Imaging, Data and Communications
Research Theme
Multi Agent Systems and Data Intelligence
Aim
To develop a principled framework for representing, communicating, and aggregating uncertainty in large language model (LLM)–based agent systems, thus improving coordination, calibration and safety in multi-agent reasoning.
Objectives
- Elicit Uncertainty: Develop and compare methods for extracting calibrated uncertainty estimates from individual LLMs.
- Represent & Communicate: Design message formats that carry uncertainty information between agents (e.g., probabilistic statements, confidence intervals, prediction sets).
- Propagate & Aggregate: Establish mathematically sound rules for combining and updating uncertainty as messages pass between agents.
- Evaluate Impact: Create benchmarks and metrics to quantify how uncertainty-aware communication affects accuracy, calibration, and robustness of multi-agent systems.
Description
State of the art solutions in many complex problem domains are often now “Agentic” systems in which collaborations of specialist LLM-based agents collectively solve problems. For example, in cyber-security, surveillance agents may monitor logs to detect potential exploits, and cooperate with patching agents that fix vulnerabilities, and testing agents that validate the results. In such systems agents communicate through natural-language messages that reflect single “best guesses.” In probabilistic terms, these are point estimates of a model’s belief. As messages are exchanged, uncertainty information is lost, leading to over-confidence, error cascades, and poor coordination under ambiguity.
This project investigates how to restore and maintain epistemic and aleatoric uncertainty throughout an agentic pipeline. The first stage will focus on discovering the most reliable way to elicit and perceive uncertainty estimates with single LLMs. The second stage will formalize uncertainty-aware message formats—for example, by attaching confidence scores, probabilistic distributions, or prediction sets to text outputs. The third stage will develop algorithms for propagating and aggregating uncertainties across agents (eg Bayesian, Dempster-Shafer, conformal fusion). Finally, we will build benchmarks that measure team-level uncertainty calibration. Expected outcomes include (i) theoretical understanding of uncertainty composition in LLM collectives, (ii) practical mechanisms for safer, more transparent AI collaboration, and (iii) open-source tools for uncertainty calibration and propagation in agentic frameworks.
Minimum entry qualification - an Honours degree at 2:1 or above (or International equivalent) in a relevant science or engineering discipline, possibly supported by an MSc Degree. Further information on English language requirements for EU/Overseas applicants.
Full funding is available for this position.
Research Theme
Aim
This project aims to significantly reduce the manual labelling of underwater data (images or acoustics) using other available data sources via transfer learning, while understanding the uncertainty of both the learned transfer functions and the uncertainty in the resulting predictions by adapting a Bayesian framework.
Objectives
- Literature review on Bayesian transfer learning (BTL)
- Develop underwater BTL framework, incorporating known physical effects (such as wave distortion by water); data source can be e.g. imagery or acoustics, depending on the student’s and project partner’s interest
- Analyse existing underwater data set and test methodology
- Develop online BTL algorithm that can process data in real time
Description
Modern machine learning methods require large amounts of training data for reliable predictions, and can degrade significantly in performance when the test data differs from the training data distribution. In transfer learning, one has access to a source data set (e.g. vessel sounds in UK coastal waters) as well as a small amount of data from a target data set for which data collection is costly or difficult (e.g. acoustic signature of unknown foreign vessels in remote waters). The goal of this project is to develop an underwater Bayesian transfer learning framework, which allows for rigorous uncertainty quantification of both the transfer function and the resulting predictions – i.e. we can understand both how well we can expect to perform on our target data in general, as well as how certain we are about predictions for specific observations.
Minimum entry qualification - an Honours degree at 2:1 or above (or International equivalent) in a relevant science or engineering discipline, possibly supported by an MSc Degree.
A solid background in statistics or a related quantitative discipline is of benefit. The student will be expected to work in the intersection of statistics, machine learning, and marine science, and should thus be interested in interdisciplinary work.
Further information on English language requirements for EU/Overseas applicants.
Full funding is available for this position.
Research Themes:
Aim
We will develop fast hardware to deploy real-time Bayesian experiment design to optimise the performance of quantum sensors
Objectives
- Integrate a GPU with a programmable real0time arbitrary waveform generator (QuantumMAchines OPX+) to create a real-time control system capable of fast Bayesian inference and experiment design, with millisecond-scale latency
- Integration of hybrid “grey-box” models (combining an analytical model with data-driven inclusion of imperfections) with Bayesian experiment design and deployment on the hardware system in (1)
- Benchmarking of the system of different quantum sensing modalities
Description
Quantum sensors represent the generation-after-the-next technology for sensing, exploiting the properties of quantum particles (such as the spin of a single electron) to achieve atomic-scale spatial resolution and/or unprecedented sensitivities.
Despite the demonstrated advantages of quantum sensors in applications ranging from defense to healthcare, they are often limited by weak signal (resulting in slow data acquisition and low bandwidth), fragile operation in well-controlled laboratory environment and difficulty of use by non-experts.
This project will develop world-first adaptive quantum control hardware, enabling the generation of microwave quantum control sequences with microsecond latency based on GPU-accelerated Bayesian inference and experiment design. The system will deploy real-time Bayesian experiment design to optimize the performance of a spin-based quantum sensor in terms of speed and accuracy, and integrating hybrid analytical/data-driven (grey-box) approaches to improve robustness in hostile and changing environment. Our work will enable the deployment of quantum sensors in real-world applications, such as defense, and could be further extended to different types of “classical” sensors.
Minimum entry qualification - an Honours degree at 2:1 or above (or International equivalent) in a relevant science or engineering discipline, possibly supported by an MSc Degree.
Further information on English language requirements for EU/Overseas applicants.
Home rate fees and stipend are available for this position.
Research Theme
Multi-agent Systems and Data Intelligence
UK Nationals Only
Aim
To research adaptive multiagent decision-making in complex multi-sensor (land and naval) platforms for practical applications.
Objectives
- Understand requirements of decision making in complex, multi-sensor (land and naval) platforms.
- Design hierarchical multiagent reasoning and learning framework for adaptive decision-making in multi-sensor platforms.
- Implement and evaluate the framework and algorithms with simulated and real-world data and platforms.
Description
This thesis research project seeks to develop an AI framework that will enable reliable and efficient decision making by humans operating complex, multi-sensor (land and naval) platforms. This framework will support hierarchical multiagent reasoning and reinforcement learning at different levels of abstraction and over long time horizons, under open-world uncertainty and resource limitations. It will automatically identify and use relevant sensor streams (e.g., camera, LIDAR, GPS, acoustic, meteorological) and commonsense domain knowledge from human experts, presenting information such that it reduces the cognitive burden on the human crew. We will illustrate and evaluate our framework in simulation and on physical data streams from multi-sensor platforms of interest to our project partner.
Industry Partner
This project is co-funded by Thales UK.
Placements possible at Thales Glasgow to understand more about the work of the Digital Crew team, subject to appropriate clearances being in place.
Degree 2(i) or better in a computing discipline such as Computer Science, Engineering, Mathematics, or Physics.
UK national
Full funding is available for this position.
Research Theme
Aim
Develop new particle filter solutions for tracking of interacting multiple targets.
Objectives
- Develop new interacting particle filters that can tackle multi-target tracking in a scalable manner
- Understand the hidden low dimensional structures within multi-target systems to enable high dimensional scalable Monte Carlo filtering
- Explore how these models can further be used to predict intent for groups of cooperating and/or adversarial targets.
Description
High-dimensional multi-target tracking poses severe computational challenges for sequential inference. Although particle filters provide a principled Monte Carlo framework, their efficiency collapses with dimension. This project aims to identify and exploit hidden low-dimensional structures within multi-target systems, enabling adaptive filtering methods that retain Monte Carlo flexibility while mitigating the curse of dimensionality. We will develop local particle filters, block-based filtering, and topology discovery methods to automatically detect weakly interacting groups of targets. We may consider scenarios involving tens to a few hundred targets. At this stage, we will primarily use simulated data, with the flexibility to incorporate real data from partners in later stages. The outcome will be scalable algorithms for structured high-dimensional tracking, combining statistical efficiency with real-time feasibility in sensing and defence applications.
Minimum entry qualification - an Honours degree at 2:1 or above (or International equivalent) in a relevant science or engineering discipline, possibly supported by an MSc Degree.
This project requires a student with a high level of mathematics or numeracy.
Further information on English language requirements for EU/Overseas applicants.
Full funding is available for this project.
Research Theme
Aim
This project consists of developing single and multi-object detection (and ultimately tracking) algorithms from neuromorphic camera data.
Objectives
- Identification of tractable (front-end) network architectures able to capture spatio-temporal patterns in spiking data
- Discrimination of spatio-temporal signatures of objects of interest from dynamic clutter.
- Efficient fusion of neuromorphic and frame-based imaging sensors for real time tracking of fast-moving objects.
Description
Neuromorphic cameras have enabled low-cost and fast imaging by only recording asynchronously intensity changes. However, efficient processing of the massive streams of polarised events remains a challenge for real-time computer vision tasks such as object detection. Moreover, platform motion also generates dynamic clutter events that can impede object detection and tracking. In this project, we will investigate artificial architectures trained to encode event streams into low temporal resolution frames while also capturing high-resolution spatio-temporal features (e.g., periodicity). Such features can then be used to improve detection and classification performance (e.g., for UAV detection) or platform motion estimation (for enhanced simultaneous localisation and mapping). Finally we will investigate how neuromorphic cameras, operating at different spatial and temporal resolutions can be fused with traditional frame-based cameras (multimodal fusion).
Minimum entry qualification - an Honours degree at 2:1 or above (or International equivalent) in a relevant science or engineering discipline, possibly supported by an MSc Degree. Further information on English language requirements for EU/Overseas applicants.
Home rate fees and stipend are available for this position.
Research Themes
Aim
To develop a software and hardware system that uses real and predicted motion data to simulate how mobile satellite terminals behave, enabling realistic, lab-based testing for defence communications using Heriot-Watt’s ground station.
Objectives
- Develop an IMU-based system to record real motion data from vehicles, such as cars, UAVs, or marine platforms.
- Design a software package that uses this motion data to control an antenna positioner and simulate real-world movement.
- Create machine learning models that can detect motion patterns and predict future movement to improve tracking accuracy.
- Test and evaluate satellite terminals under simulated motion using Heriot-Watt’s ground station to measure performance in realistic defence scenarios.
Description
This project focuses on improving how satellite communication systems for defence are tested and developed. Mobile platforms such as military vehicles, UAVs, and naval vessels require satellite links that remain stable under motion and in harsh environments, such as urban canyons, mountainous terrain, or contested electromagnetic environments with jamming or multipath interference. Testing these systems in the field can be expensive and hard to repeat. This PhD will develop a lab-based testing system that uses real motion data, recorded from vehicles using inertial sensors (IMUs), to control a robotic antenna positioner (hexapod). The system will also use machine learning to recognise patterns and predict future motion, allowing more advanced testing of how terminals behave. By using Heriot-Watt’s satellite ground station and hexapod equipment, the system will allow realistic and repeatable testing of terminals for defence use, helping to improve resilience, tracking, and performance before deployment.
Scientists to develop next-generation satellite user terminal for affordable global connectivity
UK technology consortium for Mobility and Autonomy Market User Terminal (MAMUT)
Minimum entry qualification - an Honours degree at 2:1 or above (or International equivalent) in a relevant science or engineering discipline, possibly supported by an MSc Degree. Further information on English language requirements for EU/Overseas applicants.
Home rate fees and stipend are available for this position.
Research Themes
Aim
To develop ambient backscatter communication systems enhanced with AI-driven signal processing for secure, low-power data exchange in defence environments, enabling covert sensing and communication among autonomous platforms.
Objectives
- Design and optimise ambient backscatter transceivers for secure, energy-autonomous data exchange in contested RF environments.
- Develop machine learning algorithms for adaptive signal detection, classification, sensing and interference mitigation.
- Integrate secure communication protocols suitable for emissions-controlled and covert operational scenarios.
- Build and evaluate a prototype system demonstrating real-time performance and resilience in representative defence settings.
Description
This PhD project builds on previous pioneering work (LORAB) in ambient FM backscatter to enable secure, battery-less wireless communications in defence scenarios. These systems reuse existing RF signals (e.g. FM, cellular) for data modulation, producing signals that are several dB below the carrier and nearly undetectable, making them ideal for stealth operations. Their ultra-low power operation allows them to run on supercapacitors and energy harvested from the environment. The project will enhance CSS-based modulation and machine learning for improved range, security, and adaptability. With the current interest in integrated sensing and communications (ISAC)[JT1] , this work will also explore possible sensing functions of such signals. The goal is a portable, autonomous sensor tag capable of silent communication over hundreds of metres, using simple SDR-based receivers. This silent and thermally low-signature technology enables covert sensing of the environment or personnel without revealing the location or activity of the node. Its resilience will be tested under jamming and interception conditions, providing a low-profile solution for secure battlefield and intelligence applications.
Relevant references:
- Daskalakis, S. N., et al., “Ambient FM Backscatter: Towards Long-Range, Zero-Power IoT Communication,” Sensors, MDPI, 2022. https://daskalakispiros.com/files/daskalakispis_mdpi_2022.pdf
- Kimionis, J., Daskalakis, S. N., et al., “Backscatter Communications for Wireless Connectivity in IoT,” Nature Electronics, 2021. https://daskalakispiros.com/files/kimionis_nature_2021.pdf
- Daskalakis, S. N., et al., “Dual-Polarized Backscatter Sensor for RF and Wireless Applications,” IEEE Transactions on Microwave Theory and Techniques, 2018.
https://daskalakispiros.com/files/daskalakis_mtt_2018.pdf
- D. Galappaththige, et al., "Integrated Sensing and Backscatter Communication," in IEEE Wireless Communications Letters, vol. 12, no. 12, pp. 2043-2047, Dec. 2023.
Minimum entry qualification - an Honours degree at 2:1 or above (or International equivalent) in a relevant science or engineering discipline, possibly supported by an MSc Degree. Further information on English language requirements for EU/Overseas applicants.
Home rate fees and stipend are available for this position.
Research Theme
Aim
Enabling anti-spoofing, anti-jamming receivers for current and future PNT systems
Objectives
- Extract unique radio frequency fingerprints (RFF) of GNSS satellites for spoofing resilient PNT receivers.
- Develop efficient multi-chain signal processing algorithms or AI tools to identify malicious parties in complex electromagnetic environment.
- Design autonomous null-steering RF frontend for anti-jamming PNT receivers.
- Showcase the security performance of developed PNT receiver in both lab and field testing, without compromising PNT accuracy.
Description
The increasingly dependent on GNSS-enabled positioning, navigation, and timing (PNT) in both military and civil applications demands more robust GNSS receivers which are resilient to both unintentional and intentional interference.
In collaboration with Spirent, which leads the global market for GNSS test equipment and emulation software (SimGNSS), we aim to develop both hardware- and software-enabled security approaches to PNT services. This includes 1) exploiting minute and unique radio frequency (RF) frontend differences as unclonable ID for genuine GNSS satellites to combat spoofing attack; 2) developing array-based autonomous beamforming designs, and associated efficient signal sensing and processing capabilities to reject potential malicious jamming and spoofing parties; and 3) integrate and prototyping security-enhanced GNSS receivers for practical demonstration.
It is importance to mention that these GNSS system challenges are recently documented in the ESA ITT (see the links below), highlighting the urgent needs in defence sectors.
- https://esastar-publication-ext.sso.esa.int/ESATenderActions/details/140407
- https://esastar-publication-ext.sso.esa.int/ESATenderActions/details/89700
- https://esastar-publication-ext.sso.esa.int/ESATenderActions/details/159629
Minimum entry qualification - an Honours degree at 2:1 or above (or International equivalent) in a relevant science or engineering discipline, possibly supported by an MSc Degree. Further information on English language requirements for EU/Overseas applicants.
Home rate fees and stipend are available for this position.
Research Themes
Aim
Research on the algorithm development for automatic target detection, tracking and characterisation using event camera and spike neural networks.
Objectives
- Space Situation Awareness with Neuromorphic Systems
- Onboard processing with SNN
- Going beyond detection based on object shapes
- Object characterisation using micro-vibration
Description
Space Situational Awareness (SSA) has become a necessity in this congested space era. The fact is that finding moving objects, which may or may not be threats, in space is like finding a needle in a haystack. Neuromorphic imagers capture only changes in scene luminance and are quiescent when there is no change. This massively reduces data generation, but it requires a different processing chain to perform the task. Neuromorphic processing is power-efficient for embedded implementations of event data.
This project investigates neuromorphic processing for SSA applications using event-camera data. The aim is to use real data to characterise how objects of interest can be detected and characterised using shape, motion-track, and micro-vibration features. This can provide a step change over current capabilities, in which most processing occurs on the ground, and help avoid the bandwidth bottleneck of transmitting full imagery to Earth, and real-time detection may be possible.
A UK first class, or equivalent international degree on Electrical Engineering, Physics, Computer Science, Mathematics or similar.
Further information on English language requirements for EU/Overseas applicants.
Full funding available for this position