PhD Studentship: Resilient Federated Learning for Autonomous Systems under Distribution Shifts

Description

Applications are invited for a fully funded PhD studentship in the School of Electrical and Electronic Engineering at the University of Sheffield, in collaboration with the Defence Science and Technology Laboratory (Dstl).

Due to funding restrictions, the position is open to candidates eligible for UK home student fees. It offers an enhanced tax-free stipend of approximately £25,000 per year, subject to annual increases.

About the Project (Background & Methodology)

Autonomous systems such as drone fleets, mobile robots, and sensor networks increasingly use federated learning (FL) to train shared models without centralising raw data. In practice, however, individual platforms operate under diverse environmental conditions, sensor calibrations, and system configurations. These differences introduce distribution shifts that can degrade model performance and reliability over time, particularly in “one-to-many” supervision settings where a single human operator oversees multiple agents. Ensuring resilience in such scenarios is critical for safety-critical applications including environmental monitoring, infrastructure inspection, and emergency response.

This PhD will develop a mathematical and algorithmic framework to assess and improve the resilience of FL-enabled autonomous systems under such heterogeneity, explicitly incorporating the human in the loop. The project will draw on information theory, machine learning, and control to analyse how local variability, sensor drift, and platform differences affect both global model performance and human supervisory factors such as workload and intervention behaviour.

Building on this, the research will design robust FL algorithms and adaptive supervisory strategies, including dynamic aggregation, confidence-based thresholds, and escalation mechanisms. A comparative study across civil and defence scenarios, using real and synthetic data, will identify both generalisable and domain-specific resilience mechanisms.

School of Electrical and Electronic Engineering at the University of Sheffield. A leading centre for machine learning, robotics, and autonomous systems. The student will join a research group working at the interface of machine learning, control and information theory, with opportunities to collaborate with partners in robotics, autonomous systems and AI safety. Access to modern computing facilities and experimental platforms (e.g. robotic testbeds or simulators) will be available depending on the final focus of the work.

Defence Science and Technology Laboratory (Dstl). As the Ministry of Defence (MOD)’s in-government science and technology organisation, Dstl provides unique expertise, insight and innovation to maintain UK warfighting readiness in an increasingly dangerous and complex world. As MOD science and technology leaders, Dstl provides expert advice, analysis and capability across a wide range of applications including Robotics & Autonomous Systems, AI and Data Science.

Eligibility and Desired Background

Applicants should hold (or expect to obtain) a first‑class or strong upper‑second‑class degree, or a Master’s degree, in a relevant discipline such as Control/Systems Engineering, Electrical or Electronic Engineering, Computer Science, Applied Mathematics or a closely related field. A strong mathematical background (probability, linear algebra, optimisation) and proficiency in programming (preferably Python/Matlab) are essential. Prior exposure to one or more of: machine learning, reinforcement learning, robotics/autonomous systems, information theory, or human–machine interaction will be an advantage.

Inquiries. Interested candidates are encouraged to contact Dr Iñaki Esnaola or Dr Morgan Jones by email to discuss the position informally and should include a brief CV detailing their suitability for the role. Formal applications should be made through the University of Sheffield application portal, and include a CV and covering letter.

Advert information