凤凰网科技讯 11月21日，小米汽车在今日广州车展活动中正式对外发布了小米HAD增强版，并披露了 其在智能驾驶领域的最新研发进展与人才布局。小米汽车方面表示，公司在AI领域的战略投入持续加 码，2025年仅AI研发投入预算就将超过70亿元，目前的辅助驾驶专家团队规模已达1800人，其中包含 108名博士。 在技术架构层面，此次发布的小米HAD增强版不仅基于原有的1000万clips训练基础，更核心的变化在 于引入了强化学习算法与世界模型，试图通过"端到端"技术路径提升驾驶表现。据介绍，通过世界模 型，系统能够在数字空间中生成包括极端天气（如大雾、大雪）、复杂路况及突发碰撞等在内的多种场 景，利用奖励函数机制训练算法，使其从"规则驱动"转向"学习驱动"。官方数据显示，该世界模型技术 已获得ICCV和NeurIPS等国际学术会议的认可。 本次发布的智驾更新将包含在小米HyperOS 1.11.0版本中，由于审核的进度差异，不同车型的推送时间 可能会略有不同，官方将全力推进，尽早的将版本推送给大家。 针对用户实际驾驶痛点，新版本重点优化了纵向与横向的控制体验。在应对旁车加塞场景时，系统通过 大模型预测并线意图，减少 ...

Core Viewpoint - The article discusses the escalating rivalry between two autonomous driving companies, Xiaoma Zhixing and Wenyuan Zhixing, as they prepare for their upcoming listings on the Hong Kong stock exchange. The conflict centers around claims of data scale and technological superiority, which are critical for valuation in the autonomous driving industry [6][7][9]. Group 1: Company Rivalry - Xiaoma Zhixing recently presented a comparison in a roadshow that labeled Wenyuan Zhixing's operational city as "Beijing" and its order volume as "zero," prompting a strong rebuttal from Wenyuan's CFO, Li Xuan, who accused Xiaoma of spreading false information [6][9]. - Both companies are competing for market share in the autonomous driving sector, particularly focusing on Robotaxi commercialization and L2+ advanced driving assistance systems [12][14]. Group 2: Key Metrics and Data - Xiaoma Zhixing reported a total driving mileage of 48.6 million kilometers, while Wenyuan Zhixing's latest figures show over 55 million kilometers, indicating that both companies are closely matched in terms of operational scale [9][14]. - Financially, Xiaoma Zhixing incurred a net loss of 681 million yuan in the first half of 2025, a year-on-year increase of approximately 75.07%, while Wenyuan Zhixing's loss was 792 million yuan, a decrease of 10.32% [14]. Group 3: Technological Focus - The debate over technological paths centers on the "end-to-end" solution, which has gained traction in the industry as the next-generation approach to autonomous driving. Wenyuan Zhixing claims to have achieved mass production with its "end-to-end" solution, while Xiaoma Zhixing's claims are questioned [10][12]. - The article highlights the importance of total driving mileage as a key metric for data quality in autonomous driving training, emphasizing that more mileage leads to richer data for algorithm development [9][10]. Group 4: Market Dynamics - The article notes that both companies are facing challenges in the commercialization of Robotaxi services, with regulatory policies acting as a significant barrier to rapid growth [12][13]. - Despite their similar financial struggles, Xiaoma Zhixing has a market capitalization of approximately $7.08 billion, while Wenyuan Zhixing stands at about $3.41 billion, indicating a disparity in investor confidence [14].

Core Viewpoint - The competition between autonomous driving companies Xiaoma Zhixing and Wenyuan Zhixing has escalated into a public dispute over their respective capabilities and technologies as they prepare for their upcoming IPOs in Hong Kong, with both companies aiming to secure significant capital through their listings [1][2][3]. Group 1: Company Comparison - Xiaoma Zhixing and Wenyuan Zhixing are both vying for market share and technological superiority in the autonomous driving sector, focusing on data scale and technological pathways as key competitive factors [3][5]. - Xiaoma Zhixing's prospectus indicates a total driving mileage of 48.6 million kilometers, while Wenyuan Zhixing's latest figures show over 55 million kilometers, highlighting their comparable capabilities in data collection [4][5]. - The dispute centers around the validity of their respective technological approaches, with Wenyuan Zhixing asserting its "end-to-end" solution in collaboration with Bosch and Chery, while challenging Xiaoma Zhixing's claims regarding its technology [5][6]. Group 2: Financial Performance and Market Position - Xiaoma Zhixing reported a net loss of 681 million yuan in the first half of 2025, a year-on-year increase of approximately 75.07%, while Wenyuan Zhixing's loss was 792 million yuan, a decrease of 10.32% [12]. - As of the latest reports, Xiaoma Zhixing's market capitalization stands at approximately $7.08 billion, compared to Wenyuan Zhixing's $3.41 billion, despite Wenyuan Zhixing having a higher gross margin [12]. - Xiaoma Zhixing plans to raise around 6.71 billion HKD (approximately 864 million USD) through its IPO, focusing on scaling operations and enhancing research and development [12][13].

摘要：路演PPT成首发子弹，CFO深夜檄文回击，全方位的正面交锋，只为争夺一张宝贵的资本门票。若全额行使超额配售，二者此次募 资至多合计超90亿元。 科技 出品 作者｜林苑 编辑｜赵子坤 公开活动远程互怼已经是"车圈传统"了。 前有何小鹏与余承东就AEB系统隔空交锋，后有智己汽车发布会误标小米汽车参数。如今，这场"PPT之战"再次升级，主角换成了两家即 将登陆港股的自动驾驶明星公司——小马智行与文远知行。 事件起因是，小马智行在近期一场面向香港投资人的路演中，使用了一页PPT进行友商对比，其中将文远知行的开放运营城市仅标注为"北 京"，订单量则标注为"零"。 这一举动迅速引燃战火。文远知行CFO李璇连夜发文，声称要正面回应小马的"虚假指控"，并就运营区域、运营数据、技术实力及技术路 线逐一进行驳斥。她直言不讳地指出："小马的行为已经超过正常竞争范畴，有片面不实、刻意贬低的表述。" 两家同样定于11月6日（本周四）冲刺港股的智能驾驶头部玩家，在上市前夕的敏感时刻"互扯头花"，显然并非简单的意气之争。 争的是规模，更是技术路线的"正统" 李璇的长文回应看似情绪化，但双方争夺的核心无非两点：数据规模与技术路线的先进 ...

Core Viewpoint - The article discusses the transformative journey of Li Auto in the autonomous driving sector, highlighting the shift from skepticism to a strong commitment to AI-driven end-to-end solutions, culminating in the successful launch of the "end-to-end + VLM" system, which significantly boosted sales and market presence [1][6][42]. Group 1: Strategic Shift - In March 2024, Li Auto's CEO, Li Xiang, expressed dissatisfaction with the company's autonomous driving performance, emphasizing the need for a decisive shift towards end-to-end technology [2][8]. - The introduction of the "end-to-end + VLM" system in July 2024 marked a pivotal moment for Li Auto, allowing the company to transition from a follower to a leader in the autonomous driving space [3][4]. Group 2: User Reception and Sales Impact - The "end-to-end + VLM" system received overwhelmingly positive feedback during initial trials, leading to a significant increase in user engagement, with 65% of test drives featuring the new technology by October 2024 [5][6]. - By the end of 2024, the delivery share of models equipped with the AD Max system (featuring the new technology) reached 75.4% in the 300,000+ yuan segment and 84.6% in the 400,000+ yuan segment, a dramatic increase from just 20% earlier in the year [6][50]. Group 3: Team Dynamics and Development - The autonomous driving team at Li Auto faced anxiety and uncertainty at the beginning of 2024, but the successful implementation of the end-to-end system led to a turnaround in morale and performance [8][12]. - Li Auto's strategy involved rapidly expanding its autonomous driving team from around 600 to over 1,000 by the end of 2023, although this expansion initially did not yield the expected results [9][10]. Group 4: Technological Innovation - The end-to-end approach allowed Li Auto to integrate various functions into a single model, enhancing efficiency and reducing complexity compared to traditional modular methods [57]. - The project was characterized by a rapid development cycle, with the team successfully delivering a demo version of the end-to-end system in just over a month, showcasing superior performance compared to previous iterations [31][52]. Group 5: Data-Driven Approach - The success of the end-to-end project was largely attributed to a robust data-driven strategy, which emphasized the importance of high-quality data over sheer manpower [63][71]. - Li Auto's data collection capabilities were built into every vehicle from the start, ensuring standardized and comparable data for algorithm training, which was crucial for the success of the autonomous driving system [71][72].

车企的辅助驾驶体系正加速转向AI。一个鲜明的表征是，头部新势力的车端辅助驾驶模型参数，已接 近许多AI大模型的参数量级。 36氪汽车了解到，小鹏汽车即将部署在车端的大模型，其参数量至少是70亿；另一家头部新势力理想汽 车，待明年它自研的辅助驾驶芯片上车后，其车端大模型参数也将来到70亿级。 这样的参数量已接近AI大模型的普遍参数量级。 小鹏、理想的AI布局 小鹏的车端大模型，是由内部正在开发的云端大模型——"小鹏世界基座大模型"蒸馏而来。之所以如 此，主要是为了应对车载辅助驾驶芯片算力、存储、内存带宽不足，从而无法直接在车端部署大模型的 问题。 2024年下半年，小鹏汽车开始向云端大模型迈进。目前，小鹏正在研发一个720亿参数打底的超大规模 自动驾驶大模型，即 "小鹏世界基座模型"，将在下个月的AI科技日上发布。 小鹏在今年4月的AI技术分享会上介绍，这个云端大模型，以LLM为骨干网络，使用海量多模态驾驶数 据训练，具备视觉理解、链式推理和动作生成能力。小鹏在云端完成对这一模型的训练后，会"取其精 华"，将蒸馏出的小模型部署到车端。 这种方法，参考的是DeepSeek已经使用过的知识蒸馏路线，其本质上是模型的 ...

Core Viewpoint - The appointment of Liu Xianming as the new head of Xiaopeng's autonomous driving center signals a strategic shift towards AI-driven models in the company's smart driving technology [2][3][12]. Group 1: Leadership Change - On October 9, Xiaopeng Motors announced the departure of Li Liyun from the position of head of the autonomous driving center, with Liu Xianming taking over [2]. - Liu Xianming's appointment is seen as a necessary move for Xiaopeng to build new technological barriers in the face of increasing competition from rivals like Li Auto and Huawei [3][12]. Group 2: Strategic Shift - Xiaopeng's autonomous driving strategy is transitioning from an engineering logic focus to an AI logic approach, emphasizing the development of a "base model" for smart driving [4][12]. - The company aims to leverage AI and data to regain its competitive edge in autonomous driving capabilities [12][13]. Group 3: Liu Xianming's Background - Liu Xianming joined Xiaopeng in March 2024 and has a strong background in model research, having previously worked at Facebook and Cruise [6][8]. - His experience in AI infrastructure and end-to-end development positions him well to lead Xiaopeng's efforts in building a unified AI model development department [11]. Group 4: Industry Context - The shift in Xiaopeng's leadership reflects a broader industry trend moving from traditional engineering-driven approaches to data-driven and model-centric paradigms in autonomous driving [13].

Core Insights - The article emphasizes the growing importance of Vision-Language Alignment (VLA) in the field of autonomous driving, highlighting its dominance in recent conferences and research outputs [1][3]. - VLA enables autonomous vehicles to make decisions in diverse scenarios, moving beyond traditional single-task methods, and offers potential solutions for corner cases [3][4]. Summary by Sections VLA in Autonomous Driving - VLA and its derivatives have become a primary focus for both autonomous driving companies and academic institutions, accounting for nearly half of the advancements in the field [1]. - The technology stack for autonomous driving VLA is still evolving, with numerous algorithms emerging, leading to challenges in entry and understanding [4]. Educational Initiatives - A new course titled "Practical Tutorial on Autonomous Driving VLA" has been developed in collaboration with Tsinghua University to address the challenges faced by learners in this field [5][6]. - The course aims to provide a comprehensive understanding of the VLA technology stack, covering various modules such as visual perception, language, and action [4][5]. Course Features - The course is designed to facilitate quick entry into the field by using a Just-in-Time Learning approach, making complex concepts more accessible [5]. - It aims to build a framework for research capabilities, helping students categorize papers and extract innovative points [6]. - Practical applications are emphasized, with hands-on sessions to bridge theory and practice [7]. Course Outline - The curriculum includes an introduction to VLA algorithms, foundational algorithms, and the role of Vision-Language Models (VLM) as interpreters in autonomous driving [12][14][16]. - It covers modular and integrated VLA approaches, detailing the evolution of language models from passive descriptions to active planning components [18]. - The course also addresses reasoning-enhanced VLA, focusing on long-chain reasoning and memory integration in decision-making processes [20]. Learning Outcomes - Participants are expected to gain a thorough understanding of current advancements in autonomous driving VLA and master core algorithms [25][26]. - The course requires prior knowledge in autonomous driving basics, familiarity with transformer models, and a foundation in probability and linear algebra [28]. Course Schedule - The course is set to commence on October 20, with a duration of approximately two and a half months, featuring offline video lectures and online Q&A sessions [29].

Core Insights - Bot Auto, co-founded by Hou Xiaodi, has made significant progress in achieving fully autonomous operations with its Hub to Hub commercial testing, marking a milestone that was previously outlined in TuSimple's vision [1][6] - The company has successfully completed its first product launch within two years, utilizing a total of $45 million (approximately 290 million RMB) for development, which is notably efficient compared to other autonomous driving companies [4][5] Funding and Financials - Bot Auto was established in July 2023 and has publicly disclosed a Pre-A funding round led by Linear Capital and M31 Capital, raising $20 million, with earlier funding rounds totaling over $45 million [2] - The efficient use of funds has allowed Bot Auto to achieve results that would typically take other companies significantly more time and money, with a cost efficiency ratio 50 times better than traditional autonomous driving firms [5] Testing and Development - The testing route covered 40 miles (approximately 64 kilometers) in Houston, demonstrating the system's ability to navigate complex urban and highway environments, including recognizing traffic signs and avoiding obstacles [5] - Bot Auto aims to validate multiple operational routes quickly, with the first route taking three months to verify, but future validations expected to be completed in two months or less [7] Business Model - Bot Auto operates under a Transportation as a Service (TaaS) model, focusing on running its own fleet of Level 4 autonomous trucks rather than selling vehicles to logistics companies [9] - The company is addressing a significant shortage of truck drivers in North America, positioning itself to meet the growing demand in logistics and e-commerce sectors [17][24] Technology and Safety - Bot Auto's technology stack incorporates both camera and LiDAR systems, enhancing safety for highway driving, and is inspired by Tesla's advancements in autonomous driving [11][13] - The company has developed a comprehensive safety net for its autonomous trucks, ensuring reliable operation even in adverse conditions [14] Market Position and Competition - Hou Xiaodi views the market as vast, with a significant shortage of truck drivers that cannot be filled by existing autonomous freight companies by 2030, indicating a large opportunity for growth [24] - The competitive landscape includes various players like Aurora and Waabi, but the company does not perceive them as direct competitors due to the market's size [24]

Core Viewpoint - The article highlights the steady growth of the new energy vehicle market in August 2025, with significant delivery increases for various companies, while also noting the competitive landscape and technological advancements in the industry [3][4][10]. Group 1: Market Performance - In August 2025, the retail market for narrow passenger vehicles is estimated at approximately 1.94 million units, representing a year-on-year growth of 2.0% and a month-on-month increase of 6.2% [3]. - The new energy vehicle retail sales are projected to reach 1.1 million units, with a penetration rate of about 56.7% [3]. - Six sample new force car companies (excluding Xiaomi) delivered a total of 199,279 vehicles in August, showing a year-on-year increase of 20.0% and a month-on-month increase of 5.9% [3]. Group 2: Company Deliveries - **Leap Motor**: Delivered 57,066 vehicles in August, up 88.3% year-on-year and 13.8% month-on-month, driven by strong sales of the B10 and C10 models [4]. - **Xpeng**: Reported deliveries of 37,709 vehicles, a year-on-year increase of 168.7% and a month-on-month increase of 2.7% [5]. - **NIO**: Achieved 31,305 vehicle deliveries, reflecting a year-on-year growth of 55.2% and a month-on-month increase of 15.9% [6]. - **Li Auto**: Delivered 28,529 vehicles, but experienced a year-on-year decline of 40.7% and a month-on-month decrease of 7.2% [6]. - **Aion**: Reported 27,044 deliveries, down 23.5% year-on-year but up 1.8% month-on-month [6]. - **Zeekr**: Delivered 17,626 vehicles, showing a slight year-on-year decline of 2.2% but a month-on-month increase of 3.8% [7]. - **Xiaomi**: Exceeded 30,000 vehicle deliveries in August, with strong demand for its new SUV model [7]. Group 3: Technological Advancements - The article discusses the acceleration of end-to-end technology applications in autonomous driving, marking the beginning of a new era in intelligent driving [8][10]. - Companies like Xpeng and those associated with Huawei have been leading the iteration and promotion of intelligent driving technologies since 2024 [10]. - The advancements in intelligent driving technology are expected to lower hardware barriers and expand applications in the mainstream market, particularly for vehicles priced under 200,000 yuan [10]. Group 4: Investment Recommendations - The article suggests a focus on companies with strong autonomous driving capabilities and those that are well-positioned in the new energy vehicle supply chain, including Geely, Xpeng, Li Auto, BYD, and Xiaomi [11][18]. - It emphasizes the importance of intelligent driving as a competitive factor and the potential for domestic suppliers to gain market share through cost-effective and responsive solutions [11].