The future of testing self-driving cars
Photos by Evan Krape and courtesy of IDS Lab January 25, 2023
UD Prof. Andreas Malikopoulos’ ‘Scaled Smart City’ bridges the gap between driving simulations and real-world tests
On the second floor of the University of Delaware’s Spencer Lab, a 20x20 foot model city — complete with tree-lined roundabouts and brightly-colored shops and houses — is where researchers in the Information and Decision Science (IDS) Laboratory try out the latest innovations for self-driving cars within the safe confines of a controlled setting.
The work of designing, developing and testing new algorithms all takes place in this Scaled Smart City, run by Andreas Malikopoulos, the Terri Connor Kelly and John Kelly Career Development Associate Professor in the Department of Mechanical Engineering. Now, with new publications and research grants in hand, the group is poised to continue its work toward ensuring that AI-piloted vehicles safely share the road with human drivers in the future.
After coming to UD in 2017, Malikopoulos built the Scaled Smart City to bridge the gap between driving simulations and real-world tests. He explained that entirely virtual simulations don’t accurately consider all of the errors that happen in the real world, while street-level tests with full-sized cars can be dangerous without appropriate safety measures.
“The problem with simulations is that they are doomed to succeed,” Malikopoulos said. “When you run a simulation, everything runs perfectly — there’s no miscommunication between vehicles, for example. But when you go into the real world, there’s information delays, GPS errors — and those communication errors can have severe negative implications in safety. The advantage of our Scaled Smart City is that it can help us see and address all of the drawbacks with our algorithms before we do real-world testing in a vehicle.”
One of the biggest challenges faced by researchers working in the field of connected and automated vehicles (CAVs), Malikopoulos said, is integrating all of the information that a CAV needs to drive safely while not having the algorithm be so complex that it requires a supercomputer to perform.
“We are working on how to use the information from the vehicle to coordinate with one another to avoid stop-and-go driving and to learn what to do when there is a conflict point, like an intersection, roundabout or construction zone, etc.,” he said. “To accomplish this, we use control theory to develop our algorithms that can handle these challenges efficiently.”
One component of IDS’s research program is led by doctoral student Heeseung Bang, whose project is focused on developing a control framework for CAVs that combines routing, coordination and control. His approach involves solving decision-making problems at different levels faced by CAVs on the road, such as selecting which way to go and what to do at a specific intersection, in combination with advanced machine learning algorithms to help CAVs predict traffic conditions and understand how to deal with complex human behaviors.
“We're solving a huge optimization problem by solving smaller problems and combining those results effectively,” Bang said. “For example, we can make CAVs coordinate themselves to cross intersections without stop-and-go driving, and then use their information to predict traffic congestion at the intersections and assign the fastest route for each vehicle.”
Infrastructure for testing new algorithms
Recently, the group published an article on how its Scaled Smart City can be used to test and validate new CAV control algorithms before taking them out into the real world. Featured on the cover of the December 2022 issue of IEEE Control Systems, the article also describes specific case studies that show how its algorithms help CAVs safely travel through specific traffic scenarios, including roundabouts, intersections and merging road ways.
“We showed how the testbed can help us prove a new concept for emerging mobility systems while also understanding the gap between theory and real-world implementation,” said Bang, who co-wrote the paper along with Malikopoulos and Behdad Chalaki (controls engineer at Honda Research Institute), A.M. Ishtiaque Mahbub (controls engineer at Aptiv) and Logan Beaver (a postdoc at Boston University), all of whom are graduates from the IDS Lab.
“On the educational front, we reported in the article how our Scaled Smart City has been used for training and educating graduate students by exposing them to a balanced mix of theory and practice, integrating research outcomes into existing courses, involving undergraduate students in research, creating interactive educational demos and reaching out to K-12 students,” Malikopoulos said.
The group was also recently recognized with a Best Paper Award at the IEEE second annual International Conference on Digital Twins and Parallel Intelligence for the “digital twin” of the Scaled Smart City, an entirely virtual version of the set-up that collaborators can access remotely.
Malikopoulos said that this digital twin, which was created during the height of the COVID-19 pandemic, allows researchers to develop and test new algorithms in a completely virtual setting before running them on the Scaled Smart City.
“Once the algorithm runs smoothly in the digital twin, you can implement it easily in the physical test,” he said.
Transportation equity and human-CAV interactions
The IDS Lab will be expanding its research in this area thanks to two National Science Foundation grants, one looking at equity in transportation systems and another to study human-CAV interactions.
For the transportation equity project, the lab will be working with collaborators at Boston University and MIT on a $1.2 million project that combines data on user preference, energy usage and travel time to propose travel incentives, such as train ticket discounts, that lead to transportation systems with broader economic, environmental and social benefits. The goal is to “learn from and also train the travelers,” said Malikopoulos, with experiments conducted in the Scaled Smart City followed by a larger experiment in the city of Boston.
“The goal of this project is to promote incentives that guarantee equity in transportation so that all travelers have equal opportunities to accessibility,” he said. “Equity is very abstract, though, and the challenge that we’re working on is how to integrate equity into a mathematical framework. This is a very hard constraint, but it’s been really rewarding to see how control theory can help address some of these challenges.”
To study human-CAV interactions, the lab was awarded a nearly $500,000 grant with which it will combine data on human driving behavior with control theory and reinforcement learning approaches. That way, CAVs can “learn” how to react appropriately to the wide variety of scenarios that can occur, as driving patterns vary widely from person to person — from overly cautious or uncertain drivers to ones that are more aggressive.
“When you have a human in the loop, the problem becomes quite challenging due to their unexpected behavior. Using our scaled city, we have the necessary infrastructure to conduct experiments without safety concerns,” Malikopoulos said. “We are very excited about this project because this is a critical problem in the field at the moment: How we can have humans interact with autonomous vehicles in an efficient and safe way.”
Visualizing the future of smart cities
While fleets of self-driving cars won’t appear on the streets overnight, Malikopoulos said that now is the time for researchers to begin visualizing the future of “smart cities” by developing the algorithms needed to help these complex systems operate safely and effectively.
“We have made a lot of progress, and these technologies are mature enough that you will start seeing them deployed in controlled environments pretty soon,” Malikopoulos said. “We have a long way to go, though. Everything is happening at the fast pace, which is why it’s important for us to take rigorous and solid research steps. If we do that, then I think we can get to the finish line successfully.”
Bang, who is interested in continuing research on CAVs after graduation, enjoys working on something that can provide a “sketch of a future transportation system” and a glimpse into how transportation corridors could look soon.
“By working in this field, it's exciting to see and predict how this technology could affect the future,” he said. “And if we could also improve safety or reduce energy consumption, it would be very rewarding.”