A systematic review of the collaborative intelligence in multi-agent scheduling

Traditional scheduling assumes a single actor, while real-world scenarios often involve multiple decision-making units with conflicting objectives, requiring collaboration, coordination, negotiation, and collective intelligence. Multi-agent scheduling (MAS) provides a realistic framework for modelling such interactions under shared resources. This article presents a comprehensive scientometric and systematic review of two-agent and multi-agent scheduling, examining the field’s development paths, knowledge structure, and emerging research clusters. Main path, cluster, and keyword analyses are integrated to trace knowledge diffusion, identify influential studies and research branches, and examine how collaborative intelligence is manifested across scheduling contexts. The findings reveal an evolution from early complexity-oriented single-machine and dual-agent models toward richer production environments, advanced computational methods, and application-oriented MAS in smart manufacturing, energy systems, edge computing, and decentralised project scheduling. More importantly, the synthesis reveals a broader transition from optimizing competing agent-specific objectives toward interaction-driven collaborative scheduling intelligence based on coordination, negotiation, information sharing, adaptation, and collective learning. This domain-level perspective suggests that future MAS should integrate decentralised decision-making, adaptive intelligence, and system-level coordination to reconcile local objectives with system-wide performance.