The MILA project, developed by Optit within the FAIR (Future Artificial Intelligence Research) Extended Partnership, and specifically under the University of Bologna’s “Pervasive AI” cascade funding call, set out to close the gap between optimisation models and the reality of operational data.
In logistics, data is rarely clean: master data full of errors, real speeds that differ from those on the map, planning criteria that no model records. So even the most rigorous algorithm risks optimising a scenario that doesn’t exist.
MILA’s answer was a hybrid approach. It combines the rigour of symbolic optimisation models with the ability of machine learning to read and correct real-world data. The result is new features for Optit’s platforms and a contribution to research on AI applied to logistics and the environment.
The project brought new functionality to Optit’s platforms and, at the same time, contributed to the scientific advancement of AI applied to logistics and the environment.
Results
More reliable data. Prototype models that automatically correct master data and infer the characteristics of new delivery points, even when the available information is incomplete.
Optimised loads. Algorithms that determine the optimal loading of goods into vehicles and autonomously select the most suitable configuration parameters.
Maps that reflect the road. A map-matching prototype that enriches the road network with real speeds derived from GPS data. Planning models therefore work with actual travel times, not theoretical ones.
The human factor in the models. Models that learn the preferences of those who plan. They make it possible to assess trip quality and predict visit sequences in Food Retail, building on planners’ experience rather than ignoring it.
New affinity KPIs. Indicators that measure the affinity between service and driver and bring it into optimisation models, steering planning choices more effectively.
Impact and sustainability
MILA also contributed to the cross-cutting objectives of Italy’s National Recovery and Resilience Plan (NRRP): planning on more accurate data means fewer kilometres travelled. The tools developed also lay the theoretical and operational groundwork for the transition to electric fleets, supporting climate change mitigation and the circular economy.
