Hyundai puts Data Flywheel into full operation, sets 2028-2029 self-driving roadmap

Hyundai Motor Group's announcement comes as it seeks a competitive advantage in autonomous driving through accelerated learning and technology development. The group is leveraging its annual sales of 7 million vehicles and a Data Union framework to expand its data ecosystem, while implementing a data-centric development framework based on hard example mining, continuous training, and SER. This approach aims to continuously advance AI capabilities by creating a virtuous cycle of data collection, training, validation, and deployment.

The strategy is dual-track: accelerate Level 2+ production through collaboration with NVIDIA while also internalizing key autonomous driving technologies through proprietary Atria AI. Progressive sensor standardization across Hyundai Motor, Kia, 42dot, and Motional is intended to bring together autonomous driving capabilities across the Group. Additionally, 42dot shared its Vision-Language-Action (VLA)-based autonomous driving technology, which integrates visual information and language-based reasoning to guide driving decisions. VLA research focuses on addressing edge cases, with parallel development of end-to-end autonomy and VLA models to enhance technical stability and scalability.

The outcome hinges on execution across these tracks. The NVIDIA-based Level 2+ production in H1 2028 and Level 2++ in H2 2028, followed by Atria AI-powered Level 2++ vehicles in H2 2029, represent ambitious targets. The real-world Level 4 pilot in Gwangju by year-end, in partnership with Korea's Ministry of Land, Infrastructure and Transport, is likely intended to secure large-scale validation data. For Hyundai, Kia, 42dot, and Motional, success depends on integrating these technologies and validating them on public roads. The VLA model validation and on-road testing scheduled to run through early next year will be a key milestone to watch.