Online dynamic scheduling for multi-compartment AGV and cobot in manufacturing systems via remaining processing time integration

Abstract This paper presents an online scheduling framework for a multi-compartment Automated Guided Vehicle (AGV) with optional onboard cobot, embedded in a Digital Twin environment that provides real-time machine-state data. The scheduler is a Genetic Algorithm (GA) with mission-index encoding and feasibility-by-construction initialization, generating valid sequences under capacity, traceability, and precedence constraints. To isolate the value of machine-state awareness, the GA operates in two modes: an RPT-blind baseline using logistics data only, and an RPT-aware extension that integrates machine Remaining Processing Time (RPT) into feasibility and fitness evaluation. Validation across eleven scenarios on a precision aerospace workshop reveals that the impact of RPT-awareness is regime-dependent. In non-saturated and capacity-constrained settings, readiness-driven synchronization reduces transport waste (Trip Time −34–51%, Empty Trip Time −54–70% vs. RPT-blind) but increases makespan by 5–11%, because conservative waiting can delay feeding of bottleneck resources. In coordination-sensitive and high-mix regimes, RPT-awareness improves both dimensions simultaneously, reducing makespan by up to 10% (810 $$\rightarrow$$ 728 min) alongside comparable logistics gains. Relative to the FIFO baseline representing current industrial practice, GA-based automation reduces Trip Time by 15–60% and Empty Trip Time by 15–77%. RPT-awareness also enables proactive cobot scheduling, completing all eligible inline missions (up to 29 per shift versus 4–9 in RPT-blind mode). The results indicate that RPT-aware scheduling is most effective where coordination complexity, rather than processing capacity, bounds system performance.