Advancements in Adversarially-Resilient Consensus and Safety-Critical Control for Multi-Agent Networks

Abstract

The capabilities of and demand for complex autonomous multi-agent systems, including networks of unmanned aerial vehicles and mobile robots, are rapidly increasing in both research and industry settings. As the size and complexity of these systems increase, dealing with faults and failures becomes a crucial element that must be accounted for when performing control design. In addition, the last decade has witnessed an ever-accelerating proliferation of adversarial attacks on cyber-physical systems across the globe. In response to these challenges, recent years have seen an increased focus on resilience of multi-agent systems to faults and adversarial attacks. Broadly speaking, resilience refers to the ability of a system to accomplish control or performance objectives despite the presence of faults or attacks. Ensuring the resilience of cyber-physical systems is an interdisciplinary endeavor that can be tackled using a variety of methodologies. This dissertation approaches the resilience of such systems from a control-theoretic viewpoint and presents several novel advancements in resilient control methodologies. First, advancements in resilient consensus techniques are presented that allow normally-behaving agents to achieve state agreement in the presence of adversarial misinformation. Second, graph theoretic tools for constructing and analyzing the resilience of multi-agent networks are derived. Third, a method for resilient broadcasting vector-valued information from a set of leaders to a set of followers in the presence of adversarial misinformation is presented, and these results are applied to the problem of propagating entire knowledge of time-varying Bezier-curve-based trajectories from leaders to followers. Finally, novel results are presented for guaranteeing safety preservation of heterogeneous control-affine multi-agent systems with sampled-data dynamics in the presence of adversarial agents.