The city resembles any number of urban centers – that’s the point. It has blocks of residential houses and commercial businesses, landscaped parks, roads, roundabouts, and traffic lights. There is a variety of vehicles, even a few police cruisers that have pulled over speeding motorists.
And it all fits in a single room in the basement of Hollister Hall.
Mini drones fly over the scaled city and broadcast positioning information, a tactic that could help coordinate mixed traffic, enhance safety and would be particularly valuable for protecting military convoys.
Welcome to the Information and Decision Science Laboratory. Here, a 20-by-20-foot “smart” city shrunk to 1:25 scale and its fleet of custom-built cars, drones, cameras and virtual reality technology are helping researchers design a better – and safer – future for transportation.
Specifically, the scaled smart city seeks to improve the performance of emerging mobility systems, such as connected and automated vehicles, by enabling small motorized cars to interact with their environment as well as other vehicles, some of which are piloted remotely by humans. Because the city is a controlled environment, experiments can be repeated and results verified. And yet, unpredictability is encouraged.
“If you don’t have an experimental testbed, you use simulation. And simulations are doomed to succeed. They’re always perfect,” said the lab’s director, Andreas Malikopoulos, professor of civil and environmental engineering in Cornell Engineering. “But in the real environment, you have miscommunication, errors, delays, unexpected events. This testbed can give us the opportunity to collect data and extrapolate information, something that we couldn’t do in the real world with real cars, because of safety concerns and the need for resources and space.”
The origins of the scaled city go back more than a decade, when Malikopoulos, previously a researcher at General Motors, was working at Oak Ridge ³Ô¹ÏÍøÕ¾ Laboratory. He was thinking about ways to improve energy efficiency in transportation via connected and automated vehicles.
“I said, OK, well, we can do all this,” he said. “But we need to have some hardware.”
Bad drivers wanted
In 2014, he received a grant for the efficiency project from the U.S. Department of Energy (DOE), and he pitched the creation of a scaled robotic testbed, but was told there wasn’t a need for one. Three years later, he left government research for a faculty position at the University of Delaware, where he was finally able to realize his “dream” of building the testbed. When Malikopoulos moved to Cornell in fall 2023, he scaled up – not in size, but in sophistication – by adding new features and upgrading his motorized fleet.
The lab is now on its fifth generation of the vehicles. The team members, many of them undergraduates, take off-the-shelf remote-control cars, gut their RC components and install custom electronics, sensors and Wi-Fi that connects to a mainframe computer. Each vehicle has a unique visual marker so it can be tracked by the testbed’s eight cameras and GPS system. Over the summer, the motor pool expanded to 75 cars and 15 mini drones. The aerial element allows for multiple drones to follow a vehicle and broadcast positioning information, a tactic that could help coordinate mixed traffic, enhance safety and would be particularly valuable for protecting military convoys.
The students don’t only work on hardware and software. They also get to be in the driver’s seat: The lab has six driver emulators where anyone can sit and “drive” through the scaled city, navigating from their car’s perspective and interacting with other vehicles on the road. The students can even communicate with those driverless vehicles through lab-designed modules that incorporate large language models.
Postdoctoral researcher Heeseung Bang, Ph.D. ’24, sets up test vehicles that have been equipped with custom electronics, sensors and Wi-Fi that connects to a mainframe computer.
The combination of automated and human-driven vehicles – also known as a cyber-physical system – introduces a level of uncertainty that can help the researchers better understand the ways connected and automated vehicles react in real time. The researchers can then design controllers and algorithms that will improve such responses.
Lab members also have the option of donning a VR headset to drive in a computer-simulated city and experience different traffic scenarios and conditions. The VR system can produce a digital twin of the testbed for more interactive driving.
“The benefit of VR is cost effectiveness, time efficiency and most importantly safety,” said Simon Tian, M.S. ’24 in systems engineering, the project lead of the VR testbed. “Experimenting with real vehicles is very expensive and very time consuming. Each experiment, you have to reset everything involving the vehicle, the people in the places, objects. If you want a specific experiment on a specific road, you have to build that road. If you want to do testing in Times Square, there’s no way you can have that place for yourself to test yet. But you can do it in VR. The technology is quite feasible.”
Tian and his team of six built their entire VR system from scratch using an open-source autonomous driving simulator called CARLA and their own Python coding.
Now the lab is the one place where bad driving is not only tolerated – it is required.
“We basically use machine learning to learn to model how humans make decisions,” said postdoctoral researcher Heeseung Bang, Ph.D. ’24. “In order to do that, you need a different kind of human driver’s data. But to collect those data, we need VR, so that people can actually just crash, and we see how aggressive drivers actually drive. Once we have the dataset, we can implement it on the city and see how they react.”
The vehicles introduce plenty of uncertainty of their own.
“This is a very complex system, and even if you repeat the same scenario, cars can make a different decision based on their different conditions, like, depending on their speed and time of entering the intersection,” Bang said.
Smarter vehicles, safer roads, greater equity
While the lab is focused on transportation, there is a parallel interest in social impact, and Malikopoulos is keen to explore ways to increase equity in the cyber-physical systems space. He is currently working with the Federal Highway Administration to make the testbed accessible as an educational module to other institutions, and he plans to create an online platform so researchers can log in, enter their algorithms and use the testbed to run experiments of their own.
In October, Malikopoulos received a collaborative $800,000 grant from the ³Ô¹ÏÍøÕ¾ Science Foundation’s Safe Learning-Enabled Systems program to establish a framework for smart autonomous systems to be deployed in complex operating environments while ensuring the systems can handle extreme events and monitoring them for irregular and unsafe behavior. The mathematical computing software company MathWorks recently provided a grant to use the scaled city testbed to develop real-time control toolboxes of autonomous driving systems.
The lab also recently completed a $4.3 million project from the DOE’s Advanced Research Projects Agency – Energy, in collaboration with the University of Michigan, Boston University, the Oak Ridge lab and Robert Bosch GmbH that improved the efficiency of Audi’s A3 plug-in hybrid by 25%.
The scaled city, located in the basement of Hollister Hall, seeks to improve the performance of emerging mobility systems, such as connected and automated vehicles, by enabling small motorized cars to interact with their environment as well as other vehicles, some of which are piloted remotely by humans.
“We used the scaled city to develop all algorithms and address technical challenges,” Malikopoulos said. “So when we went for field testing with the real Audi, we were ready.”
In addition to informing the design of the next generation of connected and automated vehicles, the scaled city is helping Malikopoulos train the next generation of engineers who will be working with these technologies. It turns out that a miniature city is a great recruitment tool.
That was the case for Shanting Wang, M.S. ’24, a doctoral student in systems engineering, who was invited last year by one of the lab members to check out the testbed. Wang was duly impressed, both by the lab’s technical innovations and its practical applications for civil engineering, and she joined the team.
“Transportation is about planning, but before, people did the planning based on historical data. For example, on Monday morning, there will be more traffic flow. But they just approximate,” she said. “Now, we can get the real-time data from the sensors, and from the real data we improve our algorithm and improve our controller. It can make traffic less congested. This is a place where I can find the solution, and it can be applied to the real world.”
The lab is expanding into an adjacent room to accommodate a second testbed. Malikopoulos has even toyed with the idea of increasing the scale of the city. He estimates he could possibly go as large as 1:10.
Sometimes smaller really is smarter.
For now, Malikopoulos has his dream city, which includes, as it must, a dream car.
In the real world, Malikopoulos drives a Fiat 500C.
“That’s something that my grandpa used to drive back in Greece – the original,” he said. “But if I could afford it, I would most likely have this.”
He pointed to a small, shiny car parked between miniature buildings on a fake lawn.
It was a red Ferrari.
