Research Strand 2:
Securing the “Operations” Environment
Autonomous Systems (AS) differ from classical complex systems due to their multi-function intelligent capability, that consequently manifest as different attack surfaces. While existing works secure specific functionalities, ascertaining the overall AS threats across the multiple attack surfaces requires a different research process.
Our approach is to tackle three inter-dependent AS surfaces (mission, control, communication) and use both: (1) the security foundations of RS1 and (2) integrate human behaviour from RS3 – to develop holistic mitigation strategies. We then integrate our advances to demonstrate their impact across a range of AS and multi-modal mission profiles. Common to all three themes is the concept of cross layer and Networked AS (NAS) threats.
AS pose specific requirements and challenges to the detection and mitigation of cyber security risks and attacks due to their complexity and dynamic characteristics combined with the limited and unreliable network connectivity. The mission surface is the core, where the decisions and execution take place; it is dynamic and sensitive by its definition and verifiable security is of critical importance. This complicates the traditional approach that involves continual monitoring and update with patches, which links closely to the control surface below. We will develop methods and algorithms that reduce the risks and costs associated with these challenges and in turn, improve the reliability and resilience of AS.
Best et al., How to Analyze the Cyber Threat from Drones, RAND Corporation, 2020
Rong, P. Angelov, Stability of Evolving Fuzzy Systems Based on Data Clouds, IEEE Trans. on Fuzzy Systems, 2018
RS2-Theme B: Securing the Control Surface
(Lead:G.Inalhan, Participants: P. Angelov, A. Tsourdos, B. Yuksek.
AS relies on the ability to conduct run time adaptations of control decisions over attacks which can result from information and dynamic environment uncertainties. Specifically, in the context of learning enabled AS, it is crucial for the control system to exhibit self-aware learning in which the boundaries of “safe” state-space and the control space are tracked through their evolution. This is particularly challenging when the system is undertaking dynamic decisions within the AS mission surface.
Bansal et al. “Hamilton-Jacobi reachability: A Brief Overview and Recent Advances.” IEEE Conf. Decision & Control, 2017.
Uzun, M. Demirezen, G. Inalhan, “A Framework for Updating Baseline Aircraft Performance Models – Machine Learning”, IEEE Trans. Aerospace Sys. & Elect., 2020
RS2C focus on the communication and sensory planes of ASs. Here, our research is divided between physical level attacks (Cranfield) and network level attacks (Lancaster).
Background: At the Physical (PHY) level, we know digital security can be derived from both antenna beamforming (codeless defence) [2C-1] and deriving distributed keys from channel state information (code-based defence) [2C-2]. The latter is particularly of interest as it can produce secure cipher keys without a common key pool or sharing keys. Yet, it must observe 3 conditions in the PHY channel, namely: (1) reciprocal to allow decentralised synchronous key generation, (2) dynamic to defence against brute force attacks, and (3) unique to avoid correlated attacks. The challenge is that the
idealised conditions are often not met for ASs especially in open static spaces and airborne spaces.