Any driver who has ever endured several cycles for a traffic light to change knows how frustrating signalized crossings can be. However, waiting at crossings isn’t just a nuisance for drivers — idle vehicles could account for as much as 15 percent of the carbon dioxide emissions from U.S. land transport.

A comprehensive modeling research led by MIT scholars demonstrates that eco-driving initiatives, which can encompass dynamically modifying vehicle speeds to minimize stops and harsh accelerations, could substantially decrease those CO₂ emissions.

Employing a potent artificial intelligence technique known as deep reinforcement learning, the researchers performed an extensive impact analysis of the elements influencing vehicle emissions in three major U.S. metros.

Their findings suggest that fully embracing eco-driving techniques could lower annual city-wide intersection carbon emissions by 11 to 22 percent, without hindering traffic flow or compromising vehicle and traffic safety.

Even if just 10 percent of vehicles on the road utilize eco-driving, it would lead to 25 to 50 percent of the overall decline in CO2 emissions, the researchers discovered.

Furthermore, dynamically optimizing speed limits at around 20 percent of intersections yields 70 percent of the total emission advantages. This implies that eco-driving measures could be gradually implemented while still having noticeable, positive effects on combating climate change and enhancing public health.

research is published in Transportation Research Part C: Emerging Technologies.

A multi-part modeling investigation

Traffic management strategies typically bring to mind fixed structures, such as stop signs and traffic lights. However, as vehicles become increasingly advanced, there arises an opportunity for eco-driving, a term that encompasses vehicle-based traffic management tactics such as utilizing dynamic speeds to lower energy usage.

In the short term, eco-driving could include speed guidance via vehicle dashboards or mobile applications. In the long term, eco-driving could integrate intelligent speed commands that directly manage the acceleration of semi-autonomous and fully autonomous vehicles through vehicle-to-infrastructure communication systems.

“Most previous research has concentrated on how to implement eco-driving. We changed the perspective to explore whether we should implement eco-driving. If we were to deploy this technology on a large scale, would it result in a difference?” Wu explains.

To address that question, the researchers embarked on a multifaceted modeling investigation that would take nearly four years to finalize.

They began by pinpointing 33 factors affecting vehicle emissions, including temperature, road gradient, intersection design, vehicle age, traffic demand, types of vehicles, driver conduct, traffic signal timing, road layout, etc.

“One of our primary hurdles was ensuring we were thorough and didn’t overlook any significant factors,” Wu points out.

Subsequently, they utilized data from open street maps, U.S. geological surveys, and other resources to construct digital models of over 6,000 signalized intersections in three cities — Atlanta, San Francisco, and Los Angeles — and simulated over a million traffic scenarios.

The researchers applied deep reinforcement learning to optimize each scenario for eco-driving to maximize emission benefits.

Reinforcement learning enhances vehicles’ driving behavior through trial-and-error interactions with a highly detailed traffic simulator, rewarding energy-efficient behaviors while penalizing inefficient ones.

Nevertheless, training vehicle behaviors that can generalize across varying intersection traffic scenarios proved to be a significant challenge. The researchers noted that some scenarios bore closer similarities to one another than others, such as scenarios with a consistent number of lanes or traffic signal phases.

Thus, they developed separate reinforcement learning models for different clusters of traffic scenarios, yielding better emission benefits overall.

Yet, even with AI assistance, analyzing city-wide traffic at the network level would be so computationally demanding that it could take an additional decade to resolve, Wu says.

Instead, they simplified the challenge and solved each eco-driving scenario at the individual intersection level.

“We meticulously restricted the impact of eco-driving control at each intersection on neighboring intersections. This way, we significantly simplified the issue, allowing us to conduct this analysis at scale without incorporating unpredictable network effects,” she explains.

Notable emission advantages

Upon analyzing the results, the researchers discovered that complete adoption of eco-driving could lead to intersection emissions reductions of between 11 and 22 percent.

These advantages vary based on a city’s street design. A compact city like San Francisco offers limited space to implement eco-driving between intersections, which may account for the diminished emission savings, while Atlanta might benefit more due to its higher speed limits.

Even if merely 10 percent of vehicles practice eco-driving, a city could still realize 25 to 50 percent of the total emissions advantage due to car-following dynamics: Non-eco-driving vehicles would trail controlled eco-driving vehicles as they adjust speed to pass smoothly through intersections, thereby reducing their carbon emissions as well.

In certain situations, eco-driving could also enhance vehicle throughput by lowering emissions. However, Wu warns that increased throughput might lead to more drivers on the roads, which could diminish emissions benefits.

Although their examination of commonly utilized safety metrics known as surrogate safety measures, including time to collision, indicates that eco-driving is as safe as human driving, it may provoke unexpected reactions in human drivers. More investigation is necessary to understand fully the potential safety implications, Wu notes.

Their findings also suggest that eco-driving could deliver even more significant benefits when paired with other transportation decarbonization methods. For example, a 20 percent adoption of eco-driving in San Francisco could reduce emission levels by 7 percent, but when combined with projected hybrid and electric vehicle adoption, emissions could drop by 17 percent.

“This marks a preliminary attempt to systematically quantify the network-wide environmental benefits of eco-driving. This is a crucial research initiative that will serve as an important reference for others to build upon in the evaluation of eco-driving systems,” remarks Hesham Rakha, the Samuel L. Pritchard Professor of Engineering at Virginia Tech, who was not involved in this research.

Furthermore, while the researchers concentrate on carbon emissions, the advantages highly correlate with improvements in fuel efficiency, energy consumption, and air quality.

This research is supported, in part, by Amazon and the Utah Department of Transportation.