NVIDIA pitches its robotaxi platform to Korea's pilot program as it grows from 200 cars to 3,000
NVIDIA has proposed to South Korea's transport ministry that its Hyperion platform and Halos safety system run in the government's driverless robotaxi pilot, Seoul Economic Daily reported. The pilot is due to add four to five cities in 2027, which makes it a place to become the default before Korea's 2028 commercial start.
NVIDIA has proposed to South Korea's Ministry of Land, Infrastructure and Transport that its Hyperion autonomous-driving platform and its Halos safety system be used in the government's driverless robotaxi pilot, Seoul Economic Daily reported on Nov. 8, citing the IT industry. Hyperion bundles an in-vehicle AI computer, cameras, radar and software; Halos is meant to catch system faults and validate safety. NVIDIA has also sounded out Korean autonomous-driving companies about working together.
The timing explains the pitch. The ministry plans to add four to five pilot cities in 2027 and grow the pilot fleet from 200 vehicles to 3,000, and it aims to run Level 4 driverless robotaxi services from 2028. A platform that goes into the pilot cars could become the common base on which Korean companies develop and validate their systems before the paid market opens.
The first city already has its players
The first pilot region is the Jeonnam-Gwangju Integrated Special City. If the plan is adopted there, NVIDIA's technology would go into the Ioniq 5-based autonomous vehicles that Hyundai supplies, and RideFlux and Autonomous A2Z, chosen for the Gwangju pilot in April, would run their own software on NVIDIA's platform, the paper wrote. NVIDIA has also approached Socar, which holds large sets of Korean driving and accident data. A meeting between NVIDIA senior director Madison Huang and Kakao Mobility CEO Ryu Keung-sun set for Nov. 9 was postponed for reasons on NVIDIA's side.
Newsletter
Robrief Weekly
One email a week with the robotics news that matters. Free, unsubscribe anytime.
The companies that built their own stacks are split. "To use NVIDIA's platform we would need extra work to retune our existing software to run inside it, and we haven't reached a conclusion on what exactly we gain," an executive at one company that received the offer told the paper. Another company said it is reviewing the offer favorably because NVIDIA is "very likely to become a major technology standard" in global autonomous driving.
NVIDIA's own plan with Uber, announced in March, is to take driverless robotaxi service to 28 cities worldwide by 2028, starting in Los Angeles and San Francisco next year.
Korea's rules come with a catch. Under the deregulation plan the government took up on Oct. 23, a fast-track permit is meant to put more than 3,000 vehicles into pilot cities from next year, and data collected by autonomous vehicles is to be stored in Korea by default, with transfers abroad needing separate permission, Electronic Times reported. NVIDIA would get Korean road miles; taking the data out of the country would take another approval.
Sources Sources verified
- #nvidia
- #hyperion
- #robotaxi
- #south-korea
- #gwangju
- #molit
- #hyundai
- #rideflux
- #autonomous-a2z
Share this briefing
Know someone who follows robotics? Send it their way.
Related briefings
More in Autonomous Driving & LogisticsPony.ai, inDrive and NVIDIA sign an Astana robotaxi MOU, with tests targeted for the first half of 2027
The MOU covers Level 4-ready robotaxis run through inDrive's app, with Pony.ai supplying the vehicles and NVIDIA the Hyperion platform. inDrive's CTO says testing with a safety operator would begin in the first half of 2027; the launch depends on a Kazakh government agreement.
Uber plans 2,000 driverless cars by end-2027 and pitches AV partners 30% more trips per car
Uber expects about 700 fully driverless vehicles on its app by year-end and more than 2,000 by the end of 2027, half of them outside the US. What it is selling partners is utilization: about 30% more trips per vehicle per day than first-party deployments.
Baidu's Apollo World world model trains an end-to-end robotaxi driver that has 1 million driverless km
Baidu showed Apollo World, a world model in which every road user is an AI agent, in Shenzhen on Nov. 3 and says Apollo Go already uses it for testing. The end-to-end driver it trains has about 0.4% of the fleet's 240 million driverless kilometres.
Newsletter
Robrief Weekly
One email a week with the robotics news that matters. Free, unsubscribe anytime.