Google Research has released MilleMiglia, an open-source generator that produces realistic synthetic instances of middle-mile logistics problems — the movement of freight between warehouses and local depots, before the final delivery to a door. The code is published under the OR-Tools organisation on GitHub with documentation and example instances, and an accompanying paper is on the HAL repository.
The reason it exists is a data problem rather than a modelling one. Carriers treat network topology and demand volumes as commercially sensitive, so academic work on a stage of the supply chain that accounts for a large share of cost has had almost no public datasets to test against. Published results have been hard to compare because everyone benchmarks on something different.
The generator places distribution centres using gravity models or spatial clustering, produces shipments with plausible volume and weight distributions, and encodes the constraints that make these problems awkward: fixed vehicle schedules, throughput limits and synchronisation between legs. Instances scale from small academic cases to continental networks. It was built with the University of Brescia and ENPC Paris. This is infrastructure for research rather than a result, but benchmark availability tends to decide what gets studied. The write-up is on the Google Research blog.