AI to empower self-driving cars to communicate with each other on road

Researchers have developed a groundbreaking communication system that allows self-driving cars to share critical driving data with each other on the road — without the need for direct, real-time connections.

The innovation, known as Cached Decentralized Federated Learning (Cached-DFL), represents a major step forward in the way autonomous vehicles exchange information, helping to improve their performance, efficiency, and safety in a variety of driving environments. The Live Science publication has shed light on the new implementation of this technology in their recent article.

Traditionally, autonomous vehicles have relied on centralized data storage systems and required close proximity to one another in order to share driving insights. This setup has several limitations, including restricted communication range, privacy concerns, and vulnerability to centralized data breaches. Cached-DFL offers a decentralized alternative by enabling vehicles to store and carry AI-trained models containing valuable driving knowledge. These models include information about road conditions, traffic signals, navigation strategies, and common challenges encountered during driving.

The article points out that self-driving cars no longer need to be adjacent or linked through direct communication to share useful information under the Cached-DFL system. Instead, vehicles operate like members of a decentralized, quasi-social network, with each car maintaining a digital "profile" that holds its accumulated driving experiences. Other vehicles can access this anonymized data and use it to inform their own navigation and decision-making processes. Importantly, no personal driver data or identifiable driving patterns are shared, preserving privacy and improving cybersecurity.

This innovative approach will allow cars to learn from each other’s experiences — even in areas they haven’t personally travelled. Dr. Yong Liu, the lead researcher and professor at NYU’s Tandon School of Engineering, likens it to creating a network of shared memories for self-driving cars. For example, a vehicle that has only driven in Manhattan could use information shared by other cars to better navigate the roads in Brooklyn, based on similar traffic conditions or road anomalies.

According to the article, one powerful feature of Cached-DFL is its ability to generalize learning across different locations. For instance, if vehicles in Brooklyn encounter uniquely shaped potholes and develop effective ways to navigate them, that knowledge can be shared with cars in other cities or countries facing similar road hazards. This scalable learning mechanism enables continuous improvement of autonomous systems by pooling collective knowledge from geographically dispersed vehicles.

The implementation of Cached-DFL not only enhances the efficiency and adaptability of autonomous vehicles but also reduces their dependence on cloud-based servers and minimizes the risks associated with centralized data storage. By keeping data local within the vehicle’s AI model and sharing insights through a federated network, self-driving cars become smarter, more self-sufficient, and better equipped to handle complex driving environments.

By Nazrin Sadigova