If there’s one thing that characterizes driving in any major city, it’s the constant stop-and-go as traffic lights change and as cars and trucks merge and separate and switch and park. This constant stopping and starting is incredibly inefficient, driving up the quantity of pollution, including greenhouse gases, that gets emitted per mile of driving.
One approach to counter that is generally known as eco-driving, which might be installed as a control system in autonomous vehicles to enhance their efficiency.
How much of a difference could that make? Would the impact of such systems in reducing emissions be definitely worth the investment within the technology? Addressing such questions is one in every of a broad category of optimization problems which were difficult for researchers to handle, and it has been difficult to check the solutions they give you. These are problems that involve many alternative agents, reminiscent of the many alternative sorts of vehicles in a city, and various factors that influence their emissions, including speed, weather, road conditions, and traffic light timing.
“We got interested a number of years ago within the query: Is there something that automated vehicles could do here by way of mitigating emissions?” says Cathy Wu, the Thomas D. and Virginia W. Cabot Profession Development Associate Professor within the Department of Civil and Environmental Engineering and the Institute for Data, Systems, and Society (IDSS) at MIT, and a principal investigator within the Laboratory for Information and Decision Systems. “Is it a drop within the bucket, or is it something to take into consideration?,” she wondered.
To deal with such an issue involving so many components, the primary requirement is to collect all available data in regards to the system, from many sources. One is the layout of the network’s topology, Wu says, on this case a map of all of the intersections in each city. Then there are U.S. Geological Survey data showing the elevations, to find out the grade of the roads. There are also data on temperature and humidity, data on the combination of auto types and ages, and on the combination of fuel types.
Eco-driving involves making small adjustments to attenuate unnecessary fuel consumption. For instance, as cars approach a traffic light that has turned red, “there’s no point in me driving as fast as possible to the red light,” she says. By just coasting, “I’m not burning gas or electricity within the meantime.” If one automotive, reminiscent of an automatic vehicle, slows down on the approach to an intersection, then the traditional, non-automated cars behind it is going to even be forced to decelerate, so the impact of such efficient driving can extend far beyond just the automotive that’s doing it.
That’s the fundamental idea behind eco-driving, Wu says. But to determine the impact of such measures, “these are difficult optimization problems” involving many alternative aspects and parameters, “so there’s a wave of interest at once in learn how to solve hard control problems using AI.”
The brand new benchmark system that Wu and her collaborators developed based on urban eco-driving, which they call “IntersectionZoo,” is meant to assist address a part of that need. The benchmark was described intimately in a paper presented on the 2025 International Conference on Learning Representation in Singapore.
Taking a look at approaches which were used to handle such complex problems, Wu says a crucial category of methods is multi-agent deep reinforcement learning (DRL), but an absence of adequate standard benchmarks to guage the outcomes of such methods has hampered progress in the sector.
The brand new benchmark is meant to handle a crucial issue that Wu and her team identified two years ago, which is that with most existing deep reinforcement learning algorithms, when trained for one specific situation (e.g., one particular intersection), the result doesn’t remain relevant when even small modifications are made, reminiscent of adding a motorcycle lane or changing the timing of a traffic light, even after they are allowed to coach for the modified scenario.
In truth, Wu points out, this problem of non-generalizability “will not be unique to traffic,” she says. “It goes back down all of the method to canonical tasks that the community uses to guage progress in algorithm design.” But because most such canonical tasks don’t involve making modifications, “it’s hard to know in case your algorithm is making progress on this sort of robustness issue, if we don’t evaluate for that.”
While there are various benchmarks which are currently used to guage algorithmic progress in DRL, she says, “this eco-driving problem includes a wealthy set of characteristics which are necessary in solving real-world problems, especially from the generalizability viewpoint, and that no other benchmark satisfies.” Because of this the 1 million data-driven traffic scenarios in IntersectionZoo uniquely position it to advance the progress in DRL generalizability. Consequently, “this benchmark adds to the richness of how to guage deep RL algorithms and progress.”
And as for the initial query about city traffic, one focus of ongoing work will probably be applying this newly developed benchmarking tool to handle the actual case of how much impact on emissions would come from implementing eco-driving in automated vehicles in a city, depending on what percentage of such vehicles are literally deployed.
But Wu adds that “quite than making something that may deploy eco-driving at a city scale, the principal goal of this study is to support the event of general-purpose deep reinforcement learning algorithms, that might be applied to this application, but additionally to all these other applications — autonomous driving, video games, security problems, robotics problems, warehousing, classical control problems.”
Wu adds that “the project’s goal is to supply this as a tool for researchers, that’s openly available.” IntersectionZoo, and the documentation on learn how to use it, are freely available at GitHub.
Wu is joined on the paper by lead authors Vindula Jayawardana, a graduate student in MIT’s Department of Electrical Engineering and Computer Science (EECS); Baptiste Freydt, a graduate student from ETH Zurich; and co-authors Ao Qu, a graduate student in transportation; Cameron Hickert, an IDSS graduate student; and Zhongxia Yan PhD ’24.
