Attack-Resilient Designs


This research area focuses on security challenges related to adversarial actions such as spoofing/corrupting sensor information, disrupting actuation, affecting communication between agents, or compromising agents involved in the mission. Specifically, efforts will focus on distributed methods for attack-detection and identification and security-aware mission planning that exploit the dynamics of the environment and agents, as well as our knowledge of the expected mission evolution (e.g., interaction between agents, sensing quality). Such methods include resiliency against worst-case scenarios with coordinated attacks (i.e., unlike work on fault tolerance where faults are assumed random or uncorrelated, we make no assumptions about malicious interference signals introduced via compromised components/agents). We develop a platform-aware attack-resilient architecture integrating the control and estimation techniques for resource-constrained autonomous systems (e.g., including analysis outcomes focused on intermediate iterates of their computations). Developing such architecture is complicated due to the different assumptions on attack surfaces as well as communication and computation resource requirements. We will develop methods to assure desired control performance even in systems with communication and computation limitations, as well as composition of secure control components that ensures optimal balance between control performance and security guarantees while satisfying constraints of the employed distributed platforms. We will investigate MDPs, probabilistic timed automata (PTA), and stochastic hybrid automata (SHA), as a means to model the interaction between the intrusion detection system and the controller/environment.


Attack-Resilient Design Publications