Hyundai Motor Group said it expects to overtake Tesla and other rivals on autonomous driving data within five years of putting its own system into mass production, a target it set on Friday while pushing that production date back by two years.
“Hyundai sells 7 million vehicles a year across 190 markets, which means we can accumulate data at scale and at this pace, we can overtake what our rivals have built in five years,” said Park Min-woo, President of Hyundai’s Advanced Vehicle Platform Division and CEO of 42dot, the group’s self-driving subsidiary.
The claim, made at a briefing in Pangyo, south of Seoul, rests on a timeline that has moved.
Vehicles running Atria AI, the system Hyundai and 42dot are building themselves, are now targeted for the second half of 2029, two years later than the late-2027 date the group had been working to.
Until then it will use Nvidia’s software, with Level 2+ cars targeted for the first half of 2028 and Level 2++ for the second half.
“Our partnership with Nvidia is not about leaving our destiny entirely in their hands,” Park said. He said Hyundai would co-design the technology with Nvidia and use the data it generates to train Atria.
Park did not say when the five years start, and the two tracks give two answers. Counted from the Nvidia cars the target year is 2033. Counted from Atria it is 2034.
Nor is an annual sales figure a rate of data collection.
Hyundai currently runs about 40 dedicated data-collection vehicles, according to Korea JoongAng Daily.
Tesla’s comparable figure counts miles driven with Full Self-Driving engaged rather than all miles its cars cover.
That total passed 10 billion on May 3, 13 billion on August 3 and 14 billion on August 27, according to Tesla’s own running counter and independent trackers, which are not audited disclosures.
The last of those billions took 24 days. Earlier this year the fleet added a billion miles in the first 50 days.
Almost all of that is supervised driving with a person responsible.
Tesla’s unsupervised robotaxi fleet had covered 1 million miles as of September 3, when Ashok Elluswamy, its VP of AI, gave the figure at the Cybercab event in Austin, up from 380,000 in late July.
Seoul Economic Daily quoted a company official saying Hyundai would overtake competitors’ data volumes within five years once mass production begins, which is a claim about the size of a dataset rather than the capability of a product.
A Delay Presented as a Choice
Park described the slip as deliberate.
“We believe Atria AI could move faster if we pushed with everything we have,” he said, according to Korea JoongAng Daily. “But we made a strategic decision to slow down. Rushing into mass production now risks delivering a middling Level 2++ system, far short of our real target.”
Park added that “the competition in autonomous driving is no longer about the speed. It comes down to how much data you can gather, how fast you can learn from it and how quickly that translates into real products and services.”
The group set out the plan at its first autonomous driving media day, held at 42dot’s headquarters.
Reuters reported the delay as evidence of the difficulty Hyundai has had building the software in-house, and said it had raised concerns about the group’s deepening reliance on Nvidia.
Level 2+ and Level 2++ are industry terms, not standards. SAE International defines six levels of driving automation from 0 to 5, and both Tesla’s FSD and the systems Hyundai described on Friday sit at Level 2, where the driver remains responsible.
The Man Making the Claim
Park took over the AVP division and 42dot on January 13. Before joining Hyundai he was a vice president at Nvidia working on autonomous driving software, and before that he worked on Tesla Vision, the camera-based perception system Tesla’s driver-assistance software is built on.
He replaced Song Chang-hyeon, who founded 42dot in 2019, sold it to Hyundai in 2022 and ran the AVP division until he resigned on December 3 after a meeting with Executive Chair Euisun Chung.
Song was widely reported to be taking responsibility for the group’s limited progress on autonomy. In a farewell message to staff he said it had not been easy to aim to build an AI device rather than a car, or to implant software DNA in a hardware business.
The group’s target at that point was Level 2+ by 2027.
Three days after Park’s appointment, Hyundai said it would bring in Milan Kovac, the former head of Tesla’s Optimus humanoid programme, as an adviser.
What the Flywheel Is
Hyundai calls its approach a Data Flywheel, on the principle that more cars produce more data, which improves the software, which improves the cars.
Two techniques do the work. Hard Example Mining automatically flags driving situations the model handles badly, such as construction zones, severe weather, abrupt lane changes and haphazard parking, and a continuous training pipeline feeds them back into the model.
A Special Event Recorder logs data when it detects sharp acceleration or braking, erratic following distances, or a moment when the driver or the system disengages.
The group is also standardising sensors across Hyundai, Kia, 42dot and Motional on Nvidia’s DRIVE Hyperion 10 platform, so that data gathered by any of them can train the same models.
The Nvidia partnership was expanded on March 16 to cover Level 2 through Level 4.
42dot said it has begun work on a vision-language-action system, which combines visual perception, language-based reasoning and vehicle control. In footage shown at the briefing, on-screen text explained the reasoning behind a lane change as the car made it.
“Through 2027, we’ll focus on bringing driving, parking and active-safety features to a high level of polish at the product level,” said Kwon Jung-hyun, who leads 42dot’s autonomous driving division. “2028 will be devoted to gathering vast amounts of edge-case data to build a genuinely safe product.
And 2029 is when we’ll prepare for safety certification, not just in Korea but in North America and Europe as well.”
Hyundai also said it will run a Level 4 pilot in Gwangju by the end of this year with Korea’s Ministry of Land, Infrastructure and Transport.
