About the project
There is an ongoing transformation in engineering autonomous systems that aim to achieve complex objectives with limited human intervention in applications such as robotics, self-driving cars, and industrial systems. While the design of autonomous systems has typically relied on pre-defined models, the desire to operate in complex, unknown, or varying conditions implies that models of the system and the operating environment may not be always available. As a result, machine learning and data-driven approaches are on the rise and have the potential for impact in autonomous systems. However, embedding machine learning in autonomous systems is facing significant challenges in terms of safety, robustness, and resource efficiency.
In this project, we focus on the problem of designing control policies for autonomous systems. We aim to develop new methodologies that combine control, optimization, and machine learning tools. The objectives are to a) consider resource constraints: limited computing to run complex machine learning models for example neural networks, or limited communication resources in networked autonomous systems for example in a network of robots, b) analyze the safety and robustness of the learned policies to disturbances or malicious disruptions, c) demonstrate the benefits of the design methodologies in numerical simulations. Experimental evaluation may also be considered.
You will have a background in control systems, mathematical optimization, and machine learning, as well as excellent numerical analysis and programming skills.
You will gain training in advanced mathematical and numerical skills. We will support the advancement of your career and provide opportunities for professional networking and external collaborations.
You will be based in the Cyber Physical Systems group with academics and researchers in the broad areas of computer engineering, embedded systems, control systems, and formal methods. The group is in the School of Electronics and Computer Science, which is highly ranked in the UK and worldwide in electrical engineering.
Related references by the supervisor
- Neural-network-based state estimators
- Communication-efficient Reinforcement Learning
- Learning for Networked Control Systems
