Distributed AI and Edge Computing Architectures for Autonomous UAV Systems: A PRISMA-Based Review
by HAVADER Editör Ekibi
Every bee in a hive, without a central manager, accomplishes a complex task — feeding the colony — just by interacting with its neighbors. Swarms of unmanned aerial vehicles (UAVs) are trying to reach a similar goal: each vehicle making smart decisions with limited processing power, without staying constantly connected to a central server. This study systematically compiles the approaches the literature has proposed for reaching that goal.
The goal was to systematically review, within the PRISMA protocol, distributed AI algorithms and edge computing architectures for autonomous UAVs operating under resource and communication constraints. Existing reviews mostly focus on physical-layer security; this study instead offers a unified architectural analysis integrating the CAP Theorem (the trade-off between consistency, availability, and partition tolerance in distributed systems), computational complexity, and hardware-software co-design.
Of 50 candidate studies identified from the Web of Science database, 43 were included in full-text analysis, classified along three axes: distributed AI/ML paradigm, system architecture decision, and optimized metric. The results showed FedAvg-based federated learning and deep reinforcement learning as the leading approaches for distributed swarm learning and online resource allocation, respectively. CAP Theorem analysis found that most examined architectures prioritize availability and partition tolerance over consistency — a design choice consistent with adversarial operating conditions.
What this study contributes is pulling together a scattered, fragmented literature into one unified framework, giving future researchers a clear roadmap. An everyday analogy: it's similar to a city's traffic lights managing traffic by communicating with neighboring lights, rather than staying constantly connected to a central control room — each light makes a smart decision with limited information, and the whole system becomes more resilient by not depending on a single central authority.
In the end, this research notes that topics like Byzantine-fault-tolerant federated learning and model compression remain underaddressed in the literature, offering an algorithmic taxonomy and a set of open design problems for autonomous UAV platforms — giving researchers in the field a concrete agenda for which gaps still need filling.
← Blog
The goal was to systematically review, within the PRISMA protocol, distributed AI algorithms and edge computing architectures for autonomous UAVs operating under resource and communication constraints. Existing reviews mostly focus on physical-layer security; this study instead offers a unified architectural analysis integrating the CAP Theorem (the trade-off between consistency, availability, and partition tolerance in distributed systems), computational complexity, and hardware-software co-design.
Of 50 candidate studies identified from the Web of Science database, 43 were included in full-text analysis, classified along three axes: distributed AI/ML paradigm, system architecture decision, and optimized metric. The results showed FedAvg-based federated learning and deep reinforcement learning as the leading approaches for distributed swarm learning and online resource allocation, respectively. CAP Theorem analysis found that most examined architectures prioritize availability and partition tolerance over consistency — a design choice consistent with adversarial operating conditions.
What this study contributes is pulling together a scattered, fragmented literature into one unified framework, giving future researchers a clear roadmap. An everyday analogy: it's similar to a city's traffic lights managing traffic by communicating with neighboring lights, rather than staying constantly connected to a central control room — each light makes a smart decision with limited information, and the whole system becomes more resilient by not depending on a single central authority.
In the end, this research notes that topics like Byzantine-fault-tolerant federated learning and model compression remain underaddressed in the literature, offering an algorithmic taxonomy and a set of open design problems for autonomous UAV platforms — giving researchers in the field a concrete agenda for which gaps still need filling.