[Why&Next] 'Atria AI' with 400 Trillion Calculations per Second Races to Catch Up with Tesla's FSD

Autonomous Driving Demonstration in Gwangju Begins This Year

Perceiving, Deciding, and Controlling Road Conditions with Cameras and Radar

Chasing Tesla with a Focus on Efficiency

Hyundai Motor Group's autonomous driving technology is entering a full-scale verification phase on actual roads.

The next-generation autonomous driving system, "Atria AI," developed by FortyTwoDot (42dot), Hyundai Motor Group's autonomous driving solution developer, will accumulate large-scale driving data through a demonstration project in Gwangju Metropolitan City in the second half of this year.

The Ministry of Land, Infrastructure and Transport, Gwangju Metropolitan City, and Hyundai Motor Group held the "Korea Autonomous Driving Team Business Agreement Ceremony" on the 13th at the Kim Dae-jung Convention Center in Gwangju, officially announcing its launch. Ministry of Land, Infrastructure and Transport

The company's strategy is to pursue the efficiency of Tesla, the global leader in autonomous driving, while aiming to surpass Tesla by achieving fully unmanned Level 4 autonomous driving. There is growing interest in how closely Hyundai is approaching Tesla’s camera-based autonomous driving model and what differentiated strategies it is preparing.

Hyundai Motor will launch an autonomous driving demonstration project in Gwangju Metropolitan City. Two hundred autonomous vehicles equipped with FortyTwoDot’s core autonomous driving solution, "Atria AI," will be deployed for this demonstration. The main feature of Atria AI is its thorough efficiency.

It boldly excluded expensive and difficult-to-maintain equipment such as high-definition maps (HD maps) and ultra-precise GPS, which have been essential in existing autonomous driving solutions. Instead, it perceives road conditions using only "eight cameras and one forward radar," mimicking the way humans see and drive.

Furthermore, it has adopted an end-to-end (E2E) approach, in which the entire process from visual recognition to decision-making and control is handled by a single AI model. This is the same approach as Tesla.

To process all this data in real time, the vehicles are equipped with a "400 TOPS-class NPU (Neural Processing Unit)" capable of 400 trillion operations per second. Like the human brain, it swiftly analyzes information collected by cameras and sensors and controls the movement of the vehicle.

The hardware architecture is also similar to Tesla’s. FortyTwoDot has implemented a "zonal architecture," in which a vehicle is divided into multiple zones, with each zone processing data first before sending it to a central computer (HPVC). By processing data by zone instead of all at once, they have improved efficiency. As a result, the number of controllers installed throughout the vehicle has been reduced by 66% compared to previous systems.

They have also adopted a 48-volt (V) high-voltage system and gigabit Ethernet technology, both of which are used in the Tesla Cybertruck. This simplifies the complicated wiring structure, reduces vehicle weight and manufacturing costs, and enhances the stability of power supply.

Some experts have raised concerns that the E2E approach, which relies primarily on cameras rather than lidar or high-definition maps, may not guarantee complete safety during rain, at night, or in unpredictable situations. Another challenge to overcome is that, as a latecomer, Hyundai has accumulated far less driving data than Tesla.

FortyTwoDot is training its autonomous driving AI using a data fleet operated jointly with Hyundai Motor and more than 2,000 high-performance GPU supercomputers. Large amounts of actual road driving data are collected to improve the vehicle’s ability to make autonomous decisions.

Tesla’s Full Self-Driving (FSD) system is a Level 2+ system, requiring drivers to keep their eyes on the road at all times. In contrast, FortyTwoDot aims to realize fully unmanned Level 4 robotaxis that operate without a driver.

Park Minwoo, CEO of FortyTwoDot, presenting at the Hyundai Motor Advanced Vehicle Platform (AVP) Division Town Hall Meeting. Hyundai Motor

Looking ahead, FortyTwoDot plans to apply a Vision-Language-Action (VLA) model. The goal is to advance AI beyond simply perceiving the road, enabling it to understand situations, make its own decisions, and act autonomously.

For this demonstration project in Gwangju, 200 Ioniq 5-based robotaxis integrated with Hyundai Motor’s mobility platform "Shucle" will be deployed. The industry sees this demonstration as both a test of FortyTwoDot’s technological maturity and a chance to assess whether it can catch up with Tesla.

An autonomous driving industry official said, "Currently, Tesla leads in camera-based end-to-end (E2E) technology. To be competitive, it is crucial to quickly follow this approach and then differentiate with advanced technologies such as VLA."

This content was produced with the assistance of AI translation services.