[Paper Review] Critical density of urban traffic

Adilton Jose da Silva, Borko Stošić|arXiv (Cornell University)|Sep 11, 2010

Traffic control and management3 citations

TL;DR

This study implements a modified Intelligent Driver Model on a real urban street map in Recife, Brazil, incorporating traffic lights, multiple lanes, and individual vehicle routes to simulate urban traffic. It demonstrates that critical density—evidenced by a peak in vehicle flux versus density—emerges locally on individual avenues due to dynamic interactions, but is damped when aggregated across the entire district due to phase offsets between congested and free-flowing regions.

ABSTRACT

A modified version of the Intelligent Driver Model was used to simulate traffic in the district of Afogados, in the city of Recife, Brazil, with the objective to verify whether the complexity of the underlying street grid, with multiple lane streets, crossings, and semaphores, is capable of exhibiting the effect of critical density: appearance of a maximum in the vehicle flux versus density curve. Numerical simulations demonstrate that this effect indeed is observed on individual avenues, while the phase offset among the avenues results in damping of this effect for the region as a whole.

Motivation & Objective

• To investigate whether complex urban street networks with traffic lights, multiple lanes, and individual vehicle routing can reproduce the empirically observed critical density effect in traffic flow.
• To determine if the Intelligent Driver Model (IDM), when modified with realistic vehicle and route parameters, can generate a maximum in the flux versus density curve.
• To examine how local traffic dynamics interact with regional dynamics, particularly whether the critical density effect is observable at the district level or only locally.
• To develop a GUI-based simulation tool enabling interactive model tuning and visualization of complex traffic interactions for improved credibility and experimental control.

Proposed method

• A modified version of the Intelligent Driver Model (IDM) is used, with a truncated Gaussian distribution for individual vehicle desired speeds to reflect real-world variability.
• The simulation is built on the actual street map of the Afogados district in Recife, Brazil, including real traffic rules such as traffic lights, lane merging, and multi-lane configurations.
• Each vehicle is assigned a predefined route, and traffic is introduced at varying inter-arrival times (from 0.1s to 10s) to control vehicle density across 80 distinct influx rates.
• Simulations run for 1800 seconds (30 minutes) per density level to ensure equilibrium, with 100 repetitions to compute average flux, speed, and density values.
• Data is collected per avenue segment (bounded by intersections) and aggregated over the entire district, with binning (2% density intervals) and third-order polynomial regression applied to smooth fluctuations and identify critical density.
• A graphical user interface (GUI) is implemented to visualize traffic dynamics, aid in debugging, and support interactive parameter adjustment during simulations.

Experimental results

Research questions

• RQ1 Can a modified IDM simulation on a real urban street network reproduce the critical density effect, defined by a peak in vehicle flux versus density?
• RQ2 How do traffic lights, multiple lanes, and individual vehicle routing influence the emergence of critical density in urban traffic simulations?
• RQ3 Why is the critical density effect observable at the level of individual avenues but not at the district-wide level?
• RQ4 To what extent do phase offsets between neighboring regions—where one is congested while another is free-flowing—suppress the macroscopic manifestation of critical density?
• RQ5 Is the critical density phenomenon better understood as a local, rather than global, traffic regime?

Key findings

• A clear maximum in the flux versus density curve is observed on individual avenues, with a critical density of 41.63% and a peak flux of 0.91998 cars per second (3312 cars per hour).
• The flux versus density curve for the entire district shows no maximum, indicating that the critical density effect is not observable at the macroscopic scale.
• Average vehicle speed exhibits large fluctuations at low densities, which diminish as density increases, particularly in individual avenues.
• The entire district never reaches a uniform congestion state; instead, regions alternate between free flow and congestion, leading to phase offsets that cancel out local congestion effects.
• The critical density phenomenon is identified as a local effect, arising from the dynamic interplay between a region’s outflow and neighboring regions’ inflow, rather than a global property of the system.
• This study represents the first reported numerical model capable of reproducing the critical density effect in urban traffic simulations using a realistic, data-driven setup.