Data-Driven Edge Cost Learning and Temporal Motion Modeling for Energy-Aware AGV Planning

This repository contains the code for a graph-based AGV planning framework that combines GGNN-based edge-cost learning with planner-based route generation and temporal motion realization. The implementation supports edge-level prediction of traversal time and energy, route computation over a directed factory graph, and trajectory realization using multiple temporal models, including Physics-Delta, MLP, GRU, LSTM, and Transformer-Lite. The repository is intended to support reproducible experiments for AGV route planning, cost-aware decision making, and execution-level motion modeling.