Amazon Last Mile Routing Research Challenge Dataset (2021)
Amazon
9,184 real Amazon delivery routes from 2018, released for a routing-algorithm challenge.
Route-, stop-, and package-level features for historical routes driven by Amazon delivery drivers in five U.S. metro areas, released for the 2021 Last Mile Routing Research Challenge run with MIT. Identifiers are regenerated and locations obfuscated. It is a rare real-world benchmark for learning-based traveling-salesperson methods.
Access
- Cost
- FreeNo charge to download; hosted in an AWS S3 bucket.
- License
- Open, non-commercial onlyCC BY-NC 4.0 (Amazon copyright; non-commercial use)
- Sign-up
- No sign-up
- How you get it
- API and Bulk downloadsee documentation
What's in it
- One row is
- One route, with its stops, packages, and the sequence the driver actually followed
- Covers
- Five U.S. metropolitan areas
- Years
- 2018 routes
- Updated
- One-time release
- Size
- 9,184 routes
Quality
Real operational routes are scarce in public routing research, and the challenge was formally documented in a journal paper. Location data is deliberately obfuscated.
- Non-commercial license
- Locations obfuscated, so not usable for real-world mapping
- Single year and five metros
Connects to
- New York City Taxi and Limousine Commission (TLC) Trip Record Data: Urban vehicle trips at the trip level, complementing stop-by-stop delivery routes
- OpenStreetMap: Road network that route and travel-time models can be compared against
Questions it could answer
- How much do experienced drivers deviate from the shortest-path stop order?
- Do stop sequences cluster by package time windows?