UC Irvine

With the advances in low-cost reliable electronic devices, autonomous multi-agent systems play an important role in a wide range of applications, such as sensing networks, smart grid, smart transportation, and exploration in hazardous situations. In recent years, factors such as avoiding a single failure point, demand for privacy preservation, and opting for lower computation and communication costs have created the expectation that the autonomous multi-agent systems should be operated in a distributed manner with no central control. In many operations, coordinating tasks among autonomous multi-agent systems involve some forms of distributed leader-following problems. That is, it is expected that a group of networked autonomous agents, normally referred to as followers, should use their local information and local interactions with their neighboring agents to determine their actions so that the entire network achieves a desired system-level behavior that depends on the state(s) of single or multiple leaders. When the dynamics of the leader(s) is unknown, e.g., in target tracking problems, and only a subset of the agents can measure the state(s) of the leader(s) online, the limited information increases the challenge to meet leader-following objectives. The focus of this dissertation is providing practical leader-following solutions that require the least possible assumptions on the dynamics of the leader(s). More specifically, we consider four types of leader-following objectives.

The first problem addressed is a single leader-following problem for a group of heterogeneous linear time-invariant followers, where a subset of the followers has access to the state of an unknown leader in only specific sampling times. We propose a distributed control that uses a minimum-energy control framework to enable the followers to arrive at the sampled state of the leader by the time the next sample arrives. The next problem we consider is a containment control problem, a leader-following problem with multiple leaders, where a group of mobile agents aim to stay in the convex hull spanned by a group of moving leaders with unknown dynamics. Our proposed distributed control enables a group of networked unicycle mobile agents to track the convex hull of the leaders. This algorithm requires the agents to only communicate with each other in discrete-time fashion. The innovation in our containment control design is to model the problem in the form of an active average consensus problem, for which we also propose a novel distributed solution both in continuous-time and discrete-time form. Active average consensus by itself constitutes the third leader-following problem that we study, in which a set of networked agents aim to track the average of the dynamic signals measured by the active agents. The fourth leader-following problem we consider in this dissertation is a coverage problem via a set of mobile agents for a group of dense dynamic targets whose distribution in the space is not known a priori. For this problem, we propose a novel distributed coverage control that first uses a distributed estimation process to enable all the agents to obtain an estimate of the targets' distribution when only a subset of agents in the network can observe the targets. Then, we develop a distributed deployment solution that enables the agents to re-position themselves in a way that their collective quality of service (QoS) distribution is in close accordance with the estimated density distribution of the targets. In our setting, the agents are heterogeneous in the sense that their spatial QoS distribution is different. We demonstrate our results for event detection via sensor networks and UAV-aided wireless communication coverage problems.