Google's MilleMiglia generates the middle-mile logistics instances carriers keep proprietary

In first- and last-mile problems the shipment stays in one vehicle from origin to destination, and the solver is picking which vehicle takes which shipments and in what order, across a fleet, usually over a single day ^[6]. The middle mile runs as a relay. A shipment can ride several vehicles across a continental network and arrive a week after it left ^[7]. At each intermediate distribution center it may be unloaded, sorted by destination, consolidated with other freight, and loaded onto the next vehicle ^[8]. The outbound departure therefore depends on the inbound arrival, and rescheduling one leg can break a leg two hubs later ^[13].

Google's worked route moves goods from a manufacturer in Groningen to the regional distribution center in Utrecht, on to another center in Paris, and then to a consumer in Versailles ^[10]. The post opens by asking how a Dutch poffert covers 700 km and arrives the next day ^[11]. Give that trip 24 hours and the network has to average about 29 km/h end to end ^[12]. Every sort and every wait for an outbound departure comes out of the same budget ^[8].

Lotfi and Cuvelier wrote that this part of the journey "covers the longest distance, represents a huge portion of the overall costs, and most importantly dictates whether your poffert arrives fresh or stale" ^[9]. Google says the generator captures the middle mile's constraints and produces realistic, privacy-preserving data ^[14]. Synthetic output is the only form this data can take publicly, given that most logistics companies treat their network topologies and demand volumes as highly sensitive proprietary information ^[5].

Any instance generator encodes an opinion about what a typical network looks like: the number of hubs, the demand between pairs of them, the slack at transfers. A result measured on generated instances is conditional on those parameter choices ^[17]. A heuristic tuned on one hub count and one demand skew can lose to a plainer method on a real network with a different skew. So before crediting a published gain on MilleMiglia instances, read the parameters used to produce them and compare them with your own hub count and volumes. Google calls the release a foundational building block to enable future research results ^[16].