이미지: X — 뉴스 앰프
Summary
- Liu Xiaoming, head of the General Intelligence Center at XPeng Global Group, gave an exclusive interview on VLA 2.0
- The interview reportedly covered where autonomous driving still fails, when Europe entry might happen, and how far off robotaxi commercialization remains
- The detailed answers were released via a YouTube video, and were not included in the excerpted tweet
- 인터뷰 대상
- 리우 시아오밍(Liu Xianming), XPeng Global Group 제너럴 인텔리전스 센터장
- 공개 채널
- X 계정 Chubby(뉴스 앰프) 및 유튜브
- 다룬 주제
- VLA 2.0 실제 성능, 자율주행 실패 지점, 유럽 진출 시점, 로보택시 상용화 거리
- 발행일
- 2026-08-11
What was released
News aggregator account Chubby posted an exclusive interview with Liu Xiaoming, head of the General Intelligence Center at XPeng Global Group, on X. The interview reportedly asked what XPeng's VLA 2.0 technology, currently under development, can actually do, where autonomous driving still fails, when the technology could reach Europe, and how far off full deployment of fully autonomous vehicles remains. However, the posted text only included the list of questions, and the detailed answers appear to be accessible only through a separate linked YouTube video.
What VLA is and why it's hard
VLA (Vision-Language-Action) is an approach that connects camera-captured scenes and linguistic reasoning within a single model, translating them directly into actions such as steering, acceleration, and braking. Whereas conventional autonomous driving software split perception, judgment, and control into separate modules, VLA is closer to a method that carries over the commonsense reasoning ability learned by large language models directly into driving decisions. Tesla's FSD has also evolved in a similar direction, and among Chinese EV makers, there is growing recognition of this approach as the next stage of autonomous driving. XPeng started out as an EV manufacturer but has expanded into its own autonomous driving stack, humanoid robots, and urban air mobility (eVTOL). The very name of the organization discussed in this interview, the "General Intelligence Center," signals a direction of treating autonomous driving as one pillar within physical AI as a whole.
Much like NVIDIA's DGX Spark multi-unit connection guide released on August 5, the physical AI industry has recently been focused on running large models locally and at the edge in real time. Running VLA models inside autonomous vehicles without latency is a challenge in the same vein.
So what changes
This interview appears to be the first time XPeng has publicly addressed VLA 2.0's actual performance and its Europe rollout roadmap in the form of a direct verbal response. Since opportunities to hear about autonomous driving technology directly from a development lead, rather than through standardized benchmarks, are rare, once the YouTube video is released it is expected to serve as a reference point for gauging the technical maturity of China's autonomous driving camp and the timing of its entry into the European market.
