Hyundai Motor Group Launches Autonomous Driving AI Data Flywheel, to Mass-Produce L2+ System in 2028

2026-09-14 11:33
Favorite

en.Wedoany.com Reported - On September 13, Hyundai Motor Group held an autonomous driving media day at the 42dot headquarters in Gyeonggi Province, South Korea, announcing that its "Data Flywheel" system for autonomous driving development has been put into full operation, and unveiling its implementation timeline for L2+ to L4 autonomous driving technology. The Group plans to mass-produce an L2+ system based on NVIDIA solutions in the first half of 2028, advance L2++ mass production in the second half of 2028, and mass-produce L2++ models equipped with its self-developed Atria AI in the second half of 2029.

Hyundai Motor Group is unifying the autonomous driving sensor architectures of Hyundai Motor, Kia, 42dot, and Motional, with the NVIDIA ecosystem as the standardized foundation, enabling data collected by different brands and development entities to enter the training, validation, and vehicle deployment stages under a unified standard. Its data flywheel operates as a closed loop of "vehicles collect data—AI model training—validation—deployment to mass-production vehicles—generation of new data," with mass-production vehicles serving as the data entry point for continuous training of subsequent autonomous driving models.

The Group currently has approximately 40 dedicated data collection vehicles continuously gathering real-world road information, covering scenarios including road construction, severe weather, sudden lane changes, emergency maneuvers, narrow-road passing, and complex urban traffic. Hyundai Motor and Kia sell more than 7 million vehicles annually in approximately 190 countries and regions, and the Group plans to incorporate vehicle data at a larger scale into the Data Union data system as sensor standardization gradually advances across mass-production models.

In its near-term validation plans, the Group intends to launch an L4 autonomous driving real-road pilot in Gwangju, South Korea, by the end of 2026, in cooperation with South Korea's Ministry of Land, Infrastructure and Transport to obtain large-scale road validation data. 42dot is simultaneously advancing its proprietary end-to-end autonomous driving system Atria AI, and has demonstrated L2++-level driving capabilities of an SDV test vehicle equipped with the system in complex urban traffic environments.

Atria AI adopts an end-to-end technical approach and was jointly developed by Hyundai Motor Group's Advanced Vehicle Platform Division and 42dot. In addition to the existing end-to-end model, 42dot is also developing a vision-language-action (VLA) autonomous driving model that combines vehicle visual information with language reasoning to handle complex and low-frequency scenarios. The VLA model is currently in the validation stage, with road testing and this full round of R&D validation planned to continue until early 2027.

Computing infrastructure will scale in tandem with the volume of vehicle data. Hyundai Motor previously announced that its AI data center in Saemangeum, South Korea, is scheduled to begin operation in 2029, with a facility scale of 100MW capable of housing more than 50,000 GPUs, used to process autonomous driving and software-defined vehicle data generated by the Group's mass-production vehicles worldwide. The facility will be connected to the Group's internal AI systems, mass-production vehicle data, and autonomous driving training platforms.

Under the current roadmap, the Group has formed two parallel technology paths: one advancing the mass production of L2+ and L2++ models in 2028 through the NVIDIA platform; the other involving 42dot's continued development of its proprietary Atria AI, which is planned to be equipped on mass-production L2++ models in the second half of 2029, and thereafter gradually expand to higher levels of autonomous driving capability based on real-world vehicle data.

This bulletin is compiled and reposted from information of global Internet and strategic partners, aiming to provide communication for readers. If there is any infringement or other issues, please inform us in time. We will make modifications or deletions accordingly. Unauthorized reproduction of this article is strictly prohibited. Email: news@wedoany.com