Google Research open-sources a generator for logistics benchmarks, and that is the interesting part

Google Research has published MilleMiglia, a C++ instance generator that produces realistic benchmarks for middle-mile delivery problems — the leg between first mile and last mile, where shipments move between facilities rather than to a door. The post frames it as a foundational building block for future research, the generated data is privacy-preserving, and the source and documentation are on GitHub.

What it means

The headline sounds like logistics. The story is about benchmarks, and it applies far outside delivery.

Optimisation research has a supply problem that has nothing to do with ideas: the instances are private. Real routing, scheduling and network-design data belongs to companies that will not publish it, so results get reported on whatever each group could obtain, and nobody can tell whether a new method is better or merely tested on easier cases. A generator that produces realistic instances turns that into something comparable — and realistic is the load-bearing word, because random instances have been available for decades and everybody knows they are too easy.

For an engineer outside research, there are two practical uses:

• Load-testing a planner with problems that look like yours. If you build routing or scheduling software, the hardest part of testing is getting inputs that have the shape of production without production's data. That is exactly what this generates.
• Reading vendor claims. When an optimisation product quotes an improvement, ask which instances. Public generators make that question answerable, which is why vendors have historically preferred their own datasets.

The general lesson is cheap to state and hard to act on: in any field where progress is measured by comparison, the benchmark decides what gets worked on. Publishing one is a bigger intervention than publishing another algorithm, and it gets a fraction of the attention.