Multi-Depot Capacitated VRP

This competition focuses on solving a Collaborative Vehicle Routing Problem, formally known as the Multi-Depot Capacitated Vehicle Routing Problem (MD-CVRP). In the logistics industry, low truck space utilisation is a persistent challenge - nearly one-third of trucks on the road travel empty. Enhancing collaboration between carriers offers a powerful solution to reduce inefficiencies and environmental impact.

In this challenge, participants will aim to minimize the total travel distance across a fleet of trucks. Each truck departs from its assigned depot, serves a set of customer demands, and then returns to their own depot. All trucks have a uniform capacity limit, and optimal routing decisions must respect this constraint while achieving the most efficient use of vehicle space and distance. This problem mirrors real-world logistics collaboration and invites innovative approaches to improving resource utilisation in multi-depot networks.

Dataset

The dataset for this competition comprises MDCVRP 100 instances. Each instance is randomly generated within a 1000 × 1000 Euclidean space, using configurations based on carefully selected parameters to ensure realistic complexity and diversity.

Sample Dataset

👉 Click the link to access the sample dataset, representing approximately 10% MD-CVRP instances of the final dataset.

Final Dataset for Competition

Important: Access to the dataset requires first to register this challenge.

👉 Click the link to access the FINAL dataset for competition.

Instance Structure

Each instance contains the following key properties:

• name: Unique identifier for the instance
• depots: Number of depots included
• dimensions: Total number of nodes (depots + customers), indicative of the problem’s computational complexity
• edge_weight_type: Type of distance metric; this challenge uses 2D Euclidean distance
• capacity: Maximum vehicle capacity allowed in an instance

Each instance includes a node_coord_section, detailing the locations of all nodes (depots and customers). Depots are listed first, followed by customers. Each node includes:

• id: Unique node identifier
• x: X-coordinate
• y: Y-coordinate

There are also two additional sections:

• demand_section: Specifies demand for each node; depot nodes have a demand of zero
• depot_section: Lists depot nodes (same as the first part of node_coord_section)

These structured datasets are designed to support scalable testing and benchmarking of optimisation algorithms in realistic multi-depot logistics scenarios.

Instance Files

For ease of use, each instance is provided in two formats: Plaintext (.vrp) and Structured YAML (.yaml). Both formats contain the same information, and participants are free to use whichever best suits their implementation.

To aid in visual understanding, each instance is also accompanied by a .png visualization, where Depots are shown as squares and Customers are shown as solid circles. These visualizations offer a quick, intuitive view of the spatial distribution of nodes within each instance.

Vehicle Configuration

All vehicles in each instance are homogeneous, sharing the same capacity as defined by the capacity property in the instance file. Each vehicle is:

• Assigned to a single depot
• Required to depart from and return to that same depot after completing its route

There is no limit on the number of vehicles that may be used in a solution. However, this decision directly affects the objective function, as the goal is to minimize the total travel distance across all vehicles.

Evaluation and Scoring

Participants are encouraged to develop their solvers using any programming language and open-source libraries of their choice. Each solver should aim to generate solutions for all 100 MD-CVRP instances provided.

Evaluation Criteria

Submitted solutions will be evaluated based on the following metrics:

• Coverage: Total number of instances successfully solved
• Solution quality: Total travel distance — shorter is better
• Computational efficiency: Time taken to compute each solution
• Vehicle space utilisation: Higher usage of vehicle capacity is preferred — laeger is better

Required Output for Each Instance

A complete solution for each instance should include:

• Routes of all vehicles, along with their individual load and travel distance
• Total distance and total load across all routes
• Vehicle capacity
• Machine specifications used to compute the solution (CPU and memory)
• Computation time

Important: All solution files must be:

• Saved as plaintext
• Well-formatted and clearly commented
• Named using the format of sol__[instance_id].txt or [instance_id].sol

🔔 An example solution (not necessarily optimal), along with its corresponding instance, is provided for reference. Click the link to download.

Distance Calculation

The Euclidean distance between two nodes (x1, y1) and (x2, y2) will be calculated as a floating-point number.
For simplicity and consistency in scoring, only the integer part of this value will be used. Please round down to the nearest integer.

Important: Please ensure to follow the guidelines outlined above to guarantee that your submission is eligible for evaluation and can be fairly assessed under consistent and uniform conditions.

Submission

Please compile all your solution files into a single .zip archive and submit it via email to scl.data.challenge@gmail.com. Make sure to include the word "Logistics" in the subject line. The submission deadline is 5:00 PM (GMT+2) on 2 July 2025.

Awards and Certificates

Two awards will be presented in this competition:

• 🏆 Best Performance Award: Awarded to the team or individual whose model achieves the best overall solutions based on the evaluation criteria.
• 🏆 Award for Innovation: Recognises a model that demonstrates exceptional creativity and insight in its approach.
• 🎓 Certificate of Participation: Participants can choose to opt in during registration to receive a Certificate of Participation.

Join us in solving a real-world challenge faced by logistics professionals and showcase your expertise in optimisation and operational research.