Hyundai Motor Group has pushed back the launch of its first vehicle carrying a self-developed driver-assist system to the end of 2029, two years later than originally planned. In the meantime, it is co-developing systems with Nvidia, rolling out L2+ and L2++ systems in 2028 first, with the resulting data feeding back to train its own software platform, Atria. Reuters was first to report the change on September 13.
The sensor plan has also been finalized: cameras, ultrasonic sensors and millimeter-wave radar, with no lidar in the initial lineup. That puts Hyundai in the minority within the industry right now, as mid-to-high-end models in the Chinese market have been piling on lidar for the past two years.
From 2027 to 2029
Lining up the milestones, Hyundai's original plan called for its self-developed system to debut at the end of 2027, going to market directly with a full in-house stack. When that timeline was set, the industry was still relatively optimistic about how hard it would be to mass-produce end-to-end systems.
The new plan splits the rollout into two stages: Nvidia's Hyperion 10 platform delivers L2+ and L2++ in 2028, and the self-developed system doesn't arrive until the end of 2029. The two years in between aren't dead time — real-world driving data from the production fleet becomes fuel for training Atria.
It's a trade of delivery pace for data accumulation. Hyundai executives have said publicly that the company is "deliberately slowing down" its autonomous driving timeline, betting that a gradual approach can overtake Tesla's FSD within five years. That five-year framing doesn't come with interim checkpoints, nor does it specify what metric would count as "overtaking."
On the relationship with Nvidia, Hyundai's position was stated plainly:
Working with Nvidia doesn't mean handing our fate entirely over to them.
That line responds to outside concern about Hyundai becoming a contract assembler. When an automaker hands perception and decision-making over to a chip supplier's platform, it saves development time in the short run but risks losing leverage in that part of the value chain over time. Hyundai's answer is to keep the data for itself.
The trade-off of going without lidar
On sensors, Hyundai executives stressed the need for redundancy, saying that even if one sensor fails, the rest of the system must still be able to keep the vehicle running safely. A camera-radar-ultrasonic combination can achieve redundancy structurally, but at night, in glare, or in rain and fog, its performance ceiling isn't in the same league as a lidar-equipped system.
Hyundai's stated reason is cost. Lidar prices have dropped considerably in recent years, but for a product line sold across multiple global markets at several million units a year, every $100 of per-vehicle cost gets multiplied by a very large number.
A commercialization timeline for L3 hasn't been announced. What separates L2++ from L3 isn't sensor count — it's liability. In an L3 accident, the automaker is on the hook, and crossing that line requires regulation, insurance and product maturity to line up at the same time. This announcement didn't touch that boundary.
Where this decision sits in the industry
At the same time, Tesla's FSD keeps iterating on pure vision and a massive fleet's worth of data, while Chinese automakers are fighting it out on L2++, with lidar and urban navigate-on-autopilot becoming near-standard on new cars. Hyundai's path is neither Tesla's pure-vision bet nor its Chinese rivals' hardware-stacking approach — it looks more like using a supplier's platform to hold its market position first, then using the data to close the gap on its own technology.
Two things need to hold for this path to work: Nvidia's L2++ platform needs enough real consumer uptake and install volume for the returning data to be worth anything, and the Atria team needs to get its models to a production-ready state within two years. If either one falls behind, the 2029 target could slip again.
By the time the first batch of Nvidia-based vehicles ships in 2028, install volume and real-world takeover rates will be the earliest evidence available.
Sources: Reuters, CocoLoop, Nvidia's official blog, IT Home; timelines and sensor plans follow statements from Hyundai Motor Group and Nvidia's announcements.