Drivers who have waited multiple cycles to turn traffic lights green know how troublesome an annoying signaled intersection can be. But sitting at an intersection doesn’t just cause drivers perseverance. Unproductive vehicle idling can contribute 15% of the carbon footprint from US land transport.
A large-scale modeling study led by researchers at MIT reveals that eco-driving measures that allow vehicle speeds to be dynamically adjusted to reduce shutdowns and excessive acceleration can significantly reduce their CO.2 Exhaust.
Using a powerful artificial intelligence method known as deep reinforcement learning, researchers conducted a detailed impact assessment of the factors affecting vehicle emissions in three major US cities.
Their analysis shows that fully adopting environmentally driven measures can slow traffic throughput, affect vehicle and traffic safety, and reduce cross-carbon emissions across the city by 11-22%.
Researchers found that even if only 10% of vehicles on the road use eco-drive, 25-50% of the total CO2 emission reduction occurs.
Additionally, dynamically optimizing the speed limit at about 20% of the intersection will result in 70% of the total emissions benefit. This indicates that environmental driving measures can be implemented over time, while having a measurable and positive impact on climate change mitigation and improving public health.
“Vehicle-based control strategies such as eco-driving can move needles to reduce climate change. Here we have shown that modern machine learning tools like deep reinforcement learning can accelerate the kinds of analytics that support socio-technical decision-making. This is the tip of Iceberg” (IDSS) member of the MIT and Information and Decision Systems (LID) lab.
She is featured in the paper by lead author Vindula Jayawardana, a graduate student at MIT. MIT graduate students AO QU, Cameron Hickert, and Edgar Sanchez. Katherine Tan of the Faculty of MIT. Baptiste Freydt, a graduate student in Eth Zurich. Mark Taylor and Blaine Leonard of the Utah Department of Transportation. Research will be displayed on Transportation Research Part C: Emerging Technology.
Multipart modeling research
Traffic control measurements usually recall fixed infrastructure, such as stop signs and traffic signals. However, as vehicles advance more technologically, it offers eco-driving opportunities. This is all the terminology of vehicle-based traffic control measures, such as using dynamic speeds to reduce energy consumption.
Soon, eco-driving will include speed guidance in the form of a vehicle’s dashboard or smartphone app. In the long run, eco-driving includes intelligent speed commands that directly control the acceleration of semi-autonomous and fully autonomous vehicles through inter-vehicle communication systems.
“Most previous work has focused on how Eco-drive is implemented. We shifted the frame, Eco drive is implemented. Is there any difference when deploying this technology at a large scale? ” says Wu.
To answer that question, the researchers embarked on a multifaceted modeling study, which accounted for the majority of four years to complete.
They began by identifying 33 factors affecting vehicle emissions, including temperature, road slope, intersection topology, vehicle age, traffic demand, vehicle type, driver behavior, traffic light timing, and road geometry.
“One of the biggest challenges was to ensure we were hardworking and not rule out any major factors,” Wu says.
They then used data from Open Street Maps, the US Geological Survey, and other sources to create digital replicas of over 6,000 signaled intersections in three cities: Atlanta, San Francisco and Los Angeles, simulating more than one million traffic scenarios.
Researchers used deep reinforcement learning to optimize each eco-driven scenario to achieve maximum emissions benefits.
Reinforcement learning can optimize vehicle driving behavior through trial and error interactions with high-fidelity traffic simulators, and reward more energy-efficient vehicle behavior while punishing those that do not.
However, training of vehicle behavior, generalized across diverse cross-traffic scenarios, was a major challenge. Researchers observed that some scenarios are more similar to one another than others, such as scenarios with the same number of lanes and the same number of traffic signal phases.
Therefore, researchers trained individual reinforcement learning models for different clusters of traffic scenarios, improving emission benefits overall.
However, even with the help of AI, analyzing citywide traffic at the network level is so computationally concentrated that it could take another decade to solve it.
Instead, they broke the problem and resolved each eco-drive scenario at individual crossing levels.
“We carefully constrained the impact of eco-drive control at each intersection at adjacent intersections. In this way, we dramatically simplified the problem that allowed us to perform this analysis on a large scale without introducing unknown network effects,” she says.
Significant emissions benefits
When they analyzed the results, the researchers found that full adoption of eco-driving could result in cross-emission reductions of 11-22%.
These benefits vary depending on the city’s street layout. While dense cities like San Francisco have little room to implement environmental drives between intersections and provide a possible explanation for reducing emissions nodes, Atlanta can see a major advantage given the high speed limits.
Even if only 10% of vehicles employ environmental drive, cities can still achieve 25-50% of total emissions benefits due to the dynamics of the vehicle. Non-ECO-powered vehicles follow controlled, eco-powered vehicles to smoothly pass through intersections and optimize speeds to reduce carbon emissions.
In some cases, eco-driving can also increase vehicle throughput by minimizing emissions. However, Wu warns that increasing throughput could lead to more drivers coming onto the roads and reducing emission benefits.
Additionally, analysis of widely used safety indicators known as proxy safety measures such as collision times suggests that eco-driving is as safe as human driving, but can cause unexpected behavior in human drivers. More research is needed to fully understand the potential safety impacts, Wu says.
Their results also show that ecodrives could potentially bring greater benefits when combined with alternative transport decarbonization solutions. For example, 20% eco-driven adoption in San Francisco reduces emissions levels by 7%, but when combined with the adoption of hybrid and electric vehicles, it reduces emissions by 17%.
“This is the first attempt to systematically quantify the environmental benefits of the entire eco-driving network. This is an incredible research effort that serves as an important reference that others will build in eco-driving systems.”
Researchers also focus on carbon emissions, but the benefits are highly correlated with improved fuel consumption, energy use and air quality.
“It’s almost a free intervention. We’re already using smartphones in cars and cars with more advanced automation capabilities at a rapid pace. In order to actually scale quickly, it has to be relatively easy to implement and shovel-ready.
This work is partially funded by Amazon and the Utah Department of Transportation.
