作者 | February 编辑 | 自动驾驶之心 原文链接： https://zhuanlan.zhihu.com/p/1943828436096353265 但一看量产清单：清一色本土新势力，最多出口东南亚。 点击下方 卡片 ，关注" 自动驾驶之心 "公众号 戳我-> 领取 自动驾驶近30个 方向 学习 路线 >>自动驾驶前沿信息获取 → 自动驾驶之心知识星球 本文只做学术分享，如有侵权，联系删文 谢邀，人在苏州，刚体验了 M 的产品。 主机厂员工，利益相关，先匿了。 先说结论： 比 Momenta 便宜的干不了全球车型，比 Momenta 技术好的干不了全球生意（某界），号称技术更好的压根不想干 L2+ 的脏活累活。 下面拆开聊。 1.比它便宜的，没有"全球上车"能力 大疆卓驭、轻舟智行确实便宜，10TOPS 以内就能跑高速 NOA，报价低到让采购部流泪。 欧盟 GSR 2024、北美 FMVSS 127、甚至韩国 K-NCAP 的法规场景，他们连测试用例都没建完。 Momenta 的秘诀是：把法规场景当成模型训练集，而不是交付后补补丁。 结果就是——同一代码主干，欧标/美标/中标三份标定，一套 OTA ...

一、报告核心观点 二、技术架构迭代分析 2. 中国玩家技术路径对比 技术路径：从无图方案转向端到端+VLA（视觉语言行动模型），强调用户体验（如语音控车、系统推理过程可视化）。 2024年9月推送VLA模型，训练分三阶段：基座预训练、动作微调（模仿学习）、强化学习优化。 2、元戎启行（第三方供应商）： 聚焦VLA路线，核心优势在于"思维链推理能力"，解决传统端到端黑盒问题，集成知识库提升泛化能力。 4、地平线机器人 & 莫门塔： 务实路线，侧重"端到端+强化学习"，兼顾客户算力配置多样性（如地平线硬件交付超千万片）。 莫门塔覆盖160多款车型，合作方包括合资、国企及民营企业。 三、商业模式变革：无人车（Robotaxi）业务提速 趋势：2026年基于量产车的Robotaxi业务将成为重点，渐进式路线（特斯拉、小鹏）优于跨越式路线（Waymo、文远知行）。 四、未来趋势与行业方向 1. 技术主旋律： AI智驾进入2.0阶段：以2024年"端到端范式"确立为1.0标志，当前行业基于模型能力提升和多样化数据训练，正迈向能自主应对极端场景的"智能涌现"阶 段。 2026年成为关键节点：技术架构、硬件迭代和无人车（Ro ...

汽车比较研究深度《AI 智驾 2.0，迈向智能涌 现》20251128 摘要 特斯拉宏图计划第四篇章聚焦 FSD 和人形机器人规模化，截至 2025 年 第三季度，FSD 累计行驶里程达 60 亿英里，并通过 Robotaxi 项目探 索商业模式，预计在无安全员情况下累计行驶超 25 万英里，同时规划 2026 年推出 AI 5 和 AI 6 芯片，提升算力与能耗比。 中国高级智驾发展经历了规则驱动、混合系统和纯数据驱动三个阶段， 特斯拉 FSD V14 版本代表了纯数据驱动的技术领先性，而理想汽车和 元戎启行则通过端到端加 WLM 方法向 VOA 模型进化，旨在提升智驾系 统的泛化能力和思维链推理能力。 VOA 模型通过具备沟通、思考和预判能力，以及强大的防御性驾驶行为， 显著改善智驾体验，尤其在遮挡路况、异形障碍物识别和复杂道路施工 区域应对方面表现突出，并提升了对交通环境文字路牌的识别能力和语 音交互体验。 华为 ADS 4.0 强调世界模型 VEVA，从数据驱动转向场景驱动，通过云 端世界引擎积累真实路测数据并生成高难度仿真题库，训练系统应对罕 见场景，车端世界行为模型则具备强大的泛化能力和预见性，强 ...

Core Insights - 2025 is identified as a pivotal year for automotive intelligence, initiating a three-year cycle that will elevate domestic electrification penetration rates to 50%-80%+, leading to a restructuring of the automotive landscape [4][10] - Leading intelligent driving manufacturers have successfully implemented complex urban scenarios such as roundabouts and U-turns, enhancing high-level functionalities like parking and scene understanding, thereby improving the driving experience for passengers and safety personnel [4][10] Investment Highlights - A comprehensive evaluation of six intelligent driving manufacturers, including ZunJie, Xiaopeng, Zhiji, Ideal, Xiaomi, and NIO, was conducted through large sample concentrated road tests and small sample in-depth road tests, focusing on scene implementation, takeover frequency, and comfort [5][10] - The keyword for the Beijing intelligent driving assessment is "stronger gets stronger," with Huawei and Xiaopeng leading in overall takeover counts and performance across various scenarios, particularly excelling in challenging situations [5][10] - Compared to Q1, the gap in intelligent driving capabilities among manufacturers has been narrowing by Q3, with second-tier manufacturers improving their performance in complex urban scenarios and reducing takeover frequencies [6][10] Road Test Overview - The large sample concentrated road test involved a standardized route in Beijing, assessing various performance metrics such as overall evaluation, takeover counts, stability, and efficiency [41][40] - The small sample in-depth road tests were conducted under varying traffic conditions, focusing on specific scenarios like roundabouts and complex intersections to evaluate the vehicles' decision-making and interaction capabilities [61][72] Manufacturer Performance - ZunJie achieved the highest overall evaluation score with an average takeover count of 1.16, demonstrating strong performance in challenging scenarios [44] - Xiaopeng's XOS model exhibited the lowest average takeover count at 0.94, showcasing balanced performance across various scenarios [48] - Zhiji's IM AD 3.0 received a score of 3.55 with an average takeover count of 1.44, indicating good handling of complex situations [49] - Ideal's OTA 8.0 scored 3.20 with a lower average takeover count of 1.06, reflecting a conservative driving style [52] - Xiaomi's V1.9.7 had an average takeover count of 3.86, indicating variability in performance across different scenarios [55] - NIO's cedar model recorded an average takeover count of 4.14, effectively covering most urban intelligent driving scenarios [58]

Core Viewpoint - The recent traffic accidents have led to a reevaluation of the overly promoted "intelligent driving" technology, highlighting the need for clearer communication and safety measures in the industry [1][3]. Industry Response - Following a serious accident involving intelligent driving, car manufacturers have begun to revise their marketing strategies, removing misleading terms like "automatic driving" and emphasizing the limitations of their systems [3][4]. - The Ministry of Industry and Information Technology has mandated that companies must clearly indicate system limitations in user manuals and advertisements, addressing the confusion among consumers regarding the capabilities of L2 level driving assistance [3][6]. Changes in Terminology - BYD has rebranded "high-level intelligent driving" to "driving assistance," while Li Auto emphasizes "safety rather than hands-free driving," marking a shift from a focus on technology to a focus on safety [4][6]. - Companies like Xiaopeng and Huawei are implementing training programs and transparency measures to educate users about the limitations and responsibilities associated with intelligent driving systems [5][6]. Consumer Awareness - Consumers are urged to maintain a rational understanding of intelligent driving technologies, recognizing that L2 remains an assistance tool and that L3 allows for limited hands-free driving under specific conditions [7][8]. - The industry stresses the importance of consumer awareness regarding data security and the need to keep control of the vehicle at all times, as no mass-produced vehicle currently achieves "true automatic driving" [8][9]. Future Outlook - The ultimate goal of intelligent driving technology is to enhance safety rather than replace human drivers, with a call for the industry to adopt a more responsible approach to marketing and technology development [8][9]. - The transition from "cool" features to "safety" as the core selling point is seen as essential for the healthy development of intelligent driving [9].

Investment Rating - The report does not explicitly provide an investment rating for the automotive industry Core Insights - The 2025 Shanghai Auto Show highlighted robots as the main attraction, with a focus on smart technology and globalization as strategic priorities, while electrification has reached maturity [2] - The competition in the large six-seat SUV segment is intensifying, with many new models launched, but some popular models were absent from the show [3] - The report emphasizes the increasing homogeneity of new electric vehicles, indicating a highly competitive market landscape [5] Summary by Sections Vehicle Launch Summary - Numerous flagship SUV models were showcased, including Zeekr 9X, BYD Dynasty-D, and GAC Trumpchi S9, focusing on comfort, multi-screen intelligence, and advanced driver assistance [3][10] - Key new models include the P7+ from Xiaopeng, which features a 5C supercharging battery and AI chassis, priced between 186,800 to 208,800 yuan [23][25] - The report notes that several significant models, such as Xiaomi YU7 and Li Auto i8, did not appear at the show, limiting new vehicle information [3][10] Intelligent Component Launch Summary - Third-party autonomous driving suppliers are actively launching new solutions, with Huawei announcing the commercialization of L3 scenarios and new technology routes [5][6] - The report highlights the emergence of various players in the domain controller and chassis sectors, with collaborations among major Tier 1 suppliers [6] - The report mentions the introduction of new intelligent driving solutions tailored to different automaker needs, indicating a growing ecosystem of partnerships [6] Company-Specific Highlights - **Huawei**: Launched the ADS 4.0 system with significant upgrades, including a new architecture and enhanced safety features [20][21] - **Xiaopeng**: Achieved strong global market performance with a Q1 delivery of 94,000 units, leading among new force brands [23] - **Li Auto**: Introduced the L6 with significant upgrades in appearance and intelligent driving capabilities [31] - **NIO**: Emphasized the importance of smart driving chips and operating systems in enhancing user experience and safety [32] - **BYD**: Showcased new concept cars and advanced technologies, including the cloud-riding system [38][42] - **Great Wall**: Focused on global expansion and technological advancements in smart driving systems [43][45] - **Changan**: Committed to advancing electric and intelligent technologies while expanding globally [46][47] - **Geely**: Integrated battery businesses and launched the Zeekr 9X with advanced intelligent driving features [50][51] - **SAIC**: Implemented a "Glocal" strategy to enhance its global presence while maintaining local relevance [54][56] - **GAC**: Introduced the first mass-produced L4 autonomous vehicle, showcasing advanced sensor technology [61][67]

Core Insights - The 2025 Shanghai Auto Show marked a shift in focus from power batteries to intelligent driving technologies, with companies emphasizing their advancements in smart driving systems [2] - The promotion of intelligent driving technologies adhered to regulations set by the Ministry of Industry and Information Technology, avoiding misleading terms and focusing on realistic portrayals of capabilities [2][16] - Key trends observed include accelerated technological iterations, deeper scenario-based implementations, and a balance between safety and user experience [2] Intelligent Driving Technology Routes - Five distinct intelligent driving technology routes were showcased, reflecting the diversity of approaches within the industry [3] - Huawei's multi-sensor fusion route, emphasizing high-precision perception through 192-line LiDAR, was prominently featured, showcasing its latest advancements [4] - Tesla represents the pure vision technology route, while companies like Xiaopeng have shifted from LiDAR to pure vision technologies [4] End-to-End Large Model Approach - The end-to-end large model route emerged as a popular technology path, aiming to evolve from "mimicking humans" to "surpassing humans" [5] - Huawei's ADS 4.0 completed 600 million kilometers of high-speed L3 simulation testing, indicating readiness for commercial deployment [7] - Xiaopeng's 720 billion parameter autonomous driving model, significantly larger than mainstream models, demonstrates advanced decision-making capabilities [7] Scene-Based Implementation - Intelligent driving technologies have transitioned from concepts to practical applications, with nearly all participating companies demonstrating real-world implementations [9] - Ideal's L6 model features dual Thor-U chips with 1400 TOPS of computing power, enhancing performance in complex driving scenarios [10] - Leap's C16 model has reduced the price of LiDAR-equipped vehicles to the 130,000 yuan range, improving parking efficiency by 40% [13] Safety and Redundancy - Safety redundancy has become a focal point, with companies emphasizing their systems' ability to handle various driving conditions [14] - The collaboration between 4D Mapping and Qualcomm introduced a comprehensive safety architecture, ensuring 100% redundancy in core functionalities [15] Regulatory Compliance - Recent regulatory tightening has prompted companies to adopt a more rational approach to marketing their intelligent driving technologies [16] - Companies like Xiaopeng are proactively educating users about the limitations and proper use of their driving assistance features [16][17] L3 Level Autonomous Driving - The auto show highlighted the imminent arrival of L3 level autonomous driving, with some companies already meeting hardware requirements [18] - L2 level driving assistance has become standard in new vehicles, with a penetration rate of 55.7% expected to exceed 65% by 2025 [18] - L3 level autonomous driving has shown high success rates in highway scenarios but faces challenges in urban environments, particularly under adverse conditions [18]

未经许可，不得转载或者引用。 投资要点 乘用车跟踪： 行业景气度跟踪： 3月产批零表现符合预期。 产量方面：乘联会口径3月狭义乘用车产量实现 248.1万辆（同比+12.9%，环比+42.9%）。销量方面：乘联会口径批发销量实现241.2万辆（同 比+10.2%，环比+36.5%。交强险口径销量为182.7万辆，同环比分别+19.7%/+42.7%。出口方 面：乘联会口径整体出口39.1万辆，同环比分别为-3.7%/+12.0%。 新能源跟踪： 3月新能源批发、零售渗透率有所上升。 3月新能源汽车批发渗透率46.8%，环比 +0.4pct；零售口径渗透率为53.0%，环比+1.8pct。 自主跟踪： 3月整体自主批发市占率环比下降。 批发/零售自主品牌市占率分别为66.0%/62.9% 城市NOA级别智能化整车环节跟踪： 城市NOA级别级别渗透率跟踪： 3月高阶智驾渗透率环比增长，产业趋势明确。 新能源乘用车 城市NOA级别级智能驾驶渗透率为18.3%，同/环比+7.9/+1.7pct，同环比增长。重点车企来看， 3月城市NOA级别智驾渗透率问界>理想>小鹏；问界智驾渗透率水平达84.0%；小鹏3月交付稳 定 ...