18

Research topics 2021

18

19

Topic 1 An investigation of 5G technology and the threats it presents for the security community and identification of countermeasure opportunities

Topic 2 Intelligent, distributed, dynamic software defined RF spectrum sensor network for detection and identification of devices in a dense RF environment

Topic 3Does utility monitoring improve or compromise security? Considering the associated dilemma of carbon neutral agenda, balanced with the need to maintain security of assets and generated data

Topic 4 Machine learning inversion

Topic 5 Digital cities/countries for intelligence and investigative purposes

Topic 6 Cybersecurity model for visible light communications (VLC)

Topic 7 Cybersecurity of swarm robotics in smart cities

Topic 8 Satellite IoT communications

Topic 9 Mathematical approaches to complex imaging problems

Topic 10 Machine led discovery of novel materials for automated chemical synthesis

Topic 11 Predicting the unpredictable: Can you predict drone intent?

Topic 12 What would Socrates think? The legal and ethical implications behind autonomous drones and future aviation

Topic 13 Low shot training and testing of machine learning algorithms for detection of items of concern

Topic 14 Eddy diffusion modelling for enhanced hazard assessments of exposures to airborne toxic materials

Topic 15 Improving energy harvesting in IoT wireless sensor nodes

Topic 16 Automated intelligibility tests through the use of AI or novel algorithms

Topic 17 Detecting use of synthetic biology methods

Topic 18 Cyber influence on behaviour change: prevalence, predictors, progress, and prevention

Topic 19 Autonomous control for small uninhabited air vehicles enabling monitoring of infrastructure

Topic 20 Quantum engineering for quantum sensors

Topic 21 THz RF transmission for wideband atmospheric communications

20

Topic 1 An investigation of 5G technology and the threats it presents for the security community and identification of countermeasure opportunities

Key words: 5G; millimetre wave; radio physics; technical security; wireless sensing; pattern recognition

Research topic description, including problem statement:Modern and future wireless technologies, such as fifth generation (5G), are utilizing increasingly higher frequencies extending into the millimetre wave and beyond with their associated ability to support higher information bandwidths. The commercialization of this technology is leading to the availability of low-cost RF sub systems and components at these higher frequencies. This will mean, we will see an increase in technical threats operating at these higher frequencies, which will be produced at a significantly lower cost with easy deployment. This will mean the public, businesses and infrastructure are vulnerable to cyber-attack. There is also a use case within airport security and screening techniques used at border checkpoints. Therefore, we need to investigate what these threats are so that we a) detect their presence and b) put effective counter measures in place to protect the public, businesses, and national infrastructure.

The aim of the research is to:• Explore how these frequencies and waveforms interact with electronic systems at a

fundamental level• Adapt 5G technology sub systems to demonstrate: - The technical surveillance vulnerabilities posed by these - Their application for countermeasures to detect threats - Provide advice and guidance to protect the public, businesses and national

infrastructure and enhance security screening at airports/ border checkpoints.

Example approaches:• There is a growing area of research that examines security and privacy concerns, identifying

attack methods and identifying countermeasures to offer greater protection from such attacks. For example, there has been research to discover how audio from loudspeakers can be recovered from soundproof buildings due to the subtle disturbances they cause to RF transmitters such as widely available such as Wi-Fi. The research identifies the risk and then describes how to protect against this potential attack method. (Reference: “Acoustic Eavesdropping through Wireless Vibrometry” Teng Weiy, Shu Wangy, Anfu Zhou and Xinyu Zhangy University of Wisconsin - Madison, Institute of Computing Technology, Chinese Academy of Sciences)

• Research has also produced a series of portable screening prototypes mm Wave sensing. With potential applications for screening in prisons or at airports. (Reference: E-Eye: Hidden Electronics Recognition through mmWave Nonlinear Effects Zhengxiong Li1, Zhuolin Yang1, Chen Song1, Changzhi Li2, Zhengyu Peng3, Wenyao Xu1 1-CSE Department, SUNY University at Buffalo, Buffalo, NY, USA 2-ECE Department, Texas Tech University, Lubbock, TX, USA, 3-Aptiv Corporation, Kokomo, IN, USA)

21

Topic 2Intelligent, distributed, dynamic software defined RF spectrum sensor network for detection and identification of devices in a dense RF environment

Key words: RF (Radio Frequency); sensors; software; distributed networks; pattern recognition

Research topic description, including problem statement:The rise of the Internet of Things has seen an increased use of wireless technologies to provide connectivity between devices. These platforms are vulnerable to various types of attack and authentication of devices, spoofing and detecting unauthorized transmissions are a constant challenge. Some progress has been made to address this through device fingerprinting, which identifies unique elements specific to a device. However, more work is needed to provide greater security to our wireless network particularly in a dense RF environment, where detection of malicious activity is challenging. This could make it even more challenging to secure environments for legitimate devices.

This topic seeks to understand the dynamic RF landscape and build upon previous research to detect and identify a specific radio among similar devices in a dense environment and catalogue these accordingly. The ambition is to achieve an intelligent sensing capability that can detect all devices operating in a dense RF environment and define its fingerprint as legitimate or unauthorized adding it to a classifier. This will provide better security for the public spaces from malicious activity. This will also benefit the security community better protect their environments and could be of use to detect unauthorized devices in places such as prisons.

The aim of the research is to:• Build upon existing research in this field and develop a prototype for practical use to detect

threats within the dynamic RF landscape • Develop a means of identifying unauthorized devices through effective fingerprinting.• Develop a classifier to identify unique signatures for devices that is robust enough to work

in a dynamic environment.• Develop counter measure approaches for unauthorized devices, such as denial of service or

location to enable recovery/ proactive investigations.

Example approaches:Some pioneering early investigative work examined the concept of radio fingerprinting, detecting specific devices in a distance ranging between 2ft to 50ft using deep learning convolutional neural networks. This has also built upon previous research examining device fingerprinting in wireless networks.

22

Topic 3Does utility monitoring improve or compromise security? Considering the associated dilemma of carbon neutral agenda, balanced with the need to maintain Security of assets and generated data

Key words: cyber security; energy analytics; carbon friendly; psychology; patterns of life; estate management; IoT security

Research topic description, including problem statement:The UK Government has set out its agenda for carbon neutrality and the focus has steered towards creating carbon efficient products for building construction/ upgrade and secondly on greater use of utility monitoring and modifying behaviours of the associated occupants of buildings to reduce energy consumption.

This topic seeks to understand the associated security risks (and opportunities) to be gained from utility monitoring. Does this enable us to have greater security of buildings or not? Furthermore, do we understand if there are any security risks associated with the new recommended building materials.

Utility monitoring/ fabrication of materials is part of the on premise IoT landscape and any development in this environment should ideally offer security assurance.

The aim of the research is to:• Build upon existing research in this field to understand the security risks• Develop counter measure approaches to protect our assets and data generated• Provide a balanced risk approach with implementation of monitoring devices (data)• Cognizance/ awareness of security risks associated with introduction of new carbon

efficiency construction materials or devices into the property.

Example approaches:• A commercial example is smart metering of utilities and making recommendations to

achieve a reduction of energy usage or more efficient means of usage by users or occupants within the monitored environment.

• Potential for identifying ‘pattern of life’ via simple utility monitoring as a positive side benefit to assist elderly or vulnerable persons living alone. Pattern of activity, inactivity.

23

Topic 4Machine learning inversion

Key words: machine learning; artificial intelligence; model inversion

Research topic description, including problem statement:Can a machine learning model be inverted to reveal the data it was trained on?

Machine learning (ML) is ubiquitous in the current landscape. It is often operating in uncharted territory in terms of ethics and governance.

In supervised ML the model learns from an existing data set where the answer is known. This may potentially use sensitive data to train on.

To what extent can this data be exposed across the many ML techniques? What data is currently at risk and what threats and opportunities does this pose? Attempts to recreate data from existing models has been published under the title model inversion but is not an established field.

If risk is established what are the mitigating steps and what costs would they have? Example approaches: Publicly available models and data sets could be used to test problem. This would enable work to be carried out with few initial barriers.

24

Topic 5Digital cities/countries for public safety

Key words: smart cities; digital twins; intelligence; subthreshold; modelling

Research topic description, including problem statement:A digital twin is a virtual recreation of any systems. With the future rollout of Smart Cities this presents the opportunity for incredibly detailed mapping of an entire town/City function. This includes electrical systems to traffic flows, to pedestrian footfall. Visual mapping will likely become prevalent with the cameras and sensors on autonomous and connected vehicles. A National Digital Twin has already been proposed by Cambridge University’s Centre for Digital Britain.

Digital Twins of cities already exist, notably in China. Such detailed mapping will be powered by IoT (Internet of Things) and 5G.

Opportunity:It seems highly likely that creating digital twins of cities and even the entire country, would allow for extremely accurate mapping and modelling of events in real time, drawing in data from a range of open source and classified material to support decision making and planning from a public safety perspective. The combination with AI would allow for the efficient deployment of emergency services to likely hotspots, identify high risk areas and also the testing of variables to understand and predict reactions within a city to events, whether natural or otherwise.

Possible steps:Creation of a digital twin as a case study for law enforcement and the emergency services, incorporating the relevant information and data streams. This would also require mapping of what data is available and timely. An Agile approach would likely work best, small sprints producing results that layer on one another. For example, twin a street, then a university campus, then a borough etc.Incorporation of behavioural analysis and relevant AI. For example, predicative aftershock analysis (earthquakes) has been trailed by police to predict future crime hot spots.

Such a volume of data could be exploited by malicious actors, so it is important to understand the vulnerabilities to digital twins and the threats posed by their misuse. The security of such information should also be considered.

Example approaches: Such a project will require a collaborative approach as it incorporates very technical data but would also need detailed behavioural analysis drawing from a range of open source and government data.

25

Topic 6Cybersecurity model for visible light communications (VLC)

Key words: visible light communications; Li-Fi

Research topic description, including problem statement:Visible Light Communications continues to develop, with several Li-Fi products being considered for use. However, we are missing a propagation model for visible light communications that can be used as the foundation for cybersecurity risk modelling at the physical layer. Such models exist for conventional radio, but these do not reliably extrapolate to VLC because the physics of light interaction is different to radio. There are plenty of analyses for propagation on co-operative paths (i.e., luminaire to user), but none that we have seen that deals with luminaire to interceptor. The lack of basic science around cybersecurity risk modelling means that network design using VLC is currently ‘vernacular’ – that is, based on hearsay, speculation, and opinion. We need a solid reliable foundation that allows VLC networks to be designed to a known cybersecurity risk.

Example approaches: Indoor propagation model, from the perspective of an adversary, that can be re-expressed as an intercept risk model.

26

Topic 7Cybersecurity of swarm robotics in smart cities

Key words: robotics; smart cities; swarms

Research topic description, including problem statement:Robotic systems may be used in smart city infrastructure and resource supply (water, power, logistics, etc), smart manufacturing, unmanned transportation systems, and as robotic assistants and companions. This is IoT operating dynamically in uncontrolled environments off-grid, or with low bandwidth backhaul, or within restricted, isolated and/or constrained physical environments. Issues in addition to conventional fixed IoT include assurance in the field of robot operation to avoid harm, accident, or malicious performance degradation; tamper detection on moving/transforming machinery; corruption of machine perception; perversion of machine personality & behaviours; reliable secure mobile communications (mesh/multimedia) in hostile environments; implications for personal data creation & handling (in medical/personal applications, etc) 

This research should consider the implications of cyberattack on robotic individuals and swarms that are programmed to exhibit personalities (e.g., innate decision preferences in response to risk) and behaviours (e.g., situation-adaptable actions and problem solutions). It should consider the ways in which personality and behaviour may be implemented, and the ways in which this is vulnerable to corruption through cyberattack. A key question is the roles of robotic wisdom (contextual judgement) and ethics (as countermeasures).

Example approaches: Demonstration of the types of cyber vulnerability of currently researched robotic behaviour and personality models, and mitigation strategies using robotic wisdom and ethical models.

27

Topic 8Satellite IoT communications

Key words: satellite; internet of things (IoT); LoRaWan

Research topic description, including problem statement:The Internet of Things (IoT) has now reached space, and start-up companies such as Lacuna are hoping to roll out a service using modified LoRaWan that will allow users to transmit data to a satellite from low power remote devices in locations that lack terrestrial infrastructure. Applications amongst others include asset tracking (including vehicles, aircraft, and vessels), wildlife conservation, climate change monitoring, situation awareness for disaster relief, policing and border control.

The concept of worldwide universal IoT connectivity from remote locations normally not serviced by terrestrial networks is potentially a game changer for so many applications however this scheme will only offer one-way communication from the ground to the spacecraft and the initiative is predicated on a modified stack/silicon so the IoT devices must be specific to space transmission. This research topic aims to explore the theory, practicality, and limits of operating native IoT communications waveforms for bi-directional IoT communications to and from a low Earth orbit satellite. This focus on ground-based technologies for satellite IoT will investigate radio waveforms and protocol designs, maximising exploitation of entropy sources for secure cryptographic communications, and constraints from necessary power saving/harvesting and ‘wake-up’ designs. Optimisation is for power efficiency and endurance, and effective exploitation of channels with very low link budgets. Low gain antennas with limited efficiency can be assumed to be a real-world constraint of any practical system.

Example approaches: As an example of possible inclusion in the research, ultra-narrowband as typified by SIGFOX devices use a very low power transmitted waveform which coupled with digital processing gain techniques are achieving communications over many tens of kilometres in terrestrial applications. The questions the research would be addressing is could such a waveform be used in a space application? What are the limits to its use given the constraints of link budgets through the atmosphere and the effects of Doppler?

28

Topic 9Mathematical approaches to complex imaging problems

Key words: inverse problems; sensing; explosives detection; landmine detection; SAR imaging; LiDAR; non-destructive testing; imaging; acoustic sensing; radar; electro-optics

Research topic description, including problem statement:We are looking to apply new mathematical approaches to our sensing problems. We would like to investigate the application of inverse problem solutions to data from a variety of traditional sensors such as radar, electro-optic (EO), infrared (IR) and X-ray with the aim of improving our ability to sense through complex media or multiple paths (e.g., through foliage, walls, fog, and round corners). We are also interested in the fusion of data from different modalities to improve solutions and information from inverse imaging problems.

The Defence, Security and Intelligence communities require capability development of novel game-changing sensing modalities. Science & Technology (S&T) is growing faster than ever before and has become a new domain of international competition. In order to counteract this, a variety of sensing modalities are being used to give us insight for Intelligence, Surveillance, Target Acquisition and Reconnaissance (ISTAR) purposes.

The physical phenomena occurring due to the interaction of Electromagnetic (EM) waves and the scene adds to the complexity of the sensing problem. For example, in Synthetic Aperture Radar (SAR), multi-path reflections cause difficulty in imaging through obscurants, commonly urban and foliage areas. In EO/IR systems similar issues arise, particularly at low photon flux, and further out in the EM spectrum, X-ray and Gamma-ray technologies are used in non-destructive testing for security purposes.

Classically, the image formation techniques for various sensing modality datasets have been limited to the data processing tools which were highly dependent on computational power and storage. Due to the rise in efficient Size, Weight and Power (SWAP) sensors and computing technology, it has allowed the development of processing and image formation tools which were once deemed impossible. All such sensing modalities would benefit from a Fellowship in the area of developing and implementing tools for processing data that incorporates non-canonical imaging techniques to give Security and Defence the added advantage it requires. Furthermore, data fusion from multiple sensors can be exploited so that a single platform can do the work of multiple platforms and complement other sensing data, e.g. SAR data can be fused with EO/IR data. A non-exhaustive list of government departments interested in sensing modalities is mentioned hereunder:• Multistatic SAR Imaging and Ground-Penetrating Radar (Ministry of Defence [MoD]);• EO/IR Imaging (Home Office [HO] / MoD / Department of Transport [DfT]);• Quantum (LiDAR) Imaging (HO/ DfT/ MoD);• Acoustic Imaging (UK Hydrographic Office [UKHO] / MoD) – Ocean Acoustic Tomography;• Magnetic Imaging (Atomic Weapons Establishment [AWE] / National Nuclear Lab. [NNL] /

MoD / HO / DfT);• X-ray Tomography (MoD / HO / DfT)

29

Example approaches: Due to storage of data and faster computational processing power, the general trend in data processing technique developments is to exploit the rich nature of the datasets. Consider that next generation SAR is attempting to process multi-dimensional datasets (i.e., multi-static, multi-channel, multi-look and multi-polar) to form better quality images so that information is retrieved which wasn’t apparent before. [3D SAR Imaging for Multistatic GPR, M. Pereira et al, SPIE Digital Library, 2019, DOI - https://doi.org/10.1117/12.2519430]. The paper [A. Horne et al., "Exploration of Multidimensional Radio Frequency Imaging to Derive Remote Intelligence of Building Interiors," 2018 International Conference on Radar (RADAR), Brisbane, QLD, 2018, pp. 1-6, doi: 10.1109/RADAR.2018.8557263] also provides an example of a situation where inverse techniques may provide benefit.

Elsewhere in the EM spectrum, inverse problems in LiDAR imaging is found in [W. Marais, R. Holz, Y. H. Hu and R. Willett, "Atmospheric lidar imaging and poisson inverse problems," 2016 IEEE International Conference on Image Processing (ICIP), Phoenix, AZ, 2016, pp. 983-987, doi: 10.1109/ICIP.2016.7532504.]. This describes an approach to an atmospheric lidar photon-limited imaging problem in which observations are contaminated with Poisson noise. [Gariepy, G., Tonolini, F., Henderson, R. et al. Detection and tracking of moving objects hidden from view. Nature Photon 10, 23–26 (2016). https://doi.org/10.1038/nphoton.2015.234] uses LiDAR imaging to track moving objects around corners.

Baggage inspection based on X-ray imaging has been established to protect environments in which access control is of vital significance. In several public entrances, like airports, government buildings, stadiums and large event venues, security checks are carried out on all baggage to detect suspicious objects (e.g., handguns and explosives). This is an ever-increasing field of research. See the IEEE Spectrum article, “Future Baggage Scanners Will Tell Us What Things Are Made Of”.

A Through-Wall Radar Imaging (TWIR) PhD is to be completed next year, which has demonstrated some fascinating image processing schemes by exploiting synthetic data using radio frequency propagation models and DNNs (Machine Learning) at University of Manchester. Two EPSRC iCASE PhD studies (part funded by DSTL) are to start in October 2021 at Universities of Cambridge (Towards Scalable EM Solvers on High Performance Computer [HPC] Platforms) to support Full-Waveform Inversion (FWI) and Manchester (Tensor Tomographic Imaging of Foliage Penetration [FOPEN]) to support the Detection, Tracking, Recognition and Identification (DTRI) of difficult targets through foliage (dense vegetation). The PhDs further highlight the interdisciplinary nature of the topic and the relevance of a Fellowship to provide focus and drive.

30

Topic 10Machine led discovery of novel materials for automated chemical synthesis.

Key words: machine learning; artificial intelligence; neural networks; automated discovery; novel materials; automated chemical synthesis

Research topic description, including problem statement:The interest in utilizing theoretical science to predict new materials and synthesis pathways has been long standing. Computing power has accelerated in the 21st century and corresponding advancements in Artificial Intelligence (AI) and Machine Learning (ML) have increased. Recent research is applying AI, ML and autonomous systems to the field of chemical synthesis. At present, human-led discovery of new materials through manual practices can take decades of research, significant continuous funding and can result in a high degree of risk. This fellowship would focus on applying machine learning for the discovery of novel materials to increase efficiency and hence reduce cost and risk.

Recent research across academia has led to the creation of software to translate bulk text into low level instructions using natural language processing1. Development in this space allows for the optimization of experiments based on prior experiences and the progress of data-driven materials discovery2. This problem statement is specifically looking at the next stage of the pipeline to identify, implement and validate a method for machine led discovery of materials. This may include the discovery of novel materials or the discovery of novel reaction pathways for conventional or traditional materials. Research in this area would also enable the assurance of material supply and identification of potential new material threats.

Example approaches: Proposals should include the assessment and development of available algorithms for discovery of new materials using machine learning. These should be assessed against the following criteria:• Applicability of algorithms to different material types, such as: synthetic biology, energetics

and other hazardous materials, advanced materials (e.g., smart materials/nanomaterials)• Computational requirements• Requirement for transparent methods and explainable results • Mitigation of risk associated with false alarms • Potential for validation and optimization of the trained model

The most relevant and applicable method(s) should be implemented and validated against applicable data.

1ChemDataExtractor: A Toolkit for Automated Extraction of Chemical Information from Scientific Literature, M. Swaine, J. Cole, https://pubs.acs.org/doi/abs/10.1021/acs.jcim.6b002072A Design-to-Device Pipeline for Data-Driven Materials Discovery, J. Cole, https://pubs.acs.org/doi/10.1021/acs.accounts.9b00470

31

Open-source datasets for model training are available, these include but are not limited to:• PubChem NCBI: https://pubchem.ncbi.nlm.nih.gov/ • ChemSpider (RSC): http://www.chemspider.com/• NIST Webbook: https://webbook.nist.gov/• Crystallography Open Database: http://www.crystallography.net/cod/result.phpThe Cambridge Structural Database (https://www.ccdc.cam.ac.uk/structures/) may also be used; however, this is licensed. It is expected that datasets from different sources may be required to best train a model for a variety of scenarios.

Proposals are also invited to consider future developments in the pipeline such as in-line characterization (e.g. in chemical synthesis) and analysis of predictions.

Dstl will offer technical partnering from the machine learning and chemical synthesis aspect. Dstl can offer testing of models and validation of results against classified data and offer feedback for further development.

32

Topic 11Predicting the unpredictable: Can you predict drone intent?

Key words: autonomy; drones; counter-drones; artificial intelligence; predictive behaviours; patterns; drone swarms; future aviation; unmanned traffic Management

Research topic description, including problem statement:We seek research proposals that will increase our understanding of predictive drone behaviours. The unmanned aircraft (colloquially “drones”) industry is rapidly developing and has the potential to dramatically impact the UK in a positive way. Drones already positively impact UK life; delivering medical supplies to remote islands during Covid-19, supporting our emergency services and as a new leisure activity.

As drone systems become larger and more capable, their potential broadens too. Drones will be critical to shaping the future of urban mobility, revolutionizing commercial operations and how we perceive and utilize airspace. Drones have already proven themselves to be incredibly versatile additions to commercial enterprises. PwC forecast that the drone industry will contribute an extra £42bn to the UK by 2030, making up 1.9% of UK GDP and supporting over 600,000 jobs in the drones’ economy.

Artificial Intelligence will be likely be used in the management of a future Unmanned Traffic Management (UTM) system, and the identification of malicious or illegal drones. The largest advantage to drones of Artificial Intelligence (AI) will be AI’s decision-making ability to achieve autonomous flight. The next generation of drones will be able to complete many complex functions that now require a pilot and crew. AI will allow for multiple drones to cooperate and coordinate in a swarm as a single, semi-autonomous system. Many applications could benefit from the use of swarms. For example, a search and rescue swarm might permit more efficient searching in a complex environment, not only by covering a larger area quicker, but by combining sensor data. However, with autonomous drones comes greater challenge and the need for more advanced counter-drone technologies to detect, track, identify and disrupt drones.

This research topic addresses the challenge of using data sets to distinguish between ‘normal’ and ‘anomalous’ drone behaviour, drones working in silo or as part of an autonomous swarm, and drones being controlled by a real person or by AI. As our skies become busier and drone-use increases it is critical that we learn to distinguish between normal, negligent, and malicious drone-use in order to safeguard the widespread benefits drones can bring, and to effectively mitigate the threat.

This research topic explores the concept of predicting drone intent by analysis of data sets and previous drone behaviour. Can we identify and create a signature library of predictive behaviours in order to accurately predict future drone behaviour and hostile intent? This understanding and predictive ability would inform effective operational response and threat prioritization in the future.

33

Example approaches: Part of this research topic will require a data science understanding of drone patterns of life; a possible approach may include:• Collation and analysis of existing research and known drone data sets, including alignment

with international partners. • Distinguishing between what is normal drone behaviour and what is an anomaly? • Distinguishing between normal, negligent, and malicious drone behaviours.• Distinguishing how or who is controlling a drone: - Is it a real person? - Is it autonomous? - There will be various levels to explore in this part, including a set of waypoints up to

full AI. - Exploration of whether the drone is interacting with other drone in a type of swarm

(these will vary in complexity).• Explore whether there are signs a drone is working in a swarm flying / working in silo. • Explore how to track drone swarm behaviour and then identify predictive behaviours for this. • Explore smart swarms and predictive behaviour analysis on this. • Develop a tool to use this identification of drone patterns and AI to predict, inform and

assess future drone behaviour intent. • Explore possibility of feeding this predictive data into a counter-drone Detect, Track, Identify

(DTI) capability to cue a further counter-drone response to disrupt the drone.

34

Topic 12What would Socrates think? The legal and ethical implications behind autonomous drones and future aviation

Key words: autonomy; drones; drone swarms; counter-drones; artificial intelligence; future aviation; unmanned traffic management; law; ethics; philosophy; human rights; public acceptance; data privacy

Research topic description, including problem statement:We seek research proposals that will increase our understanding of the legal and ethical implications of current, emerging, and future autonomous drones (and other autonomous vehicles).

The unmanned aircraft (colloquially “drones”) industry is rapidly developing and has the potential to dramatically impact the UK in a positive way. Drones already positively impact UK life; delivering medical supplies to remote islands during Covid-19, supporting our emergency services and as a new leisure activity.

As drone systems become larger and more capable, their potential broadens too. Drones will be critical to shaping the future of urban mobility, revolutionizing commercial operations and how we perceive and utilize airspace. Drones have already proven themselves to be incredibly versatile additions to commercial enterprises. PwC forecast that the drone industry will contribute an extra £42bn to the UK by 2030, making up 1.9% of UK GDP and supporting over 600,000 jobs in the drones’ economy.

Artificial Intelligence will be likely be used in the management of a future Unmanned Traffic Management (UTM) system, and the identification of malicious or illegal drones. The largest advantage to drones of Artificial Intelligence (AI) will be AI’s decision-making ability to achieve autonomous flight.

The next generation of drones will be able to complete many complex functions that now require a pilot and crew. AI will allow for multiple drones to cooperate and coordinate in a swarm as a single, semi-autonomous system. Many applications could benefit from the use of swarms. For example, a search and rescue swarm might permit more efficient searching in a complex environment, not only by covering a larger area quicker, but by combining sensor data.

With autonomous drones, however, comes greater challenge and the need for more advanced counter-drone technologies to detect, track, identify and disrupt drones. If you were to have an autonomous drone being detected, tracked, and identified (DTI) by an autonomous counter-drone system, which could autonomously cue a counter-drone system to disrupt that autonomous drone, you begin to unravel many complex questions around potential risk, responsibility, the ethics and legality of AI and true autonomy.

35

This research topic addresses the challenge of needing to look ahead on the topic of autonomy, AI, drones, and counter-drones, to help inform and shape future policy, legislation, and regulation. A key objective of this research topic is to ensure domestic and international legislation, ethics, human rights, safety, and national security requirements are not overlooked during the rapid technological advancement of the drones and counter-drones market.

Similar debates have taken place in the past on autonomous cars, and who the legal responsibility falls to in the case of an incident or accident. There are significant similarities between these arguments and those for autonomous drones, however there are also critical nuance which this research topic will seek to explore.

The social and behavioural angle, including exploration of public perception and trust of AI/autonomous drones and of greater data capture by authorities that monitor drone movements to support incident response, is a further complex component to consider as part of this research topic.

Example approaches: Research proposals could approach this from a variety of disciplines, or as a cross-disciplinary effort. The problem touches on aspects of data science, international and domestic law, philosophy and ethics, human rights, psychology, engineering, human-centred computing, systems and design thinking, software development, with strong links across both physical and social sciences.

A possible approach may include:• Define what the critical questions are that we should be asking now of Government and

Industry.• Review of UK (and International) legislation and regulation in place.• Scope to expand to international legal frameworks.• Explore overlap between autonomous technologies and human rights legislation. • Explore varying levels of human control (e.g. a fully autonomous drone and/or counter-drone

system with no human oversight, or a counter-drone system which identifies and tracks a drone autonomously, but which requires a human to issue the command to disrupt the drone).

• Literature review of parallel issues explored for other autonomous vehicles (I.e., self-driving cars).

• Scope whether the Governance currently in place in the UK (and/or Internationally) needs to be expanded to cater for such developments (e.g., do we need bespoke Government Departments or new Institutions akin to Bretton Woods to promote international cooperation/stability).

• Make recommendations to Government and Industry.

36

Topic 13 Low shot training and testing of machine learning algorithms for detection of items of concern

Key words: machine learning; X-Ray screening; artificial intelligence; low shot learning, low shot testing synthetic imagery

Research topic description, including problem statement:X-Ray scanners are deployed for security screening of bags at various security checkpoints: in airports, other transport hubs, and at critical national infrastructure. There is a great interest to assess the effectiveness of machine learning (ML) algorithms deployed on X-Ray scanners to detect concealed items of concerns.

To date, it has been considered that a large number and variety of images are required to train a ML algorithm for effective threat detection. However, it may not be possible to provide a large image set of items of concern to developers. Therefore, it is of importance to be able to develop effective ML algorithms using a small number of training images – termed low shot training.

It is also of interest to explore the feasibility of independently testing an algorithm with a small test set of images, termed low shot testing. A further option to overcome difficulties with provision of large image sets of images of concern, is to assess the effectiveness of using synthetic imagery.

Example approaches: A small training set of images of items of concern will be provided (open format, not security classified) to the postdoc to be used to develop a low shot trained detection algorithm. A small test set of images will be provided to be used to test the algorithm. A large data set of related images will be provided to train and test an algorithm with a conventional supervised learning approach. A comparison is to be made of the low shot learning approach and the supervised learning approach using the large data set. Similarly, a comparison is to be made of the low shot testing approach with testing using the large test set.

Based on the provided images for items of concern, the postdoc will create synthetic images and develop a study to compare the effectiveness of synthetic images, real images, and a combination of both.

Combine the approaches of low shot learning and synthetic imagery to assess the most effective way to train a ML algorithm for items of concern in the absence of large data sets.

37

Topic 14Eddy diffusion modelling for enhanced hazard assessments of exposures to airborne toxic materials

Key words: modelling; particulate matter; post-event Recovery

Research topic description, including problem statement:Mathematical models are used to calculate the exposure of individuals to airborne toxic material in indoor environments to provide advice to protect civilian populations. Relevant indoor environments include public spaces, critical national infrastructure, and transport platforms. Eddy diffusion models provide spatially resolved concentrations/exposures while being simple to set up and quick to run. These models have advantages over simpler spatially averaged approaches, which can significantly over or under-predict exposures. In eddy diffusion models, a single parameter governs mixing, the eddy diffusion coefficient. Some relationships that enable the eddy diffusion coefficient to be calculated have been described in the open literature. However, approaches taken to derive these relationships have so far been simplistic. There is a requirement that work is undertaken to produce a new relationship between the eddy diffusion coefficient and room ventilation parameters based on fundamental fluid dynamics. This relationship would ideally take account of a range of processes that affect mixing such as the movement of people and thermal stratification. Improved eddy diffusion modelling would enable more accurate and timely prediction of the hazard from airborne toxic material in a range of indoor environments of interest.

Example approaches: • Development of an eddy diffusion modelling based on empirical data. Possible flow

visualization.• Validation of created or existing models against real data

38

Topic 15Improving energy harvesting in IoT wireless sensor nodes

Key words: battery; batteries; electronics; energy harvesting; electrodynamic; electromagnetic; Internet of Things (IoT); Li-ion; lithium polymer; low power; piezoelectric; photovoltaic; PV solar; thermoelectric; triboelectric; wireless sensor nodes (WSN)

Research topic description, including problem statement: The Internet of Things (IoT) is a growing market with predicted worldwide spending in this field to exceed $1.2T by 2022 (Forbes.com). Likewise, a complimentary technology, energy harvesting, is predicted to exceed worldwide spending of $70B by the same year (IDTechEx, Energy Harvesting MicroWatt to GigaWatt: Opportunities 2020 - 2040).

This topic is focused on powering wireless sensor node (WSN) hardware, which are designed to be cheap, low power and easy to install without the need to integrate into existing infrastructure. A major drawback to the use of WSNs is the battery lifetime. The need to balance energy or power consumption, the frequency of device maintenance, and the performance of the sensor inevitably leads to a device with compromised functionality or significant maintenance overheads.

A route to increase the functionality of WSNs is to incorporate energy harvesting into the device. On a large-scale energy harvesting has already been successful, with solar technology allowing many millions of people in Africa access to power where the local electricity grid is inadequate (smart-energy.com). There has been a significant amount of research and development into smaller scale energy harvesters with limited, but some success. Examples are tyre temperature and pressure monitoring systems on vehicles which can be powered from ambient vibrations (Bowen, C. R. and Arafa, M. H. Advanced Energy Materials 2014), the PowerWatch developed by Matrix Industries which utilises thermoelectric (TEG) and photovoltaic (PV) harvesters to power a smart watch (powerwatch.com), and PV powered crop monitoring ensuring farmers can monitor real-time data to improve yields (Telit.com).

The aim of this topic is to address shortcomings in the way WSNs are powered, with the broad goal of increasing the duration of the device by incorporating energy harvesters that generates power in the region of 0.1 - 10 mW, with a minimum energy density of 0.1 mW/cm3. Proposals are welcomed that tackle this goal from a variety of different routes.

Example approaches: • The following examples highlight several strategies to develop effective energy harvesters to

power WSNs:• Development of a full system that can be demonstrated to harvest energy from ambient

conditions, utilised on a vehicle or within a property for example. This should include mechanical housing, power electronics and advanced energy harvester(s). There are many energy harvesting devices available, but the topic authors are technology agnostic. All efficient systems will be considered.

39

• Focus on crossing the ‘valley of death’ to produce higher TRL energy harvesting systems. Several technologies have moved into the high TRL space, such as PV and TEGs, but promising technologies such as RF harvesting, piezoelectric and triboelectric are still mostly limited to low-TRL demonstrators.

• Development of RF energy harvesters and/or ‘power beaming’ technologies that can be used to harvest energy safely at greater than 1 metre separations. Health and safety considerations will be key to a proposal in this area.

• Focus on the fundamentals of improving the materials which can provide a step-change in various energy harvesting areas in the near future. For example, improving the lifetime of organic PVs, improving the electronic conductivity of TEG materials, or broadening the resonant frequency of piezoelectric materials.

• Increase the efficiency of power electronics used to convert harvested energy into useful power. For example, the power conversion of nanoamps of current at several kilovolts typically seen with piezoelectric harvesters results in significant power losses.

40

Topic 16Automated intelligibility tests through the use of AI or novel algorithms

Key words: AI audio testing; speech intelligibility; speech algorithms; testing efficiencies

Research topic description, including problem statement: Aim: to Investigate the ability to perform measurably realistic intelligibility listening tests using AI or novel algorithms to increase the efficiencies of tests and improve the accuracy of repeated tests.

Understanding the impact a device, component or algorithm has on speech intelligibility is vital when designing or buying a product that captures or processes speech. At present there are a few standard techniques to test intelligibility including; Speech Transmission Index (STI), Speech Intelligibility Index (SII) and listening tests. STI and SII are limited in their application and hence do not always accurately reflect the intelligibility impact from a system. The most accurate approach to date involves listening tests. Listening tests are both time and resource intensive meaning their use is often restricted. Over recent years academia have been de-veloping algorithms to calculate intelligibility metrics and some of those metrics have been shown to have a one-to-one mapping with listening test data used by the developers. Wheth-er the mappings remain consistent for different scenarios is currently unknown. Before a new algorithm can be adopted it is important to understand the reliability of the algorithm and understand the circumstances or scenarios in which the algorithm can or cannot be used.

Example approaches: • The researchers may wish to examine approaches and algorithms from different disciplines

including but not limited to data science, acoustics, and hearing aid research.

• Proposals could consider: - Understanding the latest techniques employed in listening tests - Researching a range of approaches and algorithms that exist that relate to speech

intelligibility prediction - Understanding capabilities and limitations of approaches that could be used to re-

duce dependency on listening tests - If mapping between metrics and listening test data the researchers may want to

consider the data set used and may also want to consider performing independent listening tests.

41

Topic 17Detecting the use of synthetic biology methods

Key words: biosecurity; CRISPR/Cas; pathogen; synthetic biology; detection; virus; bacteria; engineering biology; genetic engineering

Research topic description, including problem statement:The development of synthetic biology techniques has broad ranging beneficial applications across a number of areas, including medical, agricultural, and environmental domains. However, as an example of a dual use technology, synthetic biology may also be exploited for the nefarious creation or modification of harmful biological systems and products. This presents a real biosecurity concern for the development and use of this rapidly advancing technology. This has been demonstrated recently by the recreation of the horsepox virus using standard synthetic biology techniques by a Canadian research group interested in smallpox vaccine development

The exploitation of established molecular biology tools such as TALENs, the widespread interest in CRISPR-Cas systems, and the development of novel and emerging techniques such as the LEAPER method are facilitating the process of genetic engineering, rendering it simpler, faster, and more efficient. This brings with it a growing biosecurity risk that pathogens could be recreated or their pathogenic characteristic amplified by a nefarious actor with increasing ease. Identifying and detecting when and where synthetic biology techniques have been used and the impact of their application by applying an understanding of synthetic biology is therefore essential, so as to lead to the development of detection capabilities and support attribution as a countermeasure to their nefarious use.

This proposal seeks to understand how synthetic biology techniques and engineering biology can be identified once used, so as to provide a technical component to support an overarching regulatory framework for the use of synthetic biology.

Example approaches: A possible approach may include: • Reviewing existing and novel synthetic biology techniques and processes e.g. CRISPR/Cas,

LEAPER systems etc. and their applications to pathogen creation/modification.• Developing understanding, characterising, and collating markers of synthetic biology use,

as well as identifying techniques that can be used to detect created or modified pathogens, including those that resemble natural pathogens.

• Developing a set of criteria or a detection framework including required capabilities to identify when synthetic biology has taken place.

• Characterizing the impact of future, emerging synthetic biology techniques and how these may impact detection of use.

42

Topic 18Cyber influence on behaviour change: prevalence, predictors, progress, and prevention

Key words: social influence; persuasion; coercion; social cyber security; narratives; behaviour change; cyber-medicated changes

Research topic description, including problem statement:The aim of this project is to understand and forecast the impact of cyberspace on changes in human behaviour which has implications for social and political outcomes. Social cyber security has been identified as a key area where social behavioural science (SBS) can exchange knowledge and contribute to issues and challenges for the IC (US Decadal survey, 2019). How cyber may mediate behaviour change and the boundary conditions to such influence (‘when and for whom’) are core research areas for the SBS.

There are recently developed models of social influence that draw, in particular, on group-level identity processes tied to common interests, belonging and shared norms. A key step forward would be to investigate the applicability of these models to cyber and their implications for actual behaviour change in the ‘real-world’. Key questions are whether 1) existing models of social influence translate straightforwardly to the cyber environment and if not, how to redefine them accordingly and 2) cyber influence does have a direct relationship to behaviour and when this is most likely to occur. This research area is of broad scope and interest with potential to form a much larger research enterprise.

Example approaches: Proposals could consider the following approaches or perspectives:• Investigate the personal and social characteristics of users (strengths and vulnerabilities)

that are most and least likely to be cyber influenced.• Identify and demonstrate through experimental studies key factors that facilitate and

disrupt cyber influence on users’ behaviour.• Understand the factors that escalate the success of cyber influence to widespread

behaviour.• Explore methods that can prepare and inoculate users to cyber influence tactics and assess

their success.• Examine the underlying personal and social motivations that underpin certain areas of cyber

influence including extremism and promoting actions that present security threats.

43

Topic 19Autonomous control for small uninhabited air vehicles enabling monitoring of infrastructure

Key words: uninhabited air vehicles (UAV); autonomous flight control; autonomous navigation; remote monitoring; national infrastructure; radio frequencies

Research topic description, including problem statement: Having small UAVs conduct inspection and monitoring of national infrastructure (i.e., bridges, railroads, power lines, pipelines, traffic flow, etc.) is becoming more common but if the systems could do the job autonomously (flight control, obstacle avoidance and navigation) it would enable even more frequent and affordable use. This topic is interested in autonomous behaviour for the small UAVs that enable the system to operate safety with a minimum of human intervention.

Example approaches: As referenced in the topic paragraph, inspection and monitoring of national infrastructure is becoming more frequently done by small UAVs, but today that requires human hands-on control or at the very least supervised limited autonomy. In addition, human monitors the imaging (or other collections) and the human often directly controls them.

If the UAV could autonomously fly along electrical power lines scanning for hot spots in the line with an IR imagery. The UAV would also be able to scan for other things that might interfere with the power transmission (e.g., a tree branch that is leaning over the lines). The data could be stored and Geo-registered as well as any anomalies identified for human assessment later, but the autonomous vehicle has to ensure that safety will verifiably be better than the remote-controlled vehicle and the human piloted aircraft.

Another example is bridge inspection looking for damage, corrosion, etc. needing repairs to ensure the structural safety. A different example is traffic monitoring for updating guidance for automobiles on the fastest way to get to work, school, home, hospital, etc. The guidance could be to a human driver or even an autonomous system on the automobile.

44

Topic 20Quantum engineering for quantum sensors

Key words: quantum; quantum engineering; quantum sensors; atomic sensors; machine learning; control theory; quantum control; enabling technology; magnetometer; gyroscope; accelerometer; gravimeter; atomic clock; atom interferometer; NV diamond                                                                                                                               Research topic description, including problem statement:Quantum sensors are devices that encode a physical quantity into a few quantum states of the system—for example, atomic magnetometers, atom interferometer gravimeters, atomic clocks, Nitrogen Vacancy-centre Diamond (NVD) magnetometers, and so on. Quantum sensors may optionally utilize nonclassical states to increase their performance. As quantum sensors become more sensitive and accurate, a key remaining challenge is to make them more practical outside the laboratory. They need to be easy to operate, fast to turn on, robust against vibration and thermal changes, small, and low power. The emerging field of quantum engineering can address these problems by applying standards and new engineering techniques to quantum devices.

Example approaches: Example approaches will depend on the maturity of the quantum sensor and its intended application environment. Some interesting directions include, but are not limited to, using machine learning techniques to simplify the user experience, using quantum or classical control techniques to increase robustness against noise, employing digital signal processing algorithms to increase sensor speed or improve accuracy, and applying advanced packaging techniques to reduce sensor size. These techniques may also be used to improve the performance of enabling technologies for the quantum sensor, such as lasers or photon detectors, but the proposal should then include the use of these enabling technologies in an actual quantum sensor. Proposals may include work on theory, modelling, or algorithms, but must apply these to a quantum sensor in the lab during the first year of the effort.

45

Topic 21THz RF transmission for wideband atmospheric communications

Key words: terahertz; satellite communications; atmospheric distortion

Research topic description, including problem statement: Satellite communications are continually challenged by both data volume and co-channel interference; one solution to these issues have been to move higher into the spectrum (more than a few GHz). However, as frequencies approach 100 GHz and beyond, atmospheric distortion and attenuation challenges can have serious data quality impacts. One advantage to the Terahertz (THz) bands is the emergence of extremely long wave IR light (10 um; 30 THz) versus extremely high frequency radio frequency energy (3 THz; 100 microns). By employing these extremely high bands for ever larger data loads, based on much higher bandwidths, what challenges in transmission (both free space and atmospheric) must be overcome as well as benefits to be developed?

Example approaches: Inter-atmospheric, exo-atmospheric and free-space data transfer modelling and analysis to handle 1000x current data loads.