Causal Reasoning in Autonomous Driving: Overcoming the Limitations of E2E
Waymo, Nvidia, and Xpeng Join the Race for Development

Hyundai Motor Group is accelerating its efforts to secure a Vision-Language-Action (VLA) model to advance autonomous driving technology. The group's strategy is to move beyond driving performance-focused technology and shift towards 'intelligent autonomous driving,' where vehicles can understand situations and explain the basis for their decisions.


According to industry sources on March 31, Minwoo Park, Head of the Advanced Vehicle Platform (AVP) Headquarters at Hyundai Motor Company and CEO of Forty2Dot, recently posted an open recruitment announcement for VLA model researchers on his LinkedIn. Observers note that it is unusual for Forty2Dot, since Park became CEO, to specifically highlight the recruitment of VLA model developers among experienced hires.


Minwoo Park, AVP (Advanced Vehicle Platform) Head at Hyundai Motor Group and CEO of FortyTwoDot, is giving a presentation at the AVP Headquarters Town Hall Meeting held on the 5th at Techwon in Pangyo, Seongnam-si, Gyeonggi-do. Hyundai Motor Group

Minwoo Park, AVP (Advanced Vehicle Platform) Head at Hyundai Motor Group and CEO of FortyTwoDot, is giving a presentation at the AVP Headquarters Town Hall Meeting held on the 5th at Techwon in Pangyo, Seongnam-si, Gyeonggi-do. Hyundai Motor Group

View original image

He emphasized the necessity of securing the VLA model, stating, "Our goal is to convert perception, reasoning, and language understanding into reliable actions in the real world."


The VLA model, aimed at enhancing the completeness of autonomous driving technology, is drawing significant attention in the automotive industry. This reflects a growing awareness that autonomous vehicles must evolve beyond simple driving capabilities and reach a level where they can 'understand and explain' situations on their own.


VLA stands for Vision-Language-Action, referring to a model in which AI perceives situations visually, understands them linguistically, and uses logical reasoning to take appropriate actions. When applied to vehicles, VLA models enable causal reasoning, such as instructing the car to "change lanes due to road construction ahead" rather than simply "stop for an obstacle," positioning this technology as a 'game changer' in the autonomous driving field.


Until now, autonomous driving technology has evolved from the traditional 'rule-based' approach to a data-driven 'End to End (E2E)' method. While E2E delivers excellent driving performance based on massive data, it has faced criticism for the 'black box' problem, where it is difficult to explain the reasoning behind driving decisions.


By incorporating language-based reasoning into the decision-making process, VLA is expected to compensate for the shortcomings of E2E by enabling explanations of the grounds for each decision.


Forty2Dot CEO Highlights 'VLA' as Hyundai Motor Group Accelerates Intelligent Autonomous Driving View original image

Tesla, Waymo (Google's autonomous driving subsidiary), China's leading automaker Xpeng, and even Nvidia are all developing integrated stacks that combine the physical driving capabilities of E2E with the logical reasoning abilities of VLA.


Waymo is operating robotaxis by applying Google DeepMind's VLA architecture to its autonomous driving system. In January, Nvidia unveiled its autonomous driving VLA model ‘Alphayo.’ This technology is set to be installed in the next-generation Mercedes-Benz S-Class, which will begin 'Level 2++' urban driving in the United States in the second half of this year.


On March 11 (local time), Xpeng announced it would begin demonstration applications and test drives of its proprietary autonomous E2E solution 'VLA 2.0' at 732 stores across China. Xpeng aims to roll out its autonomous driving solution to global markets next year, and Volkswagen has decided to adopt Xpeng's VLA 2.0 solution.



Currently, Forty2Dot is working on advancing its E2E autonomous driving technology 'Atria AI.' The plan is to integrate the VLA model into Atria AI, transforming it into a next-generation intelligent AI capable of logical reasoning and explanation in driving situations. Consequently, Hyundai Motor Group is expected to establish an autonomous driving collaboration system involving Forty2Dot, Nvidia, and Motional.


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

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing