Multi-Agent Systems for Collaborative and Distributed Decision-Making: Design and Applications

The hard distributed problems that no one agent or centralized system can solve are increasingly looked at problems that can be addressed using multi-agent systems. Independent agents collaborate with other independent agents pursuing either the same or similar objectives. Focusing on the basics and frontiers of MAS, this survey uses structures and mechanisms of coordination and communication as points of reference. First, we cover the routes of coordination and communication that are traditional: voting, negotiation, consensus, and auction, the most primary forms of agent interactions. Then, we cover the routes of coordination and communication that are contemporary: agent communication languages. Next, we look into the main architectural styles used in MAS: hierarchical, heterarchical, and holonic, along with newer hybrid methods and recent approaches based on large language models [2], all aimed at addressing certain orchestration challenges through global planning and local execution [1]. To display MAS realization, we present four representative domains: smart grids, autonomous transport, supply chains, and disaster response. In each domain, multiple agent teams come together—for instance, smart meters coordinate power consumption, self-driving vehicles negotiate route formation, factories adjust schedules for production, and rescue drones form coalitions—demonstrating both the opportunities and unique challenges posed by MAS. System-level challenges become prominent across these domains. Scalability to large agent populations, security and privacy in open networks, explainability of agent decisions, and conflict resolution are some of the major research challenges. Recent surveys observe that privacy-aware cybersecure multi-agent frameworks [3] and privacy-preserving consensus protocols should be developed.