SAIR: Safe and Secure AI for Robotics

Welcome to SAIR Research Theme page!

The National Robotarium and the Edinburgh Center for Robotics identified several high-priority research themes. This is the page of the Research Theme SAIR: Safe and Secure AI for Robotics. The theme has just opened, and is gradually building up.

What motivates our research:

Significant advances in Machine Learning led to its ubiquitous deployment in autonomous systems. However, Machine Learning is a black-box technology; therefore, any system that adopts it potentially jeopardises its safety, security, and interoperability. The significance of this problem has been recognised internationally, giving rise to EU AI Act and The Bletchley Park AI Declaration. For AI and autonomous systems like robots to be trusted we have ensure that they work for society in a responsible, verifiable, ethical, legal and safe manner.

SAIR Theme Co-leads:

• Prof David Aspinall (UoE),
• Prof Ekaterina Komendantskaya (HWU),
• Dr Ioannis Konstas (HWU),
• Dr Marta Romeo (HWU),
• Prof Jacques Fleuriot (UoE),
• Dr Tariq Elahi (UoE)

SAIR Members

Topics we Investigate:

• Building safer and more secure machine-learning components
• Verification of neural networks and robot behaviour
• Robust use of machine learning in security applications
• Formal properties of learning and autonomous systems
• Legal requirements to autonomous systems
• The interaction between human factors (e.g. individual differences in trust) and robotic systems 
• The impact of AI on the future of work
• The economics of robotics: costs/benefits
• Safety/Ethics in Large Language Models, Vision and Language Models​
• Hate Speech and Abusive Language to/from Robotic Agent​
• Miscommunication, Repair and Adaptation in Conversational Interaction​

SAIR Seminars and Events:

All seminars advertised here are open to any interested researcher (you do not need a subscription). Below you will find information on how to join, in face-to-face mode or online. However, if you would like to join our mailing list and receive updates, please fill in this registration form.

Date	Topics
30 January 2025, 10.30 am	Professor Andrew Ireland, Mathematical and Computer Sciences, Heriot-Watt University. Title: Safety Assurance of Autonomous Systems – Mind the Gaps Abstract: The proliferation of autonomous systems brings significant benefits as well as risks to society. In terms of risks, my interests in autonomous systems relate to safety, i.e., accident prevention and mitigation. I am particularly interested in how gaps in knowledge and understanding impact on system safety. My talk will illustrate such gaps through examples from conventionally engineered autonomous systems. Addressing such gaps requires multiple levels of intervention and support. I will focus on tool support, in particular the benefits that strong integration within a toolchain can bring to assuring system safety. I will draw upon work undertaken through the recently completed UKRI Research Node on Trustworthy Autonomous Systems Governance and Regulation. Bio: Andrew Irelandis a Professor of Computer Science at Heriot-Watt University and a founder member of the Dependable Systems Group. He has collaborated with the Mathematical Reasoning Group (University of Edinburgh) for 30 years – supported by six EPSRC Platform Grants, two of which he was PI. His research focuses on automated reasoning and the verification of software intensive systems. Most recently, he was a CoI on an UKRI Research Node on Trustworthy Autonomous Systems Governance and Regulation. He has had organisational involvement with numerous international workshops and conferences, including: the Verified Software: Theories, Tools & Experiments (VSTTE), the IEEE/ACM International Conference on Automated Software Engineering (ASE)and the Automated Verification of Critical Systems workshop series (AVoCS). He was co-chair/chair of ASE 2008, VSTTE 2010, AVoCS 2015. He was co-chair of the Award Committee for the Microsoft Research Verified Software Milestone Award (2011-13). He is a founder member of the IFIP Working Group 1.9/2.15 on Verified Software. His industrial collaborators have included BAE Systems (Warton), Praxis Critical Systems (now Capgemini Engineering), QinetiQ and D-RisQ.
10th December 2024, 11.00 am Informatics Forum and Online	Inaugural SAIR lecture, by Professor Subramanian Ramamoorthy, School of Informatics, University of Edinburgh Title: Safety and Trustworthiness of Human-Centred Autonomous Systems Abstract: With AI systems becoming increasingly more capable, there is growing interest in questions around safety, trustworthiness and explainability. For embodied AI systems such as autonomous systems, especially robots operating in human-centred application domains, these considerations translate into desiderata for AI and systems design.   In this talk, I’ll describe results from a few recent projects addressing learning and adaptive decision making in human-robot interaction contexts, ranging from surgical skills to autonomous driving. I will try to distill learnings from these in terms of a few thematic questions that define my current research agenda.   Bio: Subramanian Ramamoorthy is a Professor of Robot Learning and Autonomy in the School of Informatics at the University of Edinburgh, where he is also Director of the Institute of Perception, Action and Behaviour, and Director of the UKRI AI CDT in Dependable and Deployable AI for Robotics. He holds a UKRI Turing AI World-Leading Researcher Fellowship (2024 – 29). His research focus is on robotics and machine learning, with particular emphasis on achieving safe and robust autonomy in human-centred environments. This work has attracted funding from a variety of sources including UKRI, EU, DARPA, DSTL and the Royal Academy of Engineering, and been recognised with best paper awards at international conferences including ICRA, IROS, CoRL, ICDL and EACL. In addition to his academic role, he has been involved in Five AI, a UK based technology company developing autonomous vehicles technology, as Vice President – Prediction and Planning (2017 – 2020) and Scientific Advisor (2021-23). Five AI was acquired by Bosch GmbH in 2022.