Kakao Mobility, Nvidia target Level 4 Kia deployment

By Grace Kim

Two Nvidia DRIVE AGX Thor in-vehicle AI computing platforms will form the compute core of Kakao Mobility’s planned Level 4 autonomous-driving unit as the company prepares commercialization in South Korea. Level 4 denotes driverless operation within defined conditions.

Kakao Mobility said on the 9th that it was signing a memorandum of understanding with Nvidia at the Westin Seoul Parnas in Seoul’s Gangnam district. Kim Jin-kyu, executive vice president of Kakao Mobility’s physical AI division, and Gaurav Agarwal, senior director of Nvidia’s automotive business division, attended.

The agreement links three layers of the vehicle stack:

• Kakao Mobility’s perception and decision-making model.
• Nvidia’s onboard computing and safety technology.
• Kia’s purpose-built vehicle base.

Kakao’s July partnership with Kia already covered the vehicle side; the Nvidia memorandum adds the computing and sensor architecture.

What does the Nvidia agreement cover?

The companies plan to develop a Level 4 system that meets global technical standards, combining Kakao Mobility’s in-house autonomous-driving AI model with Nvidia’s computing infrastructure and optimization technology.

Kakao is developing an end-to-end model (one AI model covering both surroundings recognition and driving decisions). The stated aims are stronger computing performance and system stability.

The planned computing unit carries two DRIVE AGX Thor platforms. The companies will also use Hyperion, Nvidia’s sensor and computing reference architecture (a defined sensor-and-computer layout), to optimize connections among the vehicle, its sensors and the computing unit.

The goal is system stability and compatibility across those interfaces. Nvidia’s integrated safety system, Halos, will apply to autonomous-driving software development and validation, with its framework extending from development through real-world service deployment.

Global technical standards sit at the center of the joint target because Kakao is preparing domestic commercialization while keeping overseas markets under review for the medium to long term.

Where will the technology deploy first?

Kia comes first. The combined technologies will reach Kia (000270.KS) vehicles before other platforms. Kakao Mobility signed a partnership with Kia in July to develop purpose-built vehicles for autonomous-driving services.

Under the Nvidia agreement, Kakao plans to integrate its software and Nvidia’s AI computing and safety technologies into those vehicles, then prepare vehicle development and supply for Level 4 services.

Kakao has tested autonomous-driving technology in Gangnam city center and is preparing for domestic commercialization. It will also review possible overseas market entry over the medium to long term.

“Kakao Mobility has internalized Level 4 technology and worked with leading companies at home and abroad on the basis of three pillars — our own autonomous driving technology, a mobility platform and a technology cooperation ecosystem,” Kakao Mobility CEO Ryu Gung-sun said.

“Through this partnership, we will prepare for the era of Level 4 commercialization as a focal point of a global technology ecosystem that connects software, components and finished vehicles,” Ryu said.

Nvidia vice president of automotive Rishi Dhall said: “Nvidia Hyperion provides the sensor and computing foundation needed to realize Level 4 autonomous driving, and Halos supports a comprehensive safety framework from development through to real-world service deployment.”

He added: “We expect this cooperation to contribute to the development of safe and scalable autonomous driving services in Korea.”

What changes for driverless services?

For driverless services, the change is a move from Gangnam testing toward a defined three-company integration. Kakao brings the driving model; Nvidia brings compute, sensors and safety architecture; Kia is the vehicle partner. The immediate commercial milestone remains preparation for South Korean deployment; the announcement gives no launch date, fleet volume or operating-area boundary.