Theory and Algorithms for Safe and Secure Dynamic Multi-Agent Networks
Abstract
Publically Releasable: The goal of this career-building project is to increase the existing levels of autonomy of networked multi-agent systems in adversarial environments by fundamentally considering both the safety and the security aspects of a dynamic multi-agent network in a unified framework. In contrast to the existing literature that focuses individually either on safety or on security of cyber-physical multi-agent systems, here we propose to: (1) explore the fundamental conditions that establish both safe and secure (resilient) operations in a dynamic multi-agent network in the presence of adversaries, and (2) develop control and estimation (filtering) methods that enable mission accomplishment via safe motion planning and coordination of multiple agents, along with their secure self-awareness and interactions with their neighbors. Towards this end we will leverage: (i) our extensive expertise in planning, coordination and control of multi-agent systems in complex environments, and in particular our novel concept of semi-cooperative interactions, and our recent work on parametric Lyapunov-like barriers that capture uncertainty in information flow, and (ii) our recent work on k-circulant communication graphs for secure networks, that establishes a novel class of resilient communication topologies for multi-agent systems under the consensus protocol. Based on these concepts we aim at exploring the fundamental conditions that establish concurrent safety and security in the network, and at developing distributed control and estimation methods that enable accomplishment of missions beyond those that can be framed under the consensus paradigm. More specifically, we propose to formally establish the novel notion of safe and secure semi-cooperative interactions: the conditions and mechanisms under which a dynamic multi-agent network can tolerate malicious information and behaviors without violating safety margins, for missions beyond consensus-based network control. Our overarching goal is to connect the physical, sensing, and communication capabilities of an agent through this concept, and to develop distributed control and estimation methods that render a multi-agent network capable of dynamically deploying in, and interacting with, an uncertain environment (neighbor agents, physical obstacles, possible adversaries). The project will produce innovative results that will push the state-of-the-art in the science of safe and secure dynamic multi-agent networks. The envisioned theoretical and algorithmic developments are applicable to a wide range of military applications such as autonomous surveillance, active exploration, and search-and-rescue in adversarial, uncertain environments (such as the battlefield or uncharted areas), as well as to civilian applications in the aerospace, automotive and robotics fields, such as integrating small UAVs safely and securely in the National Airspace, enabling safe and secure self-driving cars, and robots that interact with their environment in a trustworthy way. Thus the ArmyÕs mission will greatly benefit from advanced networks of autonomous agents with increased intelligence and autonomy, that meet certain safety and security guarantees. Also, the proposed concept of semi-cooperative interactions that brings the constrained dynamics, the uncertain information, and the diverse tasks of the agents under a unified framework, constitutes an innovative paradigm which, for the first time, tightly links and enables both safety and security for dynamic multi-agent networks. Finally, the experimental implementations with small UAVs both indoors and outdoors will offer an initial proof-of-concept that might in the future be extended to even more realistic conditions. At the educational level, the project will engage graduate, undergraduate, and K12 students, and disseminate the results through publications, workshops, outreach activities activities and onsite demonstrations.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 11, 2018
- Source ID
- W911NF1710526
Entities
People
- Dimitra Panagou
Organizations
- Army Contracting Command
- United States Army
- University of Michigan