Hu Xiu·2025-10-20 07:22Core Insights - The automatic driving industry is experiencing a battle for narrative control over next-generation technologies, with companies like Li Auto and XPeng betting on VLA (Visual Language Action) as the future architecture, while Huawei promotes its self-developed WA (World Behavior Architecture) [1][2][20] - The rapid emergence of jargon in the industry reflects the struggle for technological branding and user perception, as hardware and configurations become increasingly homogeneous [1][2][27] Group 1: Evolution of Technology - Before 2022, the evolution of automatic driving technology was primarily defined by Tesla and Waymo, with terminology focused on objective descriptions of specific functions [3] - Tesla's introduction of the BEV+Transformer architecture in 2021 marked a significant shift from rule-based systems to AI-driven approaches, enhancing perception capabilities [4][5][6] - The transition to an end-to-end paradigm was catalyzed by Tesla's AI DAY in 2022, which integrated perception and planning into a single neural network, significantly improving obstacle recognition [9][10] Group 2: Adoption of New Models - Chinese automakers quickly adopted Tesla's technology, with companies like XPeng and NIO implementing their own versions of the BEV+Transformer model for mass production [8][10] - The industry is moving towards end-to-end systems, with XPeng and Huawei initially adopting a multi-stage approach for safety reasons, before transitioning to fully integrated models [10][12] - The introduction of VLA and world models into automatic driving systems represents a new frontier, with companies like Yuanrong Qixing and NIO leading the charge in applying these concepts [17][20] Group 3: Competitive Landscape - The competition among companies is not only about technology but also about computational power, with XPeng and Li Auto investing heavily in cloud computing capabilities, boasting figures of 10 EFlops and over 13 EFlops respectively [18][19][55] - The race for computational resources extends to both vehicle and cloud platforms, with Tesla's Dojo and other companies ramping up their AI training capabilities [18][57] - The rapid evolution of VLA and world models is indicative of a broader trend where companies are leveraging advanced AI techniques to enhance their automatic driving systems [20][46] Group 4: Future Directions - The world model concept, initially used for simulation, is now being applied in real-time vehicle control by companies like NIO and Huawei, aiming for more predictive and human-like driving experiences [20][24][25] - The emergence of terms like VLA and world models highlights the industry's shift towards integrating language understanding and real-time decision-making into automatic driving systems [46][59] - The ultimate success in this competitive landscape may hinge on a company's ability to translate technological promises into tangible user experiences, rather than merely marketing jargon [30][29]