This repository contains problem files used to test models for solving the Meal Delivery Routing Problem and a solution evaluator. The sample instances have been provided by Grubhub and Georgia Tech's H. Milton Stewart School of Industrial & Systems Engineering to advance research on the subject.
The public_instances folder contains 240 instances using anonymized data from meal delivery operations at Grubhub. These were derived from a set of 10 seed instances through the following procedure.
- Reducing their size. This defines three subsets of instances:
o100: no size reductiono50: 50% reduction in number of orders and courier hours, sampling the order set directly.r50: 50% reduction in number of orders and courier hours, sampling the restaurant set and keeping all orders from selected restaurants.
- Changing the courier schedule. This defines two subsets:
s1: schedule based on historical courier shifts.s2: schedule with "optimized shifts" (subject to having about the same, and never more, courier hours than the historical counterpart)
- Increasing the travel speed. This defines two subsets:
t100: original travel timest75: travel times are multiplied by 0.75
- Increasing the preparation times of all orders. This defines two subsets:
p100: original preparation timesp125: preparation times are multiplied by 1.25
Each instance is then labeled by concatenating the labels of the sets to which it belongs. For example: 1o50s2t100p125 is the instance derived from seed 1, by reducing its size by 50% on the order set, using an optimized courier schedule, original travel speed, and longer preparation times; and seed 1 is represented as 1o100s1t100p100.
The folder MDRP_code contains the solution evaluator script. Meal Delivery Routing: The Grubhub Instances provides a complete description of the Meal Delivery Routing Problem and the test instance set.