University of Glasgow
Biological inspired control and machine learning for clinical rehabilitation and engineering systems
Abstract
dc:description.abstractHuman quiet standing has been studied over the years in order to model controllers able to replicate variability and intermittency present in human control mechanisms. The Intermittent Control (IC) framework was proposed as a computational model and can be described by a serie of open-loop trajectories with close-loop triggering. However, the original implementation is based on a deterministic approach which requires knowledge of the underlying system’s dynamics. This thesis explores the capabilities of a data-driven stochastic Intermitent Controller, using Gaussian Processes (GP), applied mainly to a Single Inverted Pendulum (SIP). Throughout this thesis, the Intermitent Controller framework has been adapted to move toward a data-driven intermittent controller using Reinforcement Learning (RL) and Data Informativity (DI) to estimate the state feedback gains. Simulations show the benefit of the open-loop trajectories created by IC in improving the overall estimation of the parameters, compared to Continuous Controllers. These results are the initial basis for a deterministic data-driven Intermittent Controller. In addition, the integration of GPs within this IC framework is able to introduce varibility in the generated control input by using probabilistic open-loop trajectories. Both these implementations have been combined to create the first data-driven stochastic intermittent controller. Results presented in this work are showing the capability of this newly controller approach to handle adaptation as well as switching between deterministic like behavior to fully probabilistic characteristics based on IC and GP parameters.
Degree
thesis:*- Level dc:type.qualificationlevel
- PhD
- Grantor dc:publisher.institution
- University of Glasgow
- Year dc:date.issued
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Doublein, Thomas
Subjects
dc:subject × 1Rights
- Language dc:language
- en
Identifiers
dc:identifier.*- DOI dc:identifier.doi
- https://doi.org/10.5525/gla.thesis.84709