Evolvable Multi-Agent Hierarchical Architecture for Contested Logistics Ecosystems
2.2 SYSTEMS-OF-SYSTEMS AND EVOLVABILITY
Maier [ 3 ] established foundational principles for architecting systems-of-systems, noting that independent components demand governance structures differing from monolithic design. Cavin and Lohse [ 4 ] introduced evolvability as a first-class design objective, providing a direct precedent for the approach taken here: designing for incremental capability change rather than episodic system replacement.
2.3 MULTI-AGENT SYSTEMS FOR LOGISTICS AND SUPPLY CHAINS
Moore [ 5 ] provides a comprehensive taxonomy of hierarchical multi-agent systems. Xu, Almahri, Mak, and Brintrup [ 6 ] demonstrate opportunities for autonomous supply chains while cautioning that governance, trust, and interoperability remain open challenges. Xu, Mak, Schoepf, Ostroumov, and Brintrup [ 7 ] present empirical results for multi-agent digital twinning. Loganathan and Chinnaraju [ 8 ] examine reinforcement learning for multi-agent logistics coordination.
2.4 AI-READY DATA AND REAL-TIME ANALYTICS
Chen, Milosevic, Rabhi, and Berry [ 9 ] survey architectures for real-time analytics, establishing that streaming data governance is central to operational decision-making. Gujjala [ 10 ] extends this, noting that AI analytics require robust quality and provenance frameworks to be trustworthy in production.
2.5 GOVERNANCE OF AUTONOMOUS AI SYSTEMS
Aguiar et al. [ 11 ] address governance and security for LLM-based multi-agent architectures, identifying trust propagation, access control, and agent authentication as critical architectural concerns. Amershi et al. [ 12 ] establish design guidelines for human-AI interaction, including graceful degradation and correction mechanisms. Biggio and Roli [ 13 ] survey a decade of adversarial machine learning, demonstrating that AI systems are vulnerable to both poisoning( training-time) and evasion( inference-time) attacks, with implications for any operational AI system whose inputs can be influenced by adversaries. Rose et al. [ 14 ] provide the foundational framework for zero-trust architecture, establishing that in distributed, multi-party environments, no actor should be implicitly trusted based on network location or identity alone, a principle directly applicable to contested logistics ecosystems. Collectively, this literature establishes that resilient, AI-enabled logistics requires not only governance and human oversight, but explicit architectural responses to adversarial manipulation.
Taken together, this literature establishes that the key barriers are architectural: interoperability, data trust, hierarchical governance, evolvable design, and adversarial resilience have not been comprehensively addressed by existing point solutions.
3 PROBLEM STATEMENT AND RESEARCH GAP
Despite sustained investment in digital transformation, many logistics systems remain illequipped to operate under disruption, uncertainty, and fragmentation. Traditional logistics architectures have evolved around assumptions of stable infrastructure, continuous data
62