Self-driving cars learn to share road knowledge through digital word-of-mouth
NYU Tandon-led research team develops system that lets vehicles pass along AI models like messages in a social network, even when they don't meet directly
NYU Tandon School of Engineering
An NYU Tandon School of Engineering-led research team has developed a way for self-driving vehicles to share their knowledge about road conditions indirectly, making it possible for each vehicle to learn from the experiences of others even when they rarely meet on the road.
The research, presented in a paper at the Association for the Advancement of Artificial Intelligence Conference on February 27, 2025, tackles a persistent problem in artificial intelligence: how to help vehicles learn from each other while keeping their data private. Typically, vehicles only share what they have learned during brief direct encounters, limiting how quickly they can adapt to new conditions.
"Think of it like creating a network of shared experiences for self-driving cars," said Yong Liu, who supervised the research led by his Ph.D. student Xiaoyu Wang. Liu is a professor in NYU Tandon’s Electrical and Computer Engineering Department and a member of its Center for Advanced Technology in Telecommunications and Distributed Information Systems and of NYU WIRELESS.
"A car that has only driven in Manhattan could now learn about road conditions in Brooklyn from other vehicles, even if it never drives there itself. This would make every vehicle smarter and better prepared for situations it hasn't personally encountered,” Liu said.
The researchers call their new approach Cached Decentralized Federated Learning (Cached-DFL). Unlike traditional Federated Learning, which relies on a central server to coordinate updates, Cached-DFL enables vehicles to train their own AI models locally and share those models with others directly.
When vehicles come within 100 meters of each other, they use high-speed device-to-device communication to exchange trained models rather than raw data. Crucially, they can also pass along models they’ve received from previous encounters, allowing information to spread far beyond immediate interactions. Each vehicle maintains a cache of up to 10 external models and updates its AI every 120 seconds.
To prevent outdated information from degrading performance, the system automatically removes older models based on a staleness threshold, ensuring that vehicles prioritize recent and relevant knowledge.
The researchers tested their system through computer simulations using Manhattan’s street layout as a template. In their experiments, virtual vehicles moved along the city’s grid at about 14 meters per second, making turns at intersections based on probability, with a 50% chance of continuing straight and equal odds of turning onto other available roads.
Unlike conventional decentralized learning methods, which suffer when vehicles don’t meet frequently, Cached-DFL allows models to travel indirectly through the network, much like how messages spread in delay-tolerant networks, which are designed to handle intermittent connectivity by storing and forwarding data until a connection is available. By acting as relays, vehicles can pass along knowledge even if they never personally experience certain conditions.
"It's a bit like how information spreads in social networks," explained Liu. "Devices can now pass along knowledge from others they've met, even if those devices never directly encounter each other."
This multi-hop transfer mechanism reduces the limitations of traditional model-sharing approaches, which rely on immediate, one-to-one exchanges. By allowing vehicles to act as relays, Cached-DFL enables learning to propagate across an entire fleet more efficiently than if each vehicle were limited to direct interactions alone.
The technology allows connected vehicles to learn about road conditions, signals, and obstacles while keeping data private. This is especially useful in cities where cars face varied conditions but rarely meet long enough for traditional learning methods.
The study shows that vehicle speed, cache size, and model expiration impact learning efficiency. Faster speeds and frequent communication improve results, while outdated models reduce accuracy. A group-based caching strategy further enhances learning by prioritizing diverse models from different areas rather than just the latest ones.
As AI moves from centralized servers to edge devices, Cached-DFL provides a secure and efficient way for self-driving cars to learn collectively, making them smarter and more adaptive. Cached-DFL can also be applied to other networked systems of smart mobile agents, such as drones, robots and satellites, for robust and efficient decentralized learning towards achieving swarm intelligence.
The researchers have made their code publicly available. More detail can be found in their technical report. In addition to Liu and Wang, the research team consists of Guojun Xiong and Jian Li of Stony Brook University; and Houwei Cao of New York Institute of Technology.
The research was supported by multiple National Science Foundation grants, the Resilient & Intelligent NextG Systems (RINGS) program — which includes funding from the Department of Defense and the National Institute of Standards and Technology — and NYU’s computing resources.
Roadway safety research, automated vehicle testing join forces at U-M
As the U-M Transportation Research Institute turns 60, it expands to include Mcity and fast-track AV technologies as the next frontier in roadway safety
University of Michigan
In an effort to bolster its research on next-generation mobility technologies that save lives, the University of Michigan is fusing its longstanding leadership in transportation safety with its distinct expertise in testing connected and automated vehicle technologies.
Beginning March 1, U-M's Transportation Research Institute (UMTRI) will expand to include the Mcity public/private partnership and test facility. Henry Liu, Mcity's director since 2022, will lead the expanded UMTRI.
"We see this as a multiplier of our impact," said Liu, the Bruce D. Greenshields Collegiate Professor of Engineering and a professor of civil and environmental engineering. "UMTRI brings top safety researchers and a focus on the human facets, while Mcity provides technology development and a one-of-a-kind test facility. Together they've shaped Michigan Engineering's reputation as a leader in mobility work that not only advances the engineering aspects, but goes beyond to examine how they impact peoples' lives."
Liu takes over for James Sayer, UMTRI's director for the past nine years, who will remain as a research scientist.
The change is a homecoming of sorts for the Mcity Test Facility. UMTRI leaders including Sayer—along with industry and government partners—envisioned, developed and launched Mcity in 2015 as the world's first purpose-built test track for connected and automated vehicles.
"What spun out of our efforts to improve roadway safety has come full circle, and at a time when it's clear that connected and automated vehicle technologies are the essential next frontier in traffic safety," Sayer said. "More than 40,000 people are still dying on U.S. roads every year."
The new era begins as UMTRI marks its 60th anniversary and the Mcity Test Facility its 10th.
Sixty years of saving lives
In 1965, the nation was facing an alarming rise in traffic fatalities. More Americans were owning faster cars designed for style and performance, and cruising them along the new interstate highways built the previous decade. From 1960-65, roadway deaths rose by 20% per 100,000 people, according to data from the National Safety Council.
The federal government recognized the issue and began deliberating solutions. So did Ford, GM and the Automobile Manufacturers Association. With $10 million, they founded UMTRI's predecessor, the Highway Safety Research Institute, which became UMTRI in the 1980s. It was time to add safety to vehicle engineering.
Since opening its doors, the institute has conducted foundational research that played a pivotal role in bringing fatalities down by an estimated 35% per capita, according to 2023 NRC data. Most recently, UMTRI has led connected and automated vehicle deployments that enabled it to amass the world's largest set of connected vehicle data, and use it to demonstrate how effective the tech can be at enhancing safety. But its work started with a focus on vehicle design, driver behavior, vehicle dynamics and crash analysis.
During the 1970s, researchers began work to make crash test dummies more realistic to better protect people in cars, including children. Eventually they published a landmark study of driver body shape and posture, Anthropometry of Motor Vehicle Occupants, that has served as the design basis for adult-sized crash test dummies for nearly 40 years. The work continues today.
In the following decade, the institute's research shaped design and policy around occupant restraint systems, particularly seatbelts, child safety seats and airbag technologies. As one example, a 1988 survey of child safety seat use and misuse across Michigan drew national attention and led to the creation of the Lower Anchors and Tethers for Children (LATCH) system for installing child safety seats.
Around the same time, UMTRI and its researchers established the International Roughness Index used across the world by builders to quantify road surface quality. The index continues to play a critical role in road safety and reducing vehicle repair costs.
Today's heavy truck safety standards can also trace their origin to UMTRI researchers' mathematical modeling and measurements.
By the 1990s, UMTRI had honed its advanced 3D modelling and simulation capabilities through decades of studying and validating vehicle dynamics, and startup Mechanical Simulation Corp. spun out to offer the auto industry a new way to accurately and realistically predict real-world vehicle behavior. When the company was purchased in 2022 by vehicle software supplier Applied Intuition, it supported more than 200 OEMs and Tier 1 suppliers. Today, Applied Intuition provides software used by automakers and others to advance autonomous vehicle technologies. It was valued at $6 billion in 2024.
UMTRI's record of safety work made it a natural partner for government and the auto industry as attention turned to the potential of connected and automated vehicles. Under UMTRI's leadership, roughly 2,800 drivers traveled 71 million miles in the world’s first large-scale connected vehicle deployment. Safety Pilot Model Deployment, a $30 million effort launched in 2012 with the U.S. Department of Transportation.
Safety Pilot showed that connected vehicles can reduce unimpaired crashes by 80%. The initial project has evolved over the past decade with infrastructure and technology enhancements and now totals more than $82 million. Today's Ann Arbor Connected Environment 2.0 and Smart Intersections Project encompass 27 square miles of Ann Arbor, 75 sites, including 69 intersections and relies on cellular-based C-V2X.
UMTRI's crash analysis research has continued to yield influential findings enabling an improved safety response. Partnering with the Office of Highway Safety Planning and others, including NHTSA, for more than 30 years, UMTRI has maintained high quality access to Michigan crash data statistics. The Michigan Traffic Crash Facts website is an award winning, powerful crash analysis tool that allows users to search traffic crash facts related to a wide-range of data including: age, deer, vehicle/driver and occupant information.
The institute's work also includes the benefits of active safety systems such as automatic emergency braking, and it informed General Motors' decision to make five active safety features standard on most 2023 vehicles. This includes systems that reduce crashes involving pedestrians and cyclists.
A safe place for technology testing
Early on, UMTRI leaders recognized the need to test cutting-edge connected and automated technologies in a safe environment, rather than on public roads. That's why they designed the Mcity Test Facility, a 32-acre site that recreates a real-world urban and suburban environment, complete with roads, intersections, traffic signs and signals, streetlights, building facades, sidewalks, construction obstacles and more.
The goal of the test facility was to create a space for rigorous, repeatable testing in a safe, controlled environment and to complement UMTRI's large-scale real-world deployments.
Automakers and others in the industry recognized the need to accelerate research and development of these technologies as well. Initially, 15 Leadership Circle companies pledged $1 million over three years, and 31 affiliate members $150,000, providing more than $19 million in industry support to operate the facility and fund research. Early Leadership Circle members included General Motors, Ford, Honda, State Farm, Toyota and Verizon.
Not long after the Mcity Test Facility opened in 2015, Ford became the first automaker to test autonomous vehicles there.
"Testing Ford's autonomous vehicle fleet at Mcity provides another challenging, yet safe, urban environment to repeatedly check and hone these new technologies," Raj Nair, then Ford's group vice president of Global Product Development, said at the time. "This is an important step in making millions of people's lives better and improving their mobility."
By combining both physical and virtual assets, Mcity can recreate almost any driving scenario—controlling vehicle behavior, simulating pedestrians and more. Its researchers are also working to provide and advocate for an automated vehicle testing structure for the industry and consumers. Its Mcity Safety Assessment Program is a two-part protocol for validating the safety of Level 4 automated vehicles before real-world deployment that could serve as a blueprint for a national standard. L4 vehicles can navigate most driving situations without human intervention.
Mcity's Driverless Shuttle was the nation's first AV shuttle research project on user behavior and a way to gauge consumer acceptance of the technology. It was followed by partnerships with May Mobility on trial AV shuttle deployments in Ann Arbor and Detroit.
Today, Mcity offers remote testing too, made possible using digital infrastructure developed with funding from the National Science Foundation. The digital and physical infrastructure work together to allow researchers around the world to use the test facility without leaving home, helping to speed up development of automated technologies.
Also, Mcity recently introduced the first open-source digital twin of the test facility, providing a faster, safer, less expensive way to test autonomous and connected vehicle software. The digital twin is free for anybody to use and does not require a physical vehicle or test facility.
A tipping point for road safety
Liu believes that bringing UMTRI and Mcity together has the potential to accelerate a tipping point for road safety. While traffic fatalities per 100,000 people are less than half of what they were at their worst in the 1970s—around 13 people vs. 28—more than 40,000 are still dying on U.S. roads every year. The USDOT calls it a "crisis" and has established the National Roadway Safety Strategy to address it. The strategy includes automated technologies, and Liu underscores their importance.
"Given all the safety features that have been added to vehicles over the years, it's my view that the only thing that will significantly reduce the number of roadway fatalities at this point is automation," Liu said.
And not automation in a vacuum.
"We need to think about how to protect AV occupants. Maybe the passengers shouldn't be facing forward. Maybe the seatbelts and airbags should be designed differently," Liu said. "In transportation, you really have to have a systems view, and this new structure will emphasize that."
No comments:
Post a Comment