Urban Delivery MAS

This report presents the architectural design, formal cognitive logic, and practical implementation of UrbanDeliveryMAS, a fully decentralized Multi-Agent System (MAS) designed to address dynamic, priority-aware last-mile logistics without reliance on a centralized routing authority or shared global locks. Operating on a discrete 10 ×10 micro-city grid topology, the system coordinates autonomous BDI (Belief-Desire-Intention) agents—comprising a central Distribution Hub and three delivery drones—negotiating task allocation via an extended Contract-Net Protocol (CNP). Agent cognitive architectures are developed in AgentSpeak(L) and executed using the Jason 3.3.0 interpreter running on Java 21. The system’s adaptive capacity is evaluated across three core operational edge cases: (1) proximitydriven spatial task division, (2) belief-triggered intention dropping and re-delegation during critical battery depletion, and (3) real-time task preemption following the mid-transit injection of high-priority Express orders. Experimental validation demonstrates 100% task completion (7/7 orders delivered) with zero unhandled goals or synchronization deadlocks. The results confirm that localized BDI deliberation paired with asynchronous message passing produces robust, self-organizing, and highly adaptive global fleet behavior.