新一轮智驾PK，迈入实战时刻

Core Viewpoint - The article discusses the recent surge in advancements in intelligent driving (智驾) technology in China, highlighting the competitive landscape and the emergence of new models that significantly enhance driving capabilities [4][5]. Group 1: Industry Dynamics - The recent actions in the intelligent driving sector are driven by several key factors, including the launch of new models by leading companies such as Li Auto, Momenta, Yuanrong, and Xpeng, all of which are introducing their versions of the VLA (Visual Language Action) model [5][10]. - The VLA model is positioned as a significant upgrade over previous end-to-end models, with its current version reportedly exceeding the capabilities of earlier models expected in mid-2024 [6][10]. - The shift from CNN-based models to Transformer-based models represents a fundamental change in the underlying technology, enhancing the model's ability to mimic human cognitive processes [7][12]. Group 2: Technical Advancements - The VLA model integrates perception, reasoning, and action execution into a single framework, moving away from the segmented approach of earlier models [7][9]. - The introduction of Chain of Thought (CoT) capabilities allows the VLA model to break down complex driving decisions into logical steps, improving decision-making reliability and transparency [11][15]. - The model's ability to learn from vast amounts of data enhances its generalization capabilities, enabling it to handle previously unseen driving scenarios effectively [15][22]. Group 3: Future Outlook - The VLA model is still in development, with ongoing efforts to enhance its CoT capabilities and overall performance [24]. - Key future tasks include collecting more data from testing fleets and refining the model based on technological trends [24][25]. - The cost of implementing VLA technology is primarily influenced by chip requirements, with higher-end models being more compatible, while lower-cost models may also have opportunities for adaptation [26].