Autonomous underwater vehicles (AUVs) are submersible underwater vehicles controlled by onboard computers. AUV formation is a cooperative control which focuses on controlling multiple AUVs to move in a group while executing tasks. In contrast to a single AUV, multi-AUV formation represents higher efficiency and better stability for many applications, such as oil and gas industries, hydrographic surveys, and military missions, etc. To achieve better formation, there are several key factors, including AUV performance, formation control, and communication capability. However, most studies in the field of AUV formation mainly focus on formation control methods. We observe that the research of communication capability and AUV performance of multiple AUV formation is still in an early stage. It is beneficial to researchers to present a comprehensive survey of the state of the art of AUV formation research and development. In this paper, we study AUV, formation control, and underwater acoustic communication capability in detail. We propose a classification framework with three dimensions, including AUV performance, formation control, and communication capability. This framework provides a comprehensive classification method for future AUV formation research. It also can be used to compare different methods and help engineers choose suitable formation methods for various applications. Moreover, our survey discusses formation architecture with communication constraints and we identify some common misconceptions and questionable research for formation control related to communication.

Multiagent systems (MASs) are distributed systems with two or more intelligent agents. Formation control is a significant control technique of MASs. To date, formation control on MASs is widely used in various fields, such as robots, spacecrafts, satellites, and unmanned aerial/surface/underwater vehicles. However, there is a relatively small body of literature that is concerned with security problems of formation control on MASs in past years. Our research represents the first step toward developing security attacks of formation control on MASs. Our study aims to investigate potential security problems of formation control on a multirobot system for the first time. We propose two kinds of control-level attacks and each kind of attack includes several specific attack forms. Then, we discuss specific features of formation control on a classical multirobot system and utilize theoretical analyses to illustrate how cyberattacks can influence the physical movements of robots. The experimental results of the proposed attacks show that attacks can easily interrupt formation movements of a multirobot system and several carefully designed attacks even can cause irreversible loss.

In this article, an anti-attack event-triggered secure control scheme for a class of nonlinear multi-agent systems with input quantization is developed. With the help of neural networks approximating unknown nonlinear functions, unknown states are obtained by designing an adaptive neural state observer. Then, a relative threshold event-triggered control strategy is introduced to save communication resources including network bandwidth and computational capabilities. Furthermore, a quantizer is employed to provide sufficient accuracy under the requirement of a low transmission rate, which is represented by the so-called a hysteresis quantizer. Meanwhile, to resist attacks in the multi-agent network, a predictor is designed to record whether an edge is attacked or not. Through the Lyapunov analysis, the proposed secure control protocol can ensure that all the closed-loop signals remain bounded under attacks. Finally, the effectiveness of the designed scheme is verified by simulation results.

In this article, an optimal control scheme against false data injection (FDI) attacks for multiagent systems with unavailable system states is presented. The output data that remains uncorrupted is extracted from a group of output measurements during FDI attacks using a secure preselector. Then, on the basis of the unattacked output data, a state observer is utilized to identify system states. By taking the communication overhead into account, an event-triggered mechanism is employed to alleviate communication burdens. The solution of the optimal control problem related to event-triggered Hamilton–Jacobi–Bellman (HJB) equations is obtained within a simplified adaptive dynamic programming (ADP) framework. The key point is that a single network is introduced to successfully avoid the problem of repeated approximations in traditional dual networks. Note that the weight vectors in the single critic network are adjusted through experience replay (ER), which helps to avoid the restrictive persistence of excitation (PE) conditions. It is strictly proven that all the signals in the closed-loop system are uniformly ultimately bounded, and the critic network weight can converge to optimal values. Simulation results illustrate the effectiveness of the control scheme.

As As an effective multi-agent system (MAS) control method, formation control is widely used in uncrewed systems, such as uncrewed surface/underwater/aerial vehicles and spacecrafts. However, the security issues of formation control have received little attention. Compared with traditional industry systems (such as smart grids or intelligence appliances), designing countermeasure approaches for distributed formation systems under attacks has several challenges, such as limited local information, robot mobility, and a chain reaction of attacks. In this paper, based on a classical formation control law on a group of robots, we concentrate on maintaining an acceptable formation performance under Man-in-the-Middle-based (MITM-based) displacement attacks. The MITM-based displacement attacks can utilize the property of formation maintenance and hijack the whole robot group. We propose two kinds of novel countermeasure approaches. An active approach, which is named the comparison-based identification and elimination (CIE) algorithm, can identify attack messages and calculate estimate values to replace attack messages based on local information. A passive approach can tolerate attack effects by designing a task priority adjustment (TPA) controller. The TPA controller can gradually adjust the formation maintenance priority to degrade attack effects. Simulation with several nonholonomic differentially driven mobile robots is conducted, and results show that our schemes can significantly decrease the impact of constant and time-varying attacks.