Security of autonomous-vehicle-team tracking dynamics: a control-theoretic and graphical perspective

Wei Wang

Recently, distributed cooperative control of multiple autonomous vehicles, including for military reconnaissance, environmental protection, and health care applications, has become a very active research topic. In many of these applications, teams of autonomous agents with distributed sensing capabilities are required to cooperatively complete a tracking task. In addition, under the most general conditions, the autonomous vehicles systems operate in harsh, changing, and even adversarial environments. Hence, the security issue of AVNs has begun to become important. Although there are many papers related to information-security in the domain of communication networks, no serious attention seems to have been given to the role of communication and sensing among the network components and to the physical dynamics of these networks, which we think is central to AVNs control. On the other hand, control theorists' viewpoints on security/robustness almost always neglect that the complex and hybrid spatio-temporal characteristics of disturbance and adversaries may have an effect on AVNs. Hence, we are motivated to develop new definitions and analysis techniques for the security of AVNs which can cover both aspects. And then we expect to develop a systematic way to make the system secure or more secure by properly designing the gains of decentralized controllers in each vehicle using the methodology in this thesis. First, we abstract the AVNs and adversaries to a canonical model. In the model, we assume that each vehicle is governed by identical dynamics (double integrator dynamics), and the whole network dynamics is largely determined by the sensing capabilities of the AVNs. The task of an adversary is to estimate the initial condition of the AVNs by observing the position of one or a small set of vehicles. Whether or not the adversary can do this, and how well, is the security problem. Based upon the model and problem formulation, we develop several results. In the time invariant model, when the adversary's observation is noise-free, it is shown that we may derive the security of AVNs directly from the control gain of distributed controller and vehicle sensing capacity from both control-theoretical and graphical perspectives. In addition, when adversary observation is noisy, we define a security level matrix to present how secure the system is and use the knowledge of the control gain of distributed controller and vehicle sensing capacity to bound the value of security level matrix (performance of estimator of adversary). Furthermore, in time varying case, based upon the results before, we give some understandings of security for a special class of time varying systems.

Security of autonomous-vehicle-team tracking dynamics: a control-theoretic and graphical perspective

Files and links (1)

Abstract

Metrics

Details