Multi-Agent Control and Estimation for Assured Autonomy against Adversaries in the Cyber and Physical Domains (resubmission of 20-PAF00728))
Abstract
The goal of this research program is to increase autonomy of air/sea systems in adversarial environments by establishing assured safety and security against adversar-ial agents and their spatiotemporal effects in the cyber and physical domains. Major challenges lie in (i) capturing complex adversarial actions and their effects in a sufciently abstract level, and in (ii) enabling scalable and, ideally, non-conservative safety and security solutions. The spatiotemporal (i.e., state- and time-dependent) behavior of the adversaries might not jeopardize safety at all times but rather be active for sufciently long time intervals so that con resolved after a certain point. Hence the main problem to be addressed is: What are the spatiotemporal ad-versarial behaviors that can jeopardize the safety of a multi-agent system? What communication structures ensure that the multi-agent system states can be safely reconstructed over specied time horizons? What controllers can ensure safety of the multi-agent system trajectories in the pres-ence of spatiotemporal attacks? What estimators can ensure that malicious information can be discarded within specied time horizons without violating safety? To answer these questions we plan to develop (i) novel semantics for signal temporal logics to capture complex spatiotemporal specications, and link them with the theory of switched and hybrid systems, (ii) secure communi-cation graphs and ltering techniques that securely reconstruct the information and specications values in prescribed-time horizons, and (iii) computationally-efcient safety controllers that fully counteract or tolerate the effects of adversarial specications.The project will innovative results that will advance the state-of-the-art in the science of autonomy by enabling safe and secure networks in the presence of spatiotemporal adversarial behavior. The theoretical and algorithmic developments apply to and will advance military mis-sions such as autonomous surveillance in adversarial, uncertain environments (e.g., multi-domain battleelds or uncharted areas). The Navys mission and applications such as Swarmboats or LOCUST will greatly benet from advanced networks of autonomous vehicles with increased intelligence and autonomy, that meet certain safety and security guarantees. More importantly, the proposed formulation that integrates spatiotemporal specications, safety control and secure communication in a common framework, constitutes an innovative paradigm that merges for the rst time adversarial actions in both the cyber and the physical domains. A major advantage is that safety and security guarantees can be obtained in the proposed framework without knowing who the adversarial agents are. In addition, the innovations of the intermediate tasks will have signicant impact in the elds of multi-agent control and estimation: The development of novel representations of spatiotemporal behavior will enable complex networks with increased levels of safety, autonomy, and intelligence. The development of secure communication graphs and lter-ing techniques against spatiotemporal adversarial behavior will be a new direction towards secure control systems. The proposed QP formulations using prescribed-time control barrier functions offers new directions in the computationally-efcient synthesis of safety-critical and time-critical controllers. Finally, the experimental implementations of a cyber-capture mission with small UAVs both indoors and outdoors will offer novel validations of the developed theory and algorithms as a proof-of-concept, which in the future might be implemented under even more realistic conditions.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 29, 2020
- Source ID
- N000142012395
Entities
People
- Dimitra Panagou
Organizations
- Board of Regents of the University of Michigan
- Office of Naval Research
- United States Navy