Secure control of multi-agent systems under cyber attacks: a comprehensive survey

Multi-agent systems (MASs) rely on open communication networks and are thus exposed to malicious cyber attacks, which not only threaten system performance but also erode the fundamental basis of collaborative operation. This paper takes the novel four-node information flow architecture as the main line, divides the cooperative control loop into four functional layers: perception, transmission, decision-making, and execution, and conducts a systematic and comprehensive review of the latest research progress in the security control of MASs. Under this architecture, we establish a unified attack classification system that captures the mathematical commonalities and operational characteristics of denial-of-service, spoofing, Byzantine, and hybrid attacks. Subsequently, we systematically sorted out the defense strategies along four information flow nodes: the perception layer covers security state estimation and physical layer trust; the transmission layer involves event-triggered, self-triggered, and quantized communication; the decision-making layer includes resilient consensus, observer-based control, adaptive control, and reinforcement learning control; the execution layer focuses on fault-tolerant control, sliding mode control, and impulsive control. A core theme running through the entire text is the safety-performance-resource tripartite paradox that restricts the design of elastic MASs. We examined the path of evolution from passive robust defense to active defense to reconcile these conflicting goals and pointed out open challenges such as cross-layer collaborative design, data-driven resilience, asynchronous multi-channel attacks, and verification on actual platforms. This survey also highlights emerging application fields, such as clusters of intelligent agricultural robots. Under adversarial communication conditions, secure collaboration is crucial for precision agriculture, autonomous harvesting, and field monitoring. By taking a unified perspective that integrates attack modeling and defense mechanisms, this survey aims to provide researchers and engineering practitioners with a systematic and forward-looking reference blueprint for building intrinsically secure collaborative MASs.