Fast Facts
-
Innovative Knowledge Sharing: NYU Tandon’s research team developed Cached Decentralized Federated Learning (Cached-DFL), enabling self-driving cars to learn from each other’s experiences about road conditions indirectly while maintaining data privacy.
-
Efficient Model Exchange: Vehicles can exchange trained AI models directly through high-speed communication when in proximity, allowing knowledge to spread even among cars that seldom meet, enhancing adaptability to diverse road scenarios.
-
Dynamic Learning System: Cached-DFL allows vehicles to maintain a cache of up to 10 models, updating their AI every 120 seconds while discarding outdated information, ensuring that they prioritize recent, relevant knowledge for efficient learning.
- Broad Application Potential: This technology not only improves self-driving cars but can also be extended to other smart mobile agents like drones and robots, promoting decentralized learning and swarm intelligence across various networks.
Self-Driving Cars Learn to Share Road Knowledge Through Digital Word-of-Mouth
A team at NYU Tandon School of Engineering has made a breakthrough in self-driving car technology. They developed a way for these vehicles to share insights about road conditions indirectly. This innovation allows each car to learn from its peers, even when they rarely encounter each other on the road.
On February 27, 2025, researchers presented their findings at the Association for the Advancement of Artificial Intelligence Conference. The study addresses a key challenge in artificial intelligence: enabling vehicles to learn collaboratively while keeping their data private. Traditionally, cars have shared insights only during brief direct interactions, which slows their ability to adapt to changing conditions.
“Think of it like creating a network of shared experiences for self-driving cars,” said Yong Liu, the research supervisor. Liu, a professor at NYU Tandon, emphasized the importance of this development. “A car that has driven only in Manhattan could learn about Brooklyn’s roads from others, making every vehicle smarter.”
The researchers coined their approach as Cached Decentralized Federated Learning, or Cached-DFL. Unlike traditional methods, Cached-DFL does not depend on a central server. Instead, vehicles can train AI models locally and share these models directly with others.
When cars come within 100 meters, they use rapid communication to exchange trained models. This process allows vehicles to share knowledge from previous encounters, spreading information further than just immediate interactions. Each car can store up to 10 external models and refresh its AI every two minutes.
To keep the data relevant, outdated models are removed based on a staleness threshold. This ensures that cars prioritize the most useful and up-to-date information.
The research team conducted simulations using Manhattan’s streets. They found that, unlike conventional methods, Cached-DFL allows knowledge to flow through the network, similar to how information spreads in social media. Vehicles can relay knowledge imparted from others, enhancing the overall intelligence of the fleet.
This innovative mechanism reduces the constraints of traditional learning methods. Vehicles now act as relays, extending learning opportunities across wider areas. This could prove vital in urban settings, where road conditions often vary significantly.
The study also highlighted several factors affecting learning efficiency, including vehicle speed and cache size. Faster speeds and frequent communications yield better results, while outdated models can impair accuracy. Additionally, a strategy focusing on diverse models from different areas may improve learning outcomes.
As AI technology moves towards decentralized systems, Cached-DFL offers a secure way for self-driving cars to learn collectively. This approach may also benefit other networks of smart mobile agents, including drones, robots, and satellites.
The team has made their code publicly available, inviting further exploration in the field. Researchers received support from National Science Foundation grants and various governmental programs. Alongside Liu and lead researcher Xiaoyu Wang, the team included experts from Stony Brook University and the New York Institute of Technology.
As self-driving technology continues to evolve, innovations like Cached-DFL could enhance vehicle performance and safety, paving the way for a smarter driving future.
Discover More Technology Insights
Learn how the Internet of Things (IoT) is transforming everyday life.
Access comprehensive resources on technology by visiting Wikipedia.
SciV1
