Urban mobility, defined as the transport of goods and people, has profoundly transformed our economy and lifestyles. Today, it must also meet sustainability requirements. The MAMUT project (Machine learning And Matheuristics algorithms for Urban Transportation) aims to tackle the challenges of urban logistics by leveraging cutting-edge research in Operations Research (OR) and Artificial Intelligence (AI). Our goal is to provide innovative solutions for more sustainable and efficient transport of goods in urban environments.

• According to the Shift Project report (March 2022):
  • On average, 27 tons of goods are transported annually over 200 km per French person.
  • Urban freight accounts for 9% of greenhouse gas emissions, with 89% from road transport and 9% from rail.

• The French Economy Transformation Plan (PTEF) proposes:
  • Organizing urban pooling centers to optimize freight loadings.
  • Increasing reliance on electric vehicles and cyclologistics for deliveries.
  • Training drivers in eco-driving practices.

Urban logistics involves dynamic and complex challenges, such as:

• Managing deliveries with heterogeneous vehicle fleets.
• Navigating restricted traffic zones in city centers.
• Addressing the inefficiency of the last mile of freight logistics.
• Responding to increased delivery demands due to crises, such as the COVID-19 pandemic.

The MAMUT project seeks to address these challenges by combining the strengths of OR and AI, creating adaptive and efficient solutions for urban freight transport.

The MAMUT project is structured around five main objectives:

• Identify and analyze various instances of urban logistics problems.
• Characterize these problems using specific indicators and explain their behavior through machine learning.

• Analyze patterns in generated instances and solutions.
• Extract rules and insights using data mining techniques.

• Develop a generic and scalable solver capable of handling real-world constraints.
• Combine metaheuristics and time-dependent optimization techniques.

• Enhance the generic solver with learned rules and insights.
• Integrate explainable AI techniques to guide optimization processes.

• Build an open-source platform featuring:
  • Resolved logistics problems.
  • Large, realistic datasets for research and industry use.
  • Algorithms, including the hybrid OR/AI solver.

The platform will be accessible to the scientific and industrial community, fostering collaboration and innovation in urban logistics.

The MAMUT project brings together academic researchers and industrial partners:

• Professor Marc Sevaux: Project coordinator, expert in optimization and graph theory.
• Associate Professor Alexandru Olteanu: Specialist in optimization and multi-criteria decision support.
• Adrien Pichon: Doctoral student in collaboration with CITI (INSA Lyon).

• Professor Romain Billot: Expert in intelligent mobility and data science.
• Lina Fahed: Lecturer specializing in explainable machine learning and temporal sequence processing.
• Florian Rascoussier: Doctoral student in collaboration with CITI (INSA Lyon).

• Professor Christine Solnon: Specialist in combinatorial optimization and urban logistics.
• Professor Olivier Simonin: Expert in multi-agent systems.
• Romain Fontaine: Post-doc specialist in combinatorial optimization and urban logistics.

• A French company specializing in software solutions for urban logistics.
• Fabien Girard: Scientific director, expert in real-world route optimization.
• MAPO by Woop will:
  • Provide expertise in agile and open-source development.
  • Generate data and design the platform.
  • Facilitate collaboration with their existing client network.

1. Characterization and classification of urban logistics problems.
2. A hybrid OR/AI solver for large-scale urban logistics problems.
3. An open-source platform hosting:
   • Algorithms and solutions.
   • Realistic datasets for academic and industrial use.
4. Documentation and tutorials for platform users and contributors.
5. Publications and workshops to disseminate results to the scientific and industrial communities.

• Generate realistic urban logistics problems based on real-world maps and constraints.
• Develop tools to analyze the complexity of generated instances.

• Detect patterns and extract rules from solutions and instances.
• Utilize data mining techniques to characterize and predict solutions.

• Build an efficient solver capable of handling dynamic and complex constraints.

• Integrate learning-based enhancements into the solver.
• Focus on explainability and adaptability to dynamic conditions.

• Develop a collaborative platform for visualization, simulation, and sharing of urban logistics solutions.

For more information, visit the repositories in this organization or contact the project coordinator:

• Marc Sevaux - Lab-STICC, Université Bretagne Sud

This project is funded by the Agence Nationale de la Recherche (ANR) and supported by the collaboration of academic and industrial partners.