[Paper Review] Instance generation tool for on-demand transportation problems

Michell Queiroz, Flavien Lucas+1 more|arXiv (Cornell University)|Aug 23, 2022

Transportation and Mobility Innovations4 citations

TL;DR

This paper introduces REQreate, a flexible instance generation tool for on-demand transportation problems that uses real-world street networks from OpenStreetMap to create realistic, diverse benchmark instances. It enables configurable control over key problem properties—size, dynamism, urgency, and geographic dispersion—demonstrating its ability to generate 5,280 synthetic instances for the On-Demand Bus Routing Problem (ODBRP) that closely match real rideshare data patterns.

ABSTRACT

We present REQreate, a tool to generate instances for on-demand transportation problems. Such problems consist of optimizing the routes of vehicles according to passengers' demand for transportation under space and time restrictions (requests). REQreate is flexible and can be configured to generate instances for a large number of problems in this problem class. In this paper, we demonstrate this by generating instances for the Dial-a-Ride Problem (DARP) and On-demand Bus Routing Problem (ODBRP). In most of the literature, researchers either test their algorithms with instances based on artificial networks or perform real-life case studies on instances derived from a specific city or region. Furthermore, locations of requests for on-demand transportation problems are mostly randomly chosen according to a uniform distribution. The aim of REQreate is to overcome these non-realistic and overfitting shortcomings. Rather than relying on either artificial or limited data, we retrieve real-world street networks from OpenStreetMaps (OSM). To the best of our knowledge, this is the first tool to make use of real-life networks to generate instances for an extensive catalogue of existing and upcoming on-demand transportation problems. Additionally, we present a simple method that can be embedded in the instance generation process to produce distinct urban mobility patterns. We perform an analysis with real life datasets reported by rideshare companies and compare them with properties of synthetic instances generated with REQreate. Another contribution of this work is the introduction of the concept of instance similarity that serves as support to create diverse benchmark sets, in addition to properties (size, dynamism, urgency and geographic dispersion) that could be used to comprehend what affects the performance of algorithms.

Motivation & Objective

• To address the lack of standardized, realistic benchmark instances for on-demand transportation problems in existing literature.
• To overcome limitations of artificial or city-specific data by using real street networks from OpenStreetMap (OSM).
• To enable generation of diverse, configurable instances that reflect realistic urban mobility patterns.
• To support algorithm evaluation by providing instance sets with controlled variations in size, dynamism, urgency, and geographic dispersion.
• To introduce instance similarity as a metric to ensure diversity and robustness in benchmark sets.

Proposed method

• REQreate retrieves real street networks from OpenStreetMap using the OSMnx library for accurate spatial modeling.
• It configures instance parameters via a JSON input file specifying attributes such as origin, destination, time windows, and resource requirements for each request.
• Geographic dispersion is controlled by setting intervals for direct travel time between origin and destination nodes.
• Dynamism is adjusted by varying the percentage of requests arriving dynamically during the planning horizon (0% to 100% in 10% steps).
• Urgency is modeled using a normal distribution with configurable mean and standard deviation for request time windows.
• A POI-based request generation method uses points of interest and distance-based probabilities to simulate realistic human mobility patterns.

Experimental results

Research questions

• RQ1 How can on-demand transportation instance generators be made more realistic by leveraging real-world street networks?
• RQ2 To what extent can synthetic instances generated by REQreate replicate statistical properties observed in real rideshare data?
• RQ3 How do controlled variations in dynamism, urgency, and geographic dispersion affect algorithm performance evaluation?
• RQ4 Can instance similarity be used as a metric to ensure diversity and avoid overfitting in benchmark instance sets?
• RQ5 What configuration of instance properties leads to more profitable or operationally feasible on-demand transit systems?

Key findings

• REQreate successfully generated 5,280 synthetic instances for the On-Demand Bus Routing Problem (ODBRP), covering multiple configurations of size, dynamism, urgency, and geographic dispersion.
• The synthetic instances generated by REQreate closely match real-world statistical properties—such as trip duration, time window distribution, and spatial clustering—observed in rideshare data from Chicago.
• The tool’s POI-based request generation method effectively reproduces distinct urban mobility patterns without requiring access to proprietary real-world data.
• Instance similarity analysis confirmed that the generated benchmark set includes diverse configurations, reducing the risk of overfitting in algorithm evaluation.
• The tool enables systematic exploration of how instance properties like urgency (mean 5–30 minutes) and geographic dispersion (10–6000 seconds) influence algorithm performance.
• REQreate supports practical planning by identifying spatial, temporal, and distance distribution conditions under which on-demand transit systems are operationally viable.