Ecological driving measures could significantly reduce vehicle emissions | MIT News

Any driver who has ever waited for traffic lights to turn green in multiple cycles knows how annoying signal intersections can be. But sitting at an intersection is not just a drag on drivers’ patience—ineffective vehicles are idle and could contribute up to 15% of U.S. land transport CO2 emissions.

A large-scale modeling study led by MIT researchers suggests that eco-driving measures may involve dynamically adjusting vehicle speeds to reduce stops and over-acceleration, which could significantly reduce these CO₂ emission.

Using a powerful artificial intelligence approach called “deep reinforcement learning,” researchers conducted an in-depth impact assessment of factors affecting vehicle emissions in three major U.S. cities.

Their analysis shows that full adoption of eco-driving measures can reduce urban cross-carbon emissions in the annual range by 11% to 22%, without slowing down traffic or affecting vehicle and traffic safety.

Researchers found that even if only 10% of vehicles on the road use eco-driving, this would result in a 25% to 50% reduction in total carbon dioxide emissions.

Additionally, at about 20% of the intersection, the dynamic optimization speed limit provides 70% of the total emission benefits. This suggests that ecological driving measures can be implemented step by step while still having a measurable positive impact on climate change mitigation and improving public health.

“Vehicle-based control strategies, such as ecological driving, can reduce climate change. We show here that modern machine learning tools, such as deep reinforcement learning, can accelerate the types of analytical support for socio-technical decisions. This is just the tip of the iceberg. This is just the tip of the iceberg. (IDSS) is a member of the MIT and the Laboratory of Information and Decision Systems (LIDS).

Vindula Jayawardana, the MIT graduate student lead author, joined the paper. and MIT graduate students AO QU, Cameron Hickert and Edgar Sanchez; MIT undergraduate Catherine Tang; Baptiste Freydt, graduate student at Eth Zurich; and Mark Taylor and Blaine Leonard of the Utah Department of Transportation. The study appeared in Transportation Research Part C: Emerging Technologies.

Multipart Modeling Research

Traffic control measures often call for fixed infrastructure, such as stop signs and traffic signals. But as vehicles become more technically advanced, it offers opportunities for ecological driving, the whole term for vehicle-based traffic control measures such as using dynamic speeds to reduce energy consumption.

In the short term, eco-driving may involve speed guidance in the form of a car dashboard or smartphone app. In the long run, eco-driven may involve smart speed commands that can directly control the acceleration of semi-autonomous and fully automatic vehicles through the vehicle-to-infrastructure communication system.

“Most of the previous work focused on how Implement ecological driving. We changed the framework and considered the problem we should We implement ecological driving. If we were to deploy this technology at a large scale, would we make a difference? “Wu said.

To answer this question, the researchers began a multifaceted modeling study that would take most of four years to complete.

They first identified 33 factors that affect vehicle emissions, including temperature, road level, intersection topology, vehicle age, traffic demand, vehicle type, driver behavior, traffic signal timing, road geometry, etc.

“One of the biggest challenges is to make sure we are diligent and not overlooking any major factors,” Wu said.

They then used data from OpenStreetMap, the U.S. Geological Survey and other sources to create digital replicas of over 6,000 signal intersections in three cities (Atlanta, San Francisco and Los Angeles) and simulated over a million traffic conditions.

Researchers use deep reinforcement learning to optimize each program to achieve an ecologically driven to achieve maximum emission benefits.

Reinforcement learning optimizes vehicle driving behavior through trial and error interaction with high-fidelity traffic simulators, thus rewarding more energy-efficient vehicle behaviors while penalizing those that do not.

The researchers viewed the problem as a decentralized cooperative multi-agent control problem, even in non-participating vehicles, vehicles collaborate to achieve overall energy efficiency, and they act in a decentralized manner, avoiding expensive communications between vehicles.

However, training vehicle behaviors summarized in various intersection traffic situations is a major challenge. The researchers observed that some scenarios were more similar than others, such as scenarios with the same number of lanes or the same number of traffic signal phases.

Therefore, the researchers trained separate reinforcement learning models for different traffic scenarios, thus making overall emission benefits better.

But even with the help of AI, analyzing the citywide traffic at the network level will be so computationally intensive, and it may take ten years to disband.

Instead, they solved the problem and solved each ecological driving scenario at various intersections.

“We carefully limit the impact of each intersection on ecological driving controls at adjacent intersections. In this way, we greatly simplify the problem, which allows us to perform this analysis at scale without introducing unknown network effects,” she said.

Significant emission benefits

When they analyzed the results, the researchers found that full adoption of ecological drives could lead to a reduction in intersection emissions between 11 and 22%.

These benefits depend on the layout of the city streets. The less room for dense cities like San Francisco to implement an eco-driven between intersections provides a possible explanation for reducing emissions savings, while Atlanta can see higher speed limits.

Even if only 10% of vehicles use eco-driving, cities can still realize that 25% to 50% of the city’s total emissions due to the dynamics of cars: non-ECO-driving vehicles will follow controlled eco-driving vehicles as they optimize the speed at which smoothly passes through the group, thus reducing the power generation of carbon.

In some cases, eco-driven can also increase vehicle throughput by minimizing emissions. However, Wu warned that increasing throughput could lead to more drivers entering the road, reducing emission benefits.

Although their widely used safety indicators (such as the timing of collisions) as known as alternative safety measures suggest that ecological driving is as safe as human driving, it can lead to unexpected behaviors of human drivers. Wu said more research is needed to fully understand the potential security impact.

Their results also show that eco-driven can provide greater benefits when used in combination with alternative transport decarbonization solutions. For example, using 20% eco-driving in San Francisco will reduce emissions by 7%, but when hybrid and electric vehicles are expected, it will reduce emissions by 17%.

“This is the first attempt to systematically quantify the environmental benefits within the network of ecological driving. This is a great research effort that will be a major reference for others in evaluating ecological driving systems,” said Samuel L. Pritchard, a Virginia engineer.

Although researchers focus on carbon emissions, benefits are highly correlated with improvements in fuel consumption, energy use and air quality.

“It’s almost a free intervention. We already have smartphones in cars and we are quickly adopting cars with more advanced automation capabilities. To quickly scale in practice, shovels must be implemented and removed relatively simply.

This work is partly funded by Amazon and the Utah Department of Transportation.