Learning-based edge-cost modeling for decision support in industrial logistics

Reliable operational cost modeling is essential for industrial logistics decision support, where transport decisions are often still based on geometric distance, nominal speed, or other simplified assumptions. In practice, however, traversal time and energy consumption depend on local geometry, repeated turning, slowdown behaviour, and execution conditions that are not captured well by static cost models. This paper presents a learning-based and data-driven framework for planner-compatible operational cost modeling from real Automated Guided Vehicle (AGV) telemetry in a manufacturing environment.The proposed framework represents the shop floor as a directed graph and converts irregularly sampled real AGV telemetry into a uniform 1 Hz representation before extracting edge-level traversal samples. A graph-aligned edge-cost estimator is then trained to predict traversal time and energy consumption for directed edges using structural and motion-related features. Because the model does not rely on route-specific identifiers, it can produce operational cost estimates across the graph under sparse real-world observations. The predicted costs are integrated with a classical graph-search planner to support explicit, interpretable, and practically deployable decision making for internal logistics tasks.The framework is evaluated on real AGV data at the edge, route, and mission levels, including hold-out analysis and comparison against static and uniform-speed baselines. The results show that the proposed approach improves operational cost modeling and provides practically useful decision-support information for route evaluation, ETA-related reasoning, and energy-aware logistics operation. A complementary temporal realization module is also included to analyse execution behaviour beyond constant-speed assumptions. Overall, the study shows how applied AI can enhance industrial logistics decision making while remaining compatible with established graph-based optimization pipelines.