Distributed Artificial Intelligence and Edge Computing Architectures in Resource-Constrained Environments: A PRISMA-Based Systematic Literature Review on Autonomous UAV Systems

We report a systematic literature review examining distributed artificial intelligence algorithms and edge computing architectures for autonomous UAV platforms operating under resource and communication constraints, conducted within the PRISMA protocol framework. Existing reviews on UAV systems focus predominantly on physical-layer security and jamming; none provides a unified architectural analysis integrating the CAP Theorem, computational complexity, and hardware-software co-design perspectives. This review addresses that gap by offering a structured design reference for distributed systems researchers working on autonomous UAV platforms. Of 50 candidate studies retrieved from the Web of Science database, 43 were included in full-text analysis following scope and quality assessment. The studies were classified along three axes: (1) distributed AI/ML paradigm (FL, DRL, MARL, CNN, Transformer), (2) system architecture decision (MEC, hierarchical FL, Gossip-based learning), and (3) optimized metric (latency, energy consumption, model size, detection accuracy). FedAvg-based FL and DRL emerge as the leading approaches in distributed swarm learning and online resource allocation, respectively. CAP Theorem analysis shows that the majority of the examined architectures prioritize availability and partition tolerance over consistency, a design decision consistent with adversarial operating conditions. Byzantine-fault tolerant FL and model compression remain under addressed in the reviewed literature. The study concludes with an algorithmic taxonomy and a set of open design problems targeting autonomous UAV platforms in adversarial settings.