Impact of navigation apps on congestion and spread dynamics on a transportation network

Alben Rome Bagabaldo, Qianxin Gan, Alexandre M. Bayen, Marta C. GonzálezData 2024 · DOI

TL;DR

While navigation apps promise to save us time, this study reveals a "sweet spot" for adoption: traffic systems perform best when only 30–60% of drivers use dynamic routing. By applying epidemiological SIR modeling to traffic flow, researchers found that 40% adoption minimizes the "infectiousness" of congestion, preventing traffic jams from spreading across the network.

What Problem Does This Paper Address?

We’ve all been there: Waze or Google Maps suggests a "shortcut" through a quiet residential neighborhood to bypass a highway jam. While this helps the individual driver, it often creates new bottlenecks on streets never designed for high volume.

This paper tackles the tension between User Equilibrium (UE)—where individuals selfishly minimize their own travel time—and System Optimum (SO)—where the total network cost is minimized. The researchers sought to quantify the "Price of Anarchy": the systemic inefficiency caused by selfish routing. Specifically, they wanted to know: at what level of app adoption does the benefit of real-time routing get canceled out by the chaos of everyone trying to take the same "secret" shortcut?

Methods

The authors employed a high-fidelity microscopic traffic simulation of the San Jose Mission District in Fremont, California, using the Aimsun Next platform.

The core of their approach involved two distinct routing behaviors:

  1. Static Routing (Fixed): Modeled using Static User Equilibrium (SUE), representing drivers who stick to known, historically fast paths.
  2. Dynamic Routing (App-guided): Modeled using Stochastic Route Choice (SRC) with a C-Logit model. This allows drivers to adjust their routes en-route based on real-time costs (travel time) and "commonality factors" (to ensure the model doesn't just pick two nearly identical paths).

The most novel aspect is the application of the Susceptible–Infected–Recovered (SIR) model, typically used in epidemiology.

  • Susceptible (S): Links with free-flowing traffic.
  • Infected (I): Congested links (defined as speed < 90% of the limit).
  • Recovered (R): Links that were congested but have returned to free flow.

By fitting simulation data to this model, they calculated the Basic Reproductive Number ($R_0$) for traffic—a metric representing the ratio of congestion propagation to dissipation.

Datasets

The study utilized a robust set of real-world data for the city of Fremont, CA:

  • Network Geometry: Sourced from OpenStreetMap (OSM) and refined with ArcGIS and Google StreetView to ensure accurate lane counts and signal locations.
  • Traffic Demand: Derived from the SF-CHAMP regional model and a StreetLight data study, totaling approximately 130,000 vehicles over a 6-hour afternoon peak.
  • Validation Data: Calibrated against 56 city flow detectors, 27 CalTrans PeMS (Performance Measurement System) detectors, and real-time speed data from the MapBox API.

Key Insights and Ideas

  • Congestion as a Contagion: The paper successfully demonstrates that traffic jams behave remarkably like viral outbreaks. Using $R_0$ allows planners to see not just where traffic is, but how "contagious" a specific bottleneck is to the rest of the network.
  • The Goldilocks Zone (30–60%): System performance is non-linear. At low adoption, the network is under-informed. At very high adoption (>70%), the "selfish" nature of the algorithms leads to a new form of Braess’s Paradox, where adding information actually increases total travel time because everyone flocks to the same alternative routes simultaneously.
  • The 40% Sweet Spot: The SIR model identified 40% dynamic routing as the point where $R_0$ is at its lowest. At this level, the "infection" of congestion is most likely to dissipate rather than spread.

Key Results

  • Optimal Performance: Traffic metrics (flow, density, delay, and queue length) were consistently best in the 30–60% adoption range.
  • Diminishing Returns: When adoption hit 90–100%, performance degraded significantly, sometimes becoming worse than having no app users at all (0% dynamic routing). For example, the mean delay time at 100% adoption was nearly double that of the 40% adoption scenario.
  • Queue Lengths: The most dramatic improvements were seen in queue lengths and delay times, which were minimized when roughly half the population followed dynamic en-route guidance.

Limitations and Future Work

The authors acknowledge several constraints:

  • Geographic Specificity: The results are based on the specific topology of Fremont, CA. Different network structures (e.g., a strict grid vs. a hub-and-spoke) might yield different "sweet spots."
  • Compliance Assumptions: The model assumes drivers follow app advice 100% of the time if they are "dynamic users." In reality, driver trust and compliance vary.
  • Environmental Impact: While the paper focuses on flow, the authors note that diverting traffic to residential areas increases noise and air pollution in those zones—a "social equity" issue that requires further study.
  • Autonomous Vehicles: Future work should explore how these dynamics change when vehicles are controlled by algorithms rather than human drivers, potentially allowing for "cooperative" rather than "selfish" routing.

Summary updated

Abstract

Abstract In recent years, the widespread adoption of navigation apps by motorists has raised questions about their impact on local traffic patterns. Users increasingly rely on these apps to find better, real-time routes to minimize travel time. This study uses microscopic traffic simulations to examine the connection between navigation app use and traffic congestion. The research incorporates both static and dynamic routing to model user behavior. Dynamic routing represents motorists who actively adjust their routes based on app guidance during trips, while static routing models users who stick to known fastest paths. Key traffic metrics, including flow, density, speed, travel time, delay time, and queue lengths, are assessed to evaluate the outcomes. Additionally, we explore congestion propagation at various levels of navigation app adoption. To understand congestion dynamics, we apply a susceptible–infected–recovered (SIR) model, commonly used in disease spread studies. Our findings reveal that traffic system performance improves when 30–60% of users follow dynamic routing. The SIR model supports these findings, highlighting the most efficient congestion propagation-to-dissipation ratio when 40% of users adopt dynamic routing, as indicated by the lowest basic reproductive number. This research provides valuable insights into the intricate relationship between navigation apps and traffic congestion, with implications for transportation planning and management.
BibTeX
@inproceedings{bagabaldo2024impact,
  author = {Alben Rome Bagabaldo and Qianxin Gan and Alexandre M. Bayen and Marta C. González},
  title = {{Impact of navigation apps on congestion and spread dynamics on a transportation network}},
  booktitle = {Data},
  year = {2024},
  doi = {10.1007/s42421-024-00099-w},
}
Figure