Beyond Optimization: A Multi-Agent Systems and Human–Agent Collaboration Perspective on Decentralized Multi-UAV Task Allocation

Decentralized task allocation among multiple Unmanned Aerial Vehicles (UAVs) is a critical NP-hard combinatorial optimization problem for enabling autonomous swarms to operate without central supervision. Multi-UAV research has advanced distributed algorithms to manage physical constraints, dynamic networks, and battery limitations. However, previous work frequently reduces task allocation to abstract numerical cost minimization, largely overlooking the critical socio-cognitive coordination dynamics both among UAVs and with human partners. Furthermore, previous surveys often blend centralized and decentralized architectures, leaving a gap for a dedicated, decentralized multi-agent systems (MASs) perspective. This review addresses these limitations by framing decentralized multi-UAV task allocation as a cooperative coordination problem rather than a simple optimization task. First, we establish a taxonomy that categorizes the literature into four foundational approaches—market-based, learning-based, distributed optimization, and clustering—comprehensively mapping their characteristics and trade-offs. Second, we bridge the multi-UAV literature with broader MAS research, highlighting how socio-cognitive models can address the current optimization-versus-coordination mismatch. Third, we contrast the dominant supervisory control approach of human–UAV collaboration (HUC) with the reciprocal, peer-level teaming models found in human–agent collaboration (HAC). We advocate for the integration of socio-cognitive factors present in cooperative coordination, such as dynamic trust calibration, shared mental models, hybrid intelligence, and ethics, within the allocation loop. Finally, we present a future research roadmap focusing on richer task structures, socially aware rational agents, mixed-initiative allocation, and multi-dimensional process-level coordination metrics. This review provides a holistic understanding for investigating how decentralized multi-UAV task allocation can evolve from rigid supervisory hierarchies toward resilient, adaptive, and human-aligned cooperative teaming models.