Decentralized Path Planning With Supervisory Coordination for Multi-AGV Systems in Non-Standardized Automated Warehouses

Abstract

A hybrid architecture with decentralized path planning and supervisory coordination is proposed for multi-Automated Guided Vehicle (AGV) systems operating in realistic, non-standardized (i.e., non grid-like) automated warehouses characterized by bidirectional roads and complex layouts. The method combines hierarchical environment modeling based on Finite Time-Expanded Network (FTEN) with an Event-Driven Predictive Control (EDPC) strategy, enabling each AGV to independently compute congestion-aware and conflict-free paths based on locally solved optimization problems. The FTEN leverages graph theory to model and analyze complex dynamic systems, whereas the EDPC strategy facilitates a proactive approach to avoiding congestion while reducing the computational cost of path planning. A central unit acts as a supervisory communication and information hub, periodically providing updated environmental and mission information to all vehicles. Conflict and deadlock resolution strategies, based on wait and restart and mission priority switch, support safe and efficient execution of missions. Validation tests on industrial warehouse scenarios demonstrate effective coordination, limited traffic congestion, and efficient mission execution. Note to Practitioners—The proposed work supports engineers, system integrators, and logistics providers in the design, validation, and deployment of AGV-based transportation systems within real-world warehouse environments. Traditional AGV coordination algorithms typically rely on grid-like layouts and unidirectional flows, which limit applicability in practical settings characterized by non-standardized environments with narrow two-way roads, irregular geometries, and dynamically evolving traffic conditions. The presented hybrid framework with decentralized path planning and supervisory coordination overcomes such limitations by enabling each AGV to independently compute its own path using updated environmental and traffic data. A central unit operates as a supervisory communication and information hub, distributing shared information without executing direct control over vehicle behavior. Integration into simulation environments, such as FlexSim, allows evaluation of throughput, congestion levels, and conflict resolution performance under realistic layout conditions and mission workloads. Application during the sales phase facilitates early validation of system feasibility and supports investment decisions through quantitative performance analysis.