MilleMiglia: A realistic instance generator for middle-mile logistics

How do you model the part of shipping that sits between factory pickup and doorstep delivery, without getting lost in company secrets? Google Research’s answer is MilleMiglia, a C++ instance generator built for middle-mile logistics, the stretch where goods move between distribution centers across regions or even continents.

That middle leg is not just a bigger version of the last mile. A shipment can change vehicles several times, wait for the next departure, and miss a connection if it arrives late at a hub. That makes timing the real problem. It’s less about sending one truck to one customer and more about keeping a chain of handoffs from breaking.

The paper behind the tool says research has lagged here because the useful data is private. Network layouts and demand volumes are exactly the sort of thing logistics companies don’t hand out. MilleMiglia tries to get around that by generating synthetic benchmarks that still look like real networks, using gravity models, spatial clustering, realistic demand pairs, and structured vehicle schedules rather than random links.

The format also bakes in the annoyances that make the problem hard: fixed timetables, distribution-center throughput limits, and synchronization between incoming and outgoing vehicles. Google says existing VRP solvers are the wrong fit because middle-mile routing behaves more like a multi-commodity flow problem on a space-time graph, with nodes as distribution centers at specific times and arcs as either vehicle movements or storage.

The project is open sourced with documentation and code on GitHub, and Google says it is meant to cover everything from small academic test cases to large continental instances, plus data sets big enough for machine learning. It is also part of a collaboration with academic partners at UniBrescia and ENPC Paris, and Google says a specialized solver and API are in progress.