Yard-Vehicle Routing Problem
The Yard-Vehicle Routing Problem (YVRP) determines how internal transport vehicles (AGVs, ALVs, straddle carriers, or terminal tractors) move containers between quay cranes, yard blocks, rail terminals, and gates. Routes must be conflict-free and respect safety, precedence, and charging constraints. The main goals are to minimize container transfer time, reduce empty travel, and balance fleet utilization under dynamic conditions.
Optimization Focus
Objective
- Minimize container transport time between quay, yard, and gate/rail points.
- Reduce empty travel and avoid route conflicts or congestion.
- Balance vehicle workload and manage AGV energy/charging cycles efficiently.
Decision Variables
- Assignment of transport jobs (quay↔yard, yard↔gate/rail) to specific vehicles.
- Conflict-free paths and start/finish times per task and route segment.
- Battery-aware dispatching and recharging schedules for automated vehicles.
Constraints
- Conflict-free routing: no collisions, safe distance, and limited lane capacity.
- Task release times, crane precedence, and time windows for synchronization.
- Vehicle kinematic limits, battery constraints, and lane direction rules.
Data Sources
- TOS logs: QC moves, yard inputs/outputs, truck/rail arrivals, container IDs.
- Network graph: arcs, intersections, buffer zones, and one-way lanes.
- Telemetry: vehicle positions, speeds, energy levels, and maintenance status.
Main Assumptions
- Tasks are known over a rolling horizon; re-optimization occurs as new events arrive.
- Lane capacities, intersections, and turning rules ensure conflict-free traffic.
- Service times are estimated or data-driven; travel times may depend on congestion.
- Integration with crane and gate scheduling improves overall terminal coordination.
Modeling Approaches
YVRP has been modeled using Mixed-Integer Linear Programming (MILP), Constraint Programming (CP), simulation-optimization, and deep reinforcement learning for adaptive routing.
- Exact models: MILP or CP on time-expanded networks for job and path allocation.
- Heuristics: Hybrid assignment-routing models for real-time and large-scale operations.
- Learning-based: DRL/DQN agents for dynamic AGV control and congestion management.
- Integrated: Joint optimization with quay and yard cranes to improve throughput and reduce idle time.
Reference Studies
🧭 Real-World Datasets
| Dataset | Description |
|---|---|
| Container-Terminal Event Logs | Rich event logs (months) with timestamps and yard fields (e.g., block/slot). Ideal to extract quay-yard pickup/delivery jobs, dwell times, and service windows for YVRP. |
| Vessel Visit API – Portbase (Rotterdam) | TProduction API for vessel-level port-call data; use ETAs/ETDs and terminal planning to build time-phased yard-move demand aligned to berth plans (Registration required). |
| ETA Terminal Open-Data API – Port of Antwerp-Bruges (NxtPort) | Open-data API exposing vessel ETA and cargo opening times per terminal. Handy to derive realistic release windows for yard jobs (Free key via NxtPort procedure.) |
| MyTerminal / Terminal APIs – ECT (Rotterdam) | Commercial APIs (e.g., Quay Planning, Object Status) providing real-time vessel status, planning, and object events. Useful to reconstruct intra-terminal transport demand. |
| Premium Terminal Data via Portbase – RWG/APMT/ECT | Feed of terminal ETAs/ETDs and discharge info via Portbase Cargo Controller. Ssupports high-fidelity demand timelines for yard moves. |
⚙️ Synthetic Benchmark Instances
| Dataset | Description |
|---|---|
| Multi-YC Instances | Ready-made instance files (multi-crane, multi-block; ranging between 6–12 tasks) suitable to synthesize YVRP job lists between quay bays and yard blocks. |
| Bulk-Port Routing Models | Routing datasets and published solutions for intra-port transport. Map directly to yard-tractor networks with node/OD definitions. |
| Dynamic PDPTW | Dynamic pickup-and-delivery instances for live re-planning. Ideal for testing online dispatch heuristics in dynamic yard environments. |
| DVRP Kilby Set | Classic dynamic VRP instances with request arrival times. Useful for evaluating event-driven yard-vehicle dispatching.. |
While few public datasets exist due to terminal confidentiality, academic benchmarks reflect realistic layouts and traffic dynamics.
