现代汽车集团发布自动驾驶长期路线图：2028年起量产L2+/L2++，2029年推自研Atria平台，无近期业绩增量

The central idea behind the strategy Hyundai Motor Group laid out at its autonomous driving media day on the 11th is the "data flywheel." Driving data collected from mass-produced vehicles is used to train and validate artificial intelligence models, and the upgraded models are then deployed back into vehicles to gather fresh data. Park Min-woo, president and head of the Advanced Vehicle Platform Division at Hyundai Motor and Kia, said the competition in autonomous driving is no longer about which company has the better individual feature. "What determines a company's competitiveness is how much data it can secure, how fast it can learn from that data, and how quickly it can reflect the results in actual products and services," he said.
Hyundai Motor Group sells more than 7 million vehicles a year in about 190 countries. Once autonomous driving systems begin going into those cars in earnest, the group will be positioned to accumulate real-world driving data quickly. Jung Sung-kyun, Atria group lead at 42dot, predicted that once mass production of autonomous driving systems begins in earnest, the group will surpass the volume of data its rivals have accumulated within five years.
The group also filled in its commercialization roadmap. In the first half of 2028, Hyundai Motor Group plans to apply Level 2+ autonomous driving functions based on Nvidia solutions to mass-produced vehicles, and in the second half of that year it plans to release vehicles equipped with Level 2++. By the group's own calculation, it could exceed the cumulative driving data held by leading rivals such as Tesla in 2033, five years after full-scale mass production begins. Level 2+ is a driver-assistance technology that allows hands-off driving under driver supervision in specific conditions such as highways, while Level 2++ extends that to general roads and urban sections.
In the second half of 2029, the group will mass-produce Level 2++ vehicles built on Atria AI, its in-house autonomous driving platform, upgrading the technology in stages. Kwon Jung-hyun, executive vice president and head of the Autonomous Driving Development Center at Hyundai Motor and Kia and lead of the AD division at 42dot, said the company will implement a range of functions covering driving, parking and active safety at a high level of completion by 2027. "After that, 2028 should be seen as the period for gathering large volumes of data on edge cases — exceptional driving situations — to build a product that is genuinely safe, and 2029 as the period for preparing to obtain safety certifications not only in Korea but in North America, Europe and elsewhere," he said.
To support this, Hyundai Motor Group is currently running about 40 vehicles dedicated to data collection, operating them day and night. Since early this year, it has fully adopted four core technologies to improve the performance of its AI models. They include "hard example mining," which automatically selects driving situations that are difficult for models to learn; "continuous training," which feeds those cases immediately into repeated training to shorten development cycles; and a "special event recorder," which automatically logs sudden situations such as maneuvering around vehicles parked on the shoulder, hard acceleration and hard braking, as well as the moments when autonomous driving is disengaged. The group is also running a "data union" that consolidates the collection systems of Hyundai Motor and Kia, 42dot and Motional, its U.S. robotaxi joint venture, which had used differing sensor specifications, onto Nvidia Hyperion 10 to build a shared, integrated network.
To strengthen what serves as the brain of its autonomous driving models, the group is adding a vision-language-action model. VLA combines visual recognition, language-based reasoning and action generation into a single model, improving judgment and explainability in complex driving situations. The company said VLA is a core technology for realizing physical AI, in which AI goes beyond simply driving to understand a situation, reason about it and act. It added that the technology will serve as a foundation that can expand from autonomous driving into areas such as robotics.