Autonomous vehicles can leverage an 'artificial intelligence-driven social platform' for communication while traversing roads.
Self-driving vehicles could significantly improve their performance through a novel data-sharing system, known as Cached Decentralized Federated Learning (Cached-DFL). This groundbreaking framework allows cars to pass each other and exchange valuable information about navigation, traffic patterns, road conditions, and traffic signals without the need for direct connections or permission.
Researchers at New York University's Tandon School of Engineering have designed Cached-DFL, an innovative artificial intelligence model-sharing system tailored for autonomous vehicles. The system creates a quasi-social network that enables cars to view and learn from each other's driving discoveries, even without exchanging personal information or driving patterns.
Typically, vehicles share data they've collected during their travels when they are virtually next to each other and grant permission. However, Cached-DFL introduces a new approach, allowing cars to store data and share insights when there's a vehicle-to-vehicle connection.
Instead of relying on a single central location for data storage, the Cached-DFL system enables vehicles to carry data in trained artificial intelligence models. These models store information about driving conditions and scenarios, making the system more robust to potential large data breaches.
In essence, vehicles can learn from one another's experiences, allowing a car that has driven only in Manhattan to learn about road conditions in Brooklyn from other vehicles. The cars can share their strategies for handling scenarios similar to those encountered in Brooklyn, which could appear on different roads worldwide.
The researchers published their study on the preprint arXiv database on August 26, 2024, and presented their findings at the Association for the Advancement of Artificial Intelligence Conference on February 27, 20XX.
Scientists found that frequent and quick communications between self-driving cars could enhance the efficiency and accuracy of driving data. They carried out a series of tests, placing 100 virtual self-driving cars into a simulated version of Manhattan and setting them to "drive" in a semi-random pattern. Each car updated its 10 AI models every 120 seconds, with the cached portion of the experiment emerging from the fact that cars hold on to data and wait to share it until they have a proper vehicle-to-vehicle (V2V) connection.
The team concluded that as long as cars were within 100 meters (328 feet) of each other, they could view and share each other's information without needing to know each other. The researchers envision Cached-DFL making self-driving technology more affordable by reducing the need for computing power and minimizing reliance on centralized servers.
Next steps involve real-world testing of Cached-DFL, improving interoperability between different brands of self-driving vehicles, and enabling communication between vehicles and other connected devices like traffic lights, satellites, and road signals—a process known as vehicle-to-everything (V2X) standards. The long-term aim is to create a broader move away from centralized servers, fostering a decentralized network of smart devices that gather and process data most efficiently, promoting rapid swarm intelligence for various connected technologies.
Artificial intelligence, in the form of the Cached Decentralized Federated Learning (Cached-DFL) system, is leveraged in self-driving vehicles to improve their performance. This novel data-sharing system enables cars to learn from each other's driving experiences, with AI models stored within vehicles sharing valuable information about driving conditions and scenarios, even across different roads worldwide.
