed. However, the findings of this work can lead to the creation of the first true soft robot that realises all the aforementioned potential.