Multi-Commodity Inventory Routing Problem with Pickup and Delivery

Lukas Seeholzer, Johan W. Joubert, Kris Braekers, Pieter VansteenwegenLirias 2026

TL;DR

This paper introduces a novel variant of the Inventory Routing Problem (IRP) tailored for the hospitality industry, incorporating multiple commodities, multiple suppliers, and integrated pickup and delivery. By centralizing logistics and allowing vehicles to "reload" at different suppliers mid-route, the researchers demonstrate significant cost savings over traditional, uncooperative delivery models.

What Problem Does This Paper Address?

In the world of logistics, the Inventory Routing Problem (IRP) is a classic challenge that combines two difficult tasks: deciding when to deliver goods (inventory management) and determining the best path to get them there (vehicle routing).

Most existing research simplifies this by looking at a single supplier delivering a single product. However, the hospitality sector doesn't work that way. A large hotel or restaurant chain deals with dozens of independent suppliers providing everything from fresh produce to linens. Currently, these suppliers usually operate their own independent fleets, leading to massive inefficiencies—multiple half-empty trucks might visit the same client on the same day.

The authors tackle this "uncooperative" mess by modeling a scenario where a central hospitality provider takes over the logistics. The challenge is that this adds layers of complexity: you aren't just routing; you are managing multiple product types and coordinating pickups from various suppliers to be delivered to various production sites within specific time windows.

Methods

The researchers approached the problem in two stages:

  1. Mathematical Modeling: They first developed a formal mathematical model to define the problem. While precise, they found that as the number of clients and suppliers grew, the problem became "practically infeasible" to solve optimally using standard solvers. The search space is simply too vast and the feasible solutions are too "isolated" (meaning it's hard for an algorithm to jump from one valid solution to another).
  2. Metaheuristic (Large Neighbourhood Search): To find high-quality solutions in a reasonable timeframe, they implemented a Large Neighbourhood Search (LNS). This algorithm works by "destroying" parts of a current solution (removing certain routes or deliveries) and then "repairing" them using a Variable Neighbourhood Descent (VND). This allows the solver to explore different configurations of delivery schedules and routes more flexibly than a rigid mathematical solver.

Datasets

The paper bases its analysis on a real-world case study of a large hospitality service provider.

  • Scope: The study considers multiple production sites (clients) and various independent suppliers.
  • Benchmarks: While the paper mentions testing on "large instances," it does not explicitly name a standard public dataset (like the Solomon or Christofides benchmarks). Instead, the evaluation focuses on comparing the new model's performance against the current "uncooperative" baseline and the results of the exact mathematical model on smaller scales.

Key Insights and Ideas

  • The "Mid-Route Reload": One of the most interesting features of this model is that it allows for pickup and delivery in the same trip. A truck can deliver goods from Supplier A to a client, then—instead of returning to base—head to Supplier B to pick up more goods for a second client. This effectively allows a vehicle to deliver a total volume greater than its physical capacity over the course of a single route.
  • Information Centralization: The "aha moment" here is that the efficiency isn't just about better math; it's about visibility. By having a central company manage the data for both suppliers and customers, they can unlock optimization opportunities that are impossible when suppliers work in silos.
  • Feasibility is the Real Enemy: The authors note that the mutual interdependence of decisions (when to pick up, where to store, which route to take) makes the search space extremely "brittle." Finding any feasible solution is often harder than optimizing an existing one.

Key Results

  • Efficiency Gains: The proposed model showed a "significant cost saving potential" compared to the status quo. By sharing vehicles across different suppliers and product types, the total distance traveled and the number of vehicles required were reduced.
  • Performance: The LNS metaheuristic achieved "near-optimal" solutions for many instances. More importantly, it provided feasible, high-quality results for large-scale scenarios where the exact mathematical model failed to produce a solution.
  • Inventory vs. Routing: The model successfully balanced the trade-off between storage costs (holding more inventory at the client site) and routing costs (making more frequent, smaller deliveries).

Limitations and Future Work

The paper is a condensed conference abstract, so some granular details are omitted:

  • Computational Scaling: While the metaheuristic is faster than the math model, the authors acknowledge that the high degree of freedom still makes the problem computationally taxing.
  • Real-world Constraints: Future work could involve adding more "messy" real-world variables, such as fluctuating traffic patterns, driver break requirements, or stricter refrigeration constraints for specific commodities.
  • Cooperation Dynamics: The model assumes a central authority takes over. In reality, convincing independent suppliers to hand over their logistics to a third party involves complex business and legal negotiations not covered by the math.

Summary updated

Abstract

The 40th conference of the Belgian Operational Research Society
BibTeX
@inproceedings{seeholzer2026multi,
  author = {Lukas Seeholzer and Johan W. Joubert and Kris Braekers and Pieter Vansteenwegen},
  title = {{Multi-Commodity Inventory Routing Problem with Pickup and Delivery}},
  booktitle = {Lirias},
  year = {2026},
}
Figure