Distributed Formation Control for Multi-agent Systems and Its Applications

Abstract

Multi-agent systems (MAS) consist of interacting entities, which can work together to solve complex problems that are difficult for an individual agent to possibly achieve. Formation control is a way to achieve collaborations in MAS by changing the motions of each agent and the distribution of the relative positions between agents. In recent years, formation control for MAS, especially for heterogeneous MAS with different entities, has been intensively studied due to its wide range of applications in aerospace, intelligent transportation, and smart logistics. However, truly distributed, and reliable operations of MAS formations are difficult in practice with multiple constraints from interaction and physical systems. For instance, their interactive information is commonly locally incomplete and unreliable due to limited communication capabilities and potential cyber-attacks, and their physical systems are inevitably subject to unmodeled dynamics, dynamic barriers, etc. Therefore, the distributed and robust formation control for MAS is significant, and the transformation from control theoretical discoveries to real-world applications is essential. In this thesis, a series of distributed formation control strategies are developed for heterogeneous multi-agent systems to ensure reliable operations, optimised performance, and flexible collaborations under interaction and physical system constraints. To evaluate the impacts of new strategies in practical systems, these discoveries are applied to autonomous vehicles in time-varying formations for target tracking and patrolling, collaborative collision avoidance, and area scanning. First, formation control problems and methods for MAS are reviewed under twolayer constraints: 1) interaction layer constraints include local information, switching topologies, limited communication resources, cyber-attacks, etc. 2) physical system layer constraints include complex heterogeneous dynamics, multiple disturbances, uncertain even unknown model, limited real-time optimization and computing capabilities, physical barriers, etc. Then, we propose novel distributed adaptive observers, event-triggered mechanisms, and resilient control methods to guarantee the stability and resilience of MAS at the interaction layer. For physical system layer constraints, robust heterogeneous formation control, optimal collision avoidance algorithm, and reinforcement learning-based model-free control strategies are provided to ensure safe operations, optimized performance, and flexible collaborations among different agents in dynamic environments. To address the practical collaborative problems, the developed control methods are applied in autonomous vehicles to perform collaborative tasks by dynamic formations. The results demonstrate the effectiveness, robustness, and resilience of the proposed strategies.