Improving traffic flow thanks to artificial intelligence

The CIRCLES (Congestion Impacts Reduction via CAV-in-the-loop Lagrangian Energy Smoothing) project aims to reduce traffic flow instabilities (called "phantom traffic jams") that cause congestion on the road and waste energy.

From November 14 to 18, 2022, the world's largest open-track traffic experiment took place on a portion of I-24 in Nashville, USA, in which three researchers from Gustave Eiffel University participated.

Mitigating "phantom traffic jams" and reducing fuel consumption

This is the ambitious goal of this project, which aims to reduce traffic jams by introducing autonomous vehicles equipped with artificial intelligence into road traffic. The stakes are high, because reducing traffic jams means less fuel consumption and fewer CO2 emissions (even for electric cars).

To demonstrate that these effects can be significantly reduced with the help of semi-autonomous vehicles and specially designed algorithms, previous tests were conducted in closed circuits before being applied to real traffic.

A fleet of 100 vehicles was gradually introduced into traffic on a 6-kilometer stretch of a busy highway to modify user behavior and reduce the "accordion" effect. For their part, the 60 researchers mobilized for the experiment collect data in real time in the control center thanks to the 300 4K digital sensors installed on the stretch to monitor the traffic.

Developing AI with human-like behavior

As part of the CIRCLES consortium lead by UC Berkeley, Multiple algorithms are developed that govern how fast these AI-powered vehicles should go. Directly connected to cruise control, these algorithms use information about general traffic conditions and the vehicle's immediate environment to determine the best speed to improve traffic flow.

Mostafa AMELI, Assistant professor at GRETTIA, leads the microsimulation team in this project to build a calibrated multi-agent simulation tool for the experiment design. This team in collaboration with all groups in CIRCLES consortium creates a calibration tool to generate the background traffic and develop a multi-lane simulator to evaluate different scenarios of adding a fleet of 100 vehicles.

An experiment like this one is key to developing "socially sustainable" artificial intelligence. Indeed, these simulations allow researchers to ensure that vehicles do not behave in ways that are dangerous and could be considered unacceptable to humans. For example, vehicles can keep traffic flowing by maintaining a slow and steady speed, rather than constantly accelerating and braking. However, slow driving can open up large gaps in traffic, which could irritate other drivers or allow other cars to get in the way.

Consortium members :

University of California, Berkeley, Vanderbilt University, Rutgers University - Camden, Temple University, University of Arizona, Gustave Eiffel University, École des Ponts ParisTech, Toyota Motor Corporation, Nissan Motor Corporation, General Motors, Tennessee Department of Transportation, and U.S. Department of Energy (DOE).

Gustave Eiffel University researchers involved in this project:

Mostafa AMELI, assistant professor at GRETTIA Laboratory, Gustave Eiffel University
Negin ALISOLTANI, researcher at GRETTIA Laboratory, Gustave Eiffel University
Maryam SAMAEI, intern at GRETTIA Laboratory, Gustave Eiffel University