Cooperative Control of Multi-Agent Systems

An Optimal and Robust Perspective

Jianan Wang, Chunyan Wang, Ming Xin, Zhengtao Ding, Jiayuan Shan

• Publisher: Academic Press
• Year: 2020
• ISBN:
  9780128204450
• Topic: Tecnología e ingeniería
• Subtopic: Ingeniería mecánica

Cooperative Control of Multi-Agent Systems: An Optimal and Robust Perspective reports and encourages technology transfer in the field of cooperative control of multi-agent systems. The book deals with UGVs, UAVs, UUVs and spacecraft, and more. It presents an extended exposition of the authors' recent work on all aspects of multi-agent technology. Modelling and cooperative control of multi-agent systems are topics of great interest, across both academia (research and education) and industry (for real applications and end-users). Graduate students and researchers from a wide spectrum of specialties in electrical, mechanical or aerospace engineering fields will use this book as a key resource.

• Helps shape the reader's understanding of optimal and robust cooperative control design techniques for multi-agent systems
• Presents new theoretical control challenges and investigates unresolved/open problems
• Explores future research trends in multi-agent systems
• Offers a certain amount of analytical mathematics, practical numerical procedures, and actual implementations of some proposed approaches

Part One

About cooperative control

Outline

1. Introduction

2. Preliminaries

1

Introduction

Abstract

In recent years, cooperative control of multi-agent systems has been witnessed as an attractive research area to many researchers from both academia and industry. In this chapter, we first introduce the background of cooperative control and the general overview of multi-agent coordination. Then, we present some related work in this area including consensus and its applications, such as formation control and flocking, etc. In particular, some future research topics are reviewed for readers' reference. In the end, the objective of this book is summarized and a brief outline is also included.

Keywords

Cooperative control; Multi-agent coordination; Optimal and robust control

1.1 Background

Multi-agent coordination is an emerging engineering field multi-disciplined by many areas as shown in Fig. 1.1. The concept of multi-agent coordination is initially inspired by the observations and descriptions of collective behaviors in nature, such as fish schooling, bird flocking and insect swarming [126]. Fig. 1.2 shows one example of fish schooling. Figs. 1.3 and 1.4 show examples of birds flocking and ‘V’ formation. These behaviors may have advantages in seeking foods, migrating, or avoiding predators and obstacles, and therefore the study of such behaviors has drawn increased attention from researchers in various fields [87]. In 1987, three simple rules – separation (collision avoidance), alignment (velocity matching) and cohesion (flock centering) – were proposed by Reynolds [151] to summarize the key characteristics of a group of biological agents. After that, a simple model was introduced by Vicsek [179] in 1995 to investigate the emergence of self-ordered motion in systems of particles with biologically motivated interaction. The flocking behaviors were later theoretically studied in [70,127,167,173].

Figure 1.1 Interdisciplinary fields of multi-agent coordination. Acronyms: AE – Aerospace; BIO – Biology; COM – Computer; CYBER – Cybernetics; PE – Photoelectric; COMM – Communication; INFO – Information; PHY – Physics.

Figure 1.2 Fish schooling.

Figure 1.3 Birds flocking.

Figure 1.4 Birds flying in ‘V’ formation.

There are many robotic control ideas coming from biological societies. For example, one of them is to use simple local control rules of various biological societies – particularly birds, fishes, bees and ants – to develop similar behaviors in cooperative robot systems. In [66], a number of generations of robotic fishes have been designed for navigation in a 3D environment based on the biologically inspired design. In [7], a number of algorithms have been developed for tracking, recognizing, and learning models of social animal behaviors.

1.1.1 Motivations

In recent decades, several researches have been engaged in the multi-agent coordination problem. The motivation of these researches is to discover the benefits compared with single-agent systems. First, it can reduce cost and complexity from hardware platform to software and algorithms, i.e., one large and expensive robot can be replaced with several small and cheap robots on task implementation with lower cost and complexity. Second, multi-agent systems are capable of many tasks which could not be effectively performed by a single-robot system, for example, the surveillance task. Moreover, multi-agent systems with decentralized control have preferred flexibility and robustness and can reduce the signal communication and computational workload by using local neighbor-to-neighbor interaction.

The development of multi-agent systems is also well supported by the technological advancement in sensor, communication, and control. As smaller, more accurate, and more reliable sensor and communication systems are available, the cooperative strategies of multi-agents to carry out certain tasks become possible and applicable [18].

1.1.2 Control architectures and strategies

For multi-agent systems, various control architectures have been proposed in literature. Most of them can be described as centralized and decentralized schemes. In centralized systems, a central unit that connects all the agents has the global team knowledge, and manages information to guarantee the achievement of the mission. Thus, advanced and expensive equipments are necessary to satisfy all the technological requirements. For decentralized schemes, all the agents are in the same level and have the same equipments. Each agent uses the local sensor to obtain the relative state information of its neighbors, then makes decision for the next step to move and explore the environment. Furthermore, each agent does not need the global information and just communicates with their neighboring agents.

Centralized and decentralized schemes have their own advantages. Regarding the centralized one, the powerful central unit can highly improve the overall performance of the multi-agent systems. Furthermore, the excellent computing capability and high-speed communication ability of the processor can send the command to all the agents quickly and effectively. On the other hand, the whole system highly relies on the central unit. The failure of the central unit will lead to the failure of the whole mission. The robustness of the centralized scheme is insufficient. Moreover, high requirements on the central unit lead to high cost of the whole system. While for decentralized systems, using low-cost sensors and processors to replace the expensive core unit, can reduce the cost effectively. The motion of the agent only relies on the local relative information of neighbors, which reduces the difficulty level of the mission. In addition, decentralized systems are more tolerant to severe environment since failure of partial agents does not affect the performance of the whole system. On the other hand, the decentralized systems rely on more complex control strategies to coordinate and optimize the execution of the mission, which limits the performance of the system. The communication bandwidth and quality limits also affect the overall performance.

1.1.3 Related applications

Cooperative control has broad potential applications in real world...