The Hyundai Group intends to accelerate the development of its automated driving technologies by leveraging an abundant resource: data generated by vehicles on the road.
During a presentation at the headquarters of 42dot in South Korea, Hyundai Motor Group detailed a strategy called the “Data Flywheel”, essentially a learning loop powered by data. It is intended to progressively improve the systems offered by Hyundai, Kia and other companies within the group.
A four-step learning loop
The principle behind the Data Flywheel relies on four steps:
- 1) collecting real-world driving data;
- 2) training artificial intelligence models;
- 3) validating their reactions;
- 4) deploying improvements to vehicles.
Once new features are rolled out, additional data can be gathered to restart the process. Hyundai hopes to accelerate the rate at which its systems learn to recognize complex situations.
The automaker specifically aims to target rare or difficult scenarios, sometimes referred to as “edge cases”. These can include unusual behaviour by other motorists, partially obscured road signs and ambiguous situations in an urban environment.
By its sheer size, the Hyundai Group detains a certain advantage over rival developers of self-driving tech. The group sells around seven million vehicles annually, and a gradual standardization of sensors across Hyundai, Kia, 42dot, and Motional could allow it to build a vast and unrivaled data ecosystem, while respecting established frameworks for data collection and usage.
A two-pronged strategy
Hyundai will pursue two parallel paths. The first relies on a collaboration with NVIDIA to accelerate the commercialization of Level 2+ systems. Production of this technology is scheduled for the first half of 2028. An evolution designated as Level 2++ would follow during the second half of that same year.
In parallel, 42dot is developing Atria AI, the group's proprietary artificial intelligence technology. Vehicles equipped with a Level 2++ system powered by Atria AI are expected in the second half of 2029.
However, the designations "2+" and "2++" are not official levels within the international SAE classification. They describe systems that are more advanced than standard Level 2 driver assistance, but the driver remains responsible for driving and must stay attentive. This is not yet a fully autonomous car.
Understanding the road through images and language
42dot is also working on a technology called Vision-Language-Action. This combines information captured by cameras with a model capable of reasoning based on language. In other words, the system does not simply seek to spot a pedestrian or a vehicle; it also attempts to understand the context before choosing an action.
Hyundai is currently conducting validation trials for this technology. Road tests and other development steps are set to continue until early 2027.
Finally, a Level 4 autonomous driving pilot project is scheduled to launch in Gwangju, South Korea, before the end of 2026. Conducted alongside the Korean Ministry of Land, Infrastructure and Transport, it will serve primarily to collect large-scale validation data. At this level, the vehicle can drive without human intervention, but only within precisely defined conditions and geographic areas.