Mei Ri Jing Ji Xin Wen·2025-12-16 12:21Core Insights - Xiaopeng Motors is set to release its VLA 2.0 (Vision Language Action) model in the next quarter, with significant pressure on the team as it is the first version [1] - A bet was placed between Xiaopeng's chairman and the autonomous driving team, aiming to match Tesla's FSD V14.2 performance by August 30, 2026, or face a challenge [1] - The autonomous driving sector is experiencing a paradigm shift, moving from traditional sensor-based systems to AI-driven models [1][2] Group 1: VLA Model Overview - The VLA model is seen as an "intelligent enhanced version" of end-to-end solutions, integrating visual perception, action execution, and language modeling [3] - It aims to overcome the black-box issue of traditional models by incorporating a reasoning chain through language models, enhancing interpretability and adaptability to complex environments [3][4] - The model's architecture allows for better integration of vast knowledge bases, improving its generalization capabilities [3] Group 2: Industry Perspectives - There is a divergence in the industry regarding the VLA and world model approaches, with companies like Li Auto and Xiaopeng favoring the VLA model [2] - Critics, such as Wang Xingxing, express skepticism about the VLA model's effectiveness in real-world interactions due to data quality concerns [4] - Li Auto emphasizes the importance of real data in developing effective autonomous driving systems, arguing that the VLA model's success relies on a robust data loop [4] Group 3: Technological Integration - The world model approach focuses on creating an internal simulation of the physical world, enabling better prediction and decision-making capabilities [5] - Companies like NIO and SenseTime are also exploring the world model technology, indicating a broader industry trend [5] - Despite differing opinions, there is a trend towards integrating VLA and world model technologies, with both approaches potentially complementing each other [6] Group 4: Future Directions - Xiaopeng is moving towards a hybrid approach, aiming to combine VLA and world model technologies, as indicated by the recent updates to their VLA model [7] - The second generation of the VLA model aims to reduce information loss by streamlining the process from visual input to action execution [7] - The industry is witnessing a shift where companies are choosing different paths based on their specific goals, whether it be selling vehicles or developing autonomous taxi services [7]