Source: Unite.AI
Hyundai Motor Group said in a September 13, 2026, announcement that it has put its Data Flywheel into full operation, detailing a dual-track autonomous driving roadmap and what it called the first showcase of its Level 2++ technology at the HMG Autonomous Driving Media Day, held at 42dot headquarters in Gyeonggi Province, South Korea.
During the event, the Group presented its autonomous driving development strategy, technology roadmap, key achievements and implementation plans. 42dot introduced key technologies and development progress for Atria AI, the Group’s proprietary autonomous driving artificial intelligence, and outlined its Vision-Language-Action (VLA) technology development initiative. The Group also unveiled footage of an Atria AI-equipped SDV Testbed navigating complex urban traffic without driver intervention, operating at Level 2++ capability.
“Competitiveness is determined by how much data you secure, how quickly you learn and how effectively you can reflect those results in actual products and services,” said Minwoo Park, President and Head of the Advanced Vehicle Platform (AVP) Division at Hyundai Motor Group and CEO of 42dot. Park said the Group will develop autonomous driving technology that customers can trust based on a virtuous cycle of data, AI and validation.
Dual-Track Strategy Sets 2028 and 2029 Production Targets
The roadmap builds on the expanded strategic partnership with NVIDIA announced by Hyundai Motor Company and Kia Corporation on March 16, 2026. That agreement covers autonomous driving development from Level 2 through Level 4 on an integrated architecture built on the NVIDIA DRIVE Hyperion platform, a unified learning pipeline spanning real-world data collection, AI model training and deployment in production vehicles, and further discussions on advancing Level 4 robotaxi capabilities through Motional, the Group’s autonomous vehicle joint venture.
Under Track One, the Group will integrate NVIDIA’s vehicle AI computing platform and autonomous driving software into its software-defined vehicle (SDV) architecture. Production vehicles equipped with NVIDIA solutions-based Level 2+ autonomous driving capabilities are targeted for the first half of 2028, with Level 2++ production vehicles targeted for the second half of 2028. Sensor systems used across Hyundai Motor, Kia, 42dot and Motional will be progressively standardized around NVIDIA DRIVE Hyperion 10, a change the Group says will support more consistent data collection and utilization for AI training and validation.
Track Two centers on Atria AI, a proprietary end-to-end autonomous driving system jointly developed by the AVP Division and 42dot under an integrated development framework. Production of Atria AI-powered Level 2++ vehicles is targeted for the second half of 2029, with capabilities progressing in phases based on real-world driving data collected from production vehicles.
How the Data Flywheel Operates
The Data Flywheel operates as a cycle in which data collected from vehicles is used to train and validate AI models, with improved models then deployed back to vehicles to generate new data. The Group says Hyundai Motor and Kia sell more than 7 million vehicles annually across approximately 190 countries and regions, and that it currently operates approximately 40 dedicated data collection vehicles around the clock. Collected datasets include routine driving scenarios as well as edge cases: road construction zones and infrastructure variations, severe weather conditions, abrupt lane changes and emergency maneuvers, parked vehicles on side streets and narrow roads, and complex urban traffic dynamics.
Since earlier in 2026, the Group has integrated new learning technologies into the system. Hard Example Mining automatically identifies challenging driving situations that AI models find difficult to recognize or interpret and prioritizes those scenarios for training. A Continuous Training Pipeline incorporates newly acquired data from real-world driving and validation into model training, with vehicle evaluation findings fed back into data collection and model development. Virtual Validation Technology reconstructs real-world driving data into three-dimensional environments, using graphics techniques such as 3D Gaussian Splatting to recreate scenarios that are difficult or potentially unsafe to reproduce in real-world testing.
A Follow-the-Sun development model connects centers in South Korea and the United States, letting teams use time-zone differences to carry out data collection, issue analysis and model improvement across continuous 24-hour cycles. The Group is gradually integrating its Special Event Recorder, which automatically records and stores significant events during autonomous driving, into the Data Flywheel, primarily to support model training and performance improvement. It is also establishing a Data Union framework based on standardized sensor architectures and data structures, so that data generated across multiple vehicles and organizations can accumulate under common standards, initially across Hyundai Motor, Kia, 42dot and Motional.
Junghyun Kwon, Executive Vice President and Head of Hyundai Motor Group’s Autonomous Driving Development Center and 42dot Autonomous Driving Division Lead, said competitiveness depends less on data volume than on how rapidly data connects to learning, validation and performance improvement. Seonggyun Jeong, Group Lead of 42dot’s Atria Group, said the Group continuously improves the performance and maturity of Atria AI through an integrated development cycle spanning data collection, model training and real-world vehicle validation.
Gwangju Level 4 Pilot
In partnership with South Korea’s Ministry of Land, Infrastructure and Transport, the Group plans to deploy the Atria AI-equipped SDV Pace Car in Jeonnam-Gwangju Special Metropolitan City by the end of 2026. The pilot will operate autonomous vehicles on actual Korean roads with complex traffic dynamics and unpredictable variables, which the Group describes as conditions fundamentally different from controlled test tracks. It says driving scenarios and contingency situations captured during the deployment will be fed back into the Data Flywheel, enhancing both Level 2+ mass-production driver assistance technology and the validation of advanced Level 4 capabilities.
Seoul Driving Footage and VLA Development
The Group also released video footage of the Atria AI-equipped SDV Testbed navigating actual Seoul traffic without driver intervention, presented in three categories. An executive ride-along video shows Park and Jeong traveling through central Seoul across expressways, major thoroughfares, bridges and urban streets, discussing Atria AI’s development process, current technical capabilities and decision-making mechanisms during the ride. One-take sequences capture congested morning rush-hour traffic in Gangnam, high-density traffic in Jamsil with frequent interactions involving large vehicles such as buses, and rainy urban driving in Pangyo, each running approximately two to four minutes without edits other than playback speed adjustments.
A third video presents ten edge-case scenarios, including avoiding vehicles parked along the roadside, responding to sudden vehicle cut-ins, navigating unprotected left turns, detecting pedestrians in congested areas and identifying oncoming vehicles on narrow neighborhood roads. Park said development speed and safety are not conflicting values, and that the Group will apply only thoroughly validated technology to its vehicles.
Beyond its existing end-to-end models, 42dot is developing Vision-Language-Action models that combine visual information recognition, language-based reasoning and action generation in a single framework. 42dot describes VLA as a key technology for Physical AI applications, including autonomous driving and robotics, and says the added language-based reasoning enables improved decision-making and explainability in complex driving scenarios. It expects VLA to improve responses to rare driving situations by drawing on language-based reasoning and large-scale pre-trained knowledge that is difficult to learn from driving data alone. 42dot also released development footage showing how the VLA model interprets driving situations and outputs its reasoning in natural language during vehicle operation.
“VLA is a core technology for implementing Physical AI where AI goes beyond simply driving to understand situations, reason through them and act,” said HeeSeok Lee, Group Lead of 42dot’s Trion Group.
42dot’s VLA-based autonomous driving technology is currently in the simulation-based model validation stage, with the full development process, including real-vehicle testing, planned from late 2026 through early 2027.