陈黎明指出，智能驾驶技术正处于关键发展阶段。从创新曲线来看，当技术处于10%到20% 的阶段时，要么迎来快速增长，要么走向消亡。智能驾驶技术已经迎来了拐点，未来将快速增 长。这一判断基于三点：一是智能驾驶是我国智能网联新能源汽车发展战略的重要组成部分，政 府在电动化和智能化方面都做出了明确规划；二是城区NOA（Navigate on Autopilot）和高速NOA 功能逐渐从高端车型下探至中低端车型，推动了技术的普及和成本降低；三是麦肯锡的统计报告 显示，智能化已成为消费者选择电动车的重要因素，尤其是在高端品牌中，智能驾驶和智能座舱 的属性备受关注。 7月10日至12日，第15届中国汽车论坛在上海嘉定成功举办。 陈黎明分享了地平线在智能驾驶领域的最新研究成果。地平线研发了Vision Mamba算法，该 算法通过引入双向量状态空间和位置编码器，解决了计算和视野问题，具有线性复杂度和全局感 受野，性能显著提升。地平线的Senna架构结合了端到端模型和VLM（Visual Language Model）模 型，提升了系统的泛化能力和决策能力，降低了规划误差和碰撞率。地平线的征程6系列芯片覆 盖低、中、高阶智能驾 ...

Core Insights - The intelligent driving industry is at a critical inflection point, with the potential for rapid growth or falling into an innovation gap as technology penetration reaches 10% to 20% [4] - The future growth of mid-to-high level intelligent driving is supported by three main factors: its central role in China's smart connected new energy vehicle strategy, the decreasing costs and increasing accessibility of advanced driving features, and strong consumer demand for intelligent features in vehicle purchases [4] Industry Trends - Intelligent driving is a core component of China's "14th Five-Year Plan" and is expected to drive the automotive industry from manufacturing to intelligent manufacturing [4] - The cost of advanced driving features has decreased significantly, with functionalities like urban NOA and highway NOA now available in vehicles priced below 150,000 yuan, accelerating the "smart driving equality" movement [4] - Consumer surveys indicate that intelligent features are among the top factors influencing the purchase of new energy vehicles, with a strong emphasis on intelligent driving and smart cockpit features in high-end models [4] Challenges - Despite advancements, intelligent driving still faces challenges, particularly in urban NOA, which is currently at a "usable" level but not meeting user expectations [5] - Issues include the system's performance in complex scenarios, limited operational design domain (ODD), and high costs, necessitating improvements in both system performance and application range [5] Technological Innovations - Horizon has made significant breakthroughs in four core areas: algorithms, software architecture, computing power, and data/engineering capabilities [6][7][8] - The Vision Mamba algorithm represents a key advancement in visual perception technology, achieving a 2.8 times increase in computation speed and an 87% reduction in GPU memory usage [6] - The new Senna multi-modal architecture integrates sensor and map data for improved vehicle control, resulting in over a 40% reduction in average planning displacement error and a 45% decrease in collision rates [6] Product Development - Horizon has launched the Horizon SuperDrive urban assistance driving system, which utilizes a combination of cameras and radar to enable full-scene intelligent driving without high-precision maps [9] - The system has demonstrated its ability to handle complex driving scenarios and is set for mass production this year [9] Future Outlook - The future of intelligent driving is expected to evolve towards hardware integration and software unification, with a focus on creating user-friendly intelligent driving products [9]

Core Insights - The 15th China Automotive Forum will be held in Shanghai from July 10 to 12, 2025, focusing on the theme "Quality Improvement and New Directions for Intelligent Driving" [1] - Dr. Chen Liming, President of Horizon Robotics, delivered a keynote speech discussing the latest research and industry insights in the field of intelligent driving [1] Industry Trends - Intelligent driving is at a critical inflection point in China, driven by three main factors: its significant role in the national strategy for smart connected new energy vehicles, the industry's push for technology democratization, and increasing consumer demand for intelligent features in electric vehicles [3][7] - The functionality of Highway Navigation on Autopilot (NOA) is becoming more user-friendly, while urban NOA is still in a "usable" stage, facing challenges such as complex traffic conditions and system performance issues [7] Technological Innovations - Horizon Robotics has made significant advancements in visual perception algorithms, evolving from traditional methods to the new Vision Mamba algorithm, which improves computational efficiency and resource optimization [11] - The Senna architecture combines end-to-end models with Visual Language Models (VLM), demonstrating over 40% reduction in average planning displacement error and 45% decrease in average collision rates compared to traditional models [14] Chip Technology - The Journey 6 series chips from Horizon Robotics cover a full range of intelligent driving applications, with the Journey 6P chip innovating in areas such as function, floating-point operations, and memory management, effectively reducing system deployment costs [17] Data and Engineering Capabilities - Horizon Robotics has introduced a solution combining reinforcement learning with world models, enabling continuous interaction with the environment for autonomous learning and optimization [19] - The company has achieved ISO 8800 certification for road vehicle safety and AI, providing over 8 million solutions and aiming to exceed 10 million this year, supporting over 200 vehicle models [19] Product Launches - The Horizon SuperDrive (HSD) for urban driving utilizes a one-stage end-to-end model and is set to enter mass production this year, featuring a configuration of 11 cameras and 1 millimeter-wave radar [22] - Horizon Robotics emphasizes its commitment to serving both domestic and global markets, collaborating with major international companies like Bosch and Denso [22]

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15个 方向 学习 路线 2025上半年已经过去了，上半年自动驾驶发展的怎么样？业内落地量产在做什么？学术界在关心什么技术方 向？转行具身还是深耕智驾？全都是问题！ 为此，自动驾驶之心采访了国内外数十位学术界和工业界的诸位大佬，和他们一起聊聊对这些问题的看法，希望 对大家有帮助~ 1. 当下前沿技术的发展路线是否已经成熟并适合量产？有哪些难点需要攻克？ 2. 如何看待新兴的技术方向：VLA/VLM、扩散模型、闭环仿真、强化学习、端到端自动驾驶、世界模型等等 3. 未来自动驾驶技术还有哪些值得探索的新方向？ 4. 深耕自动驾驶 or 投身具身智能，怎么选？ SJTU 张志鹏 BEV，Occ， E2E， VLA， 这几个词是从2022年到2025年自动驾驶中热词的一个变迁。BEV和Occ的引入某种程 度上都是因为特斯拉推崇纯视觉方案，然后借鉴Lidar和机器人的一些建模方式去表征整个驾驶环境。近期又出 现了一些用3DGS建模3D环境的方式。但是整体来看，自动驾驶中感知的场景表征的方式还是比较成熟了，各家 厂商也验证了这种模式在实际中是走的 ...

Core Viewpoint - Horizon Robotics aims to build a foundational platform that empowers various robotic applications, positioning itself as a leader in the robotics era, particularly in the automotive sector [4][10]. Group 1: Company Milestones - Horizon Robotics celebrated its 10th anniversary on June 1, marking its evolution from being the first Chinese company focused on deep neural network computing solutions to becoming the largest mass-production intelligent driving technology company in China [4]. - The company is set to officially list on the Hong Kong Stock Exchange on October 24, 2024, achieving the largest IPO for a Chinese tech company since 2022 and the fastest entry into the Hang Seng Tech Index [4]. - Horizon has been included in the FTSE Global Equity Index Series, effective June 20, 2024, enhancing its visibility to global investors [4]. Group 2: Market Position and Collaborations - Horizon has established partnerships with over 40 automotive manufacturers, including the top ten Chinese car companies, and has achieved over 8 million units of pre-installed production [12][14]. - The company holds a market share exceeding 40% in the Chinese OEM Advanced Driver Assistance Systems (ADAS) market for 2024, maintaining its position as the leading supplier of intelligent driving solutions [14]. - Horizon's technology is currently integrated into one out of every three intelligent vehicles in China, with expectations to surpass 10 million units shipped by the end of the year [12][14]. Group 3: Technological Innovations - Horizon Robotics has consistently focused on the integration of hardware and software, achieving over 1000 times improvement in computing performance over the past decade [23][25]. - The company has introduced significant advancements in its core algorithms, including the end-to-end autonomous driving model UniAD and the next-generation Vision Mamba model, which has gained recognition in the AI research community [24]. - Horizon emphasizes the importance of continuous technological breakthroughs to establish a long-term competitive advantage, with a commitment to high R&D investment, projected at approximately 5 billion yuan for 2024 [21][25]. Group 4: Future Vision - The long-term vision of Horizon Robotics, referred to as "Horizon 2.0," focuses on expanding into the general robotics field, with automotive applications being the initial focus [10][19]. - The company aims to leverage its technology to enhance automation capabilities in various sectors, including elder care, household service robots, agriculture, manufacturing, and logistics [10]. - Horizon's strategy includes deepening collaborations with a wide range of partners to drive advancements in intelligent driving and ensure safer, more efficient transportation solutions for everyone [19].

行业趋势：智驾平权下受益于渗透率提升及产业链自主可控智能化已成汽车行业下半场竞争胜负手， 2025 年车企智驾平权战略密集发布，乘用车中高阶智驾配置下沉至售价10 万元型，新车渗透率有望从 2024 年约10%大幅提升。在产业链自主可控趋势下，公司征程6 系列开启国产替代新范式，加速智驾面 向全场景、全生态量产应用与标配普及，成为主机厂选择的最大公约数。 竞争优势：软硬一体能力及战略、商业模式及全球化领先性公司核心竞争优势在于软硬一体的稀缺模 式，具备高阶智驾全栈解决方案能力，覆盖产业链高技术壁垒及高附加值环节。当前高阶智驾行业仍处 于技术快速迭代及用户价值认知阶段， 公司以J6P 芯片及Nash BPU 打造了坚实的系统底座，以先进算 法及工程化能力追求最优性能及量产效率，以快人一步的战略及开放柔软的商业模式实现价值共创，以 全球化的体系能力及资源破圈出海。2025 年智驾平权趋势下，公司有望逐步确立市场优势地位。 投资价值：智驾芯片自主可控龙头+高阶智驾价值创造 当前市场对公司的投资价值认知或主要集中在智驾芯片自主可控且仍担心主机厂自研芯片导致公司可拓 展的市场空间受限；此外，目前新势力主机厂高阶智驾以全 ...

地平线机器人-W系列 车中旭霞 智能驾驶芯片深度 《 智能驾驶芯片行业专题-智能驾驶进入快车道，地平线机器人-W和黑芝麻智能的投资价值分析 》——20250207 公司点评 《 地平线机器人-W（09660.HK）-2024年收入同比增长54%，智能平权时代征程6系列芯片加速放量》 ——20250328 《地平线机器人-W（09660.HK）-地平线征程6量产上车，共启全民智驾时代新篇》 ——20250309 核心观点 2024年公司实现营业收入23.8亿元，同比增长54% 2024年，地平线机器人-W实现总收入为23.8亿元，同比增长53.6%；净利润为23.5亿元，2023年为-67.39亿元，主要系 优先股及其他金融负债的公允价值变动收益46.77亿元影响（2023年为-47.6亿元）。2024年公司经调整经营亏损14.95 亿，相比2023年的16.87亿元有所收窄。 分业务来看， 2024年，汽车解决方案收入23.12亿元，同比增加57.2%，1）汽 车产品解决方案收入6.64亿元，同比增加31.2%，主要系公司全年交付量增加所推动；2）授权及服务业务收入16.58亿 元，同比增加70.9%，主要是 ...

评论 智能驾驶芯片深度 《 智能驾驶芯片行业专题-智能驾驶进入快车道，地平线机器人-W和黑芝麻智能的投资价值分析 》——20250207 公司点评 《地平线机器人-W（09660.HK）-地平线征程6量产上车，共启全民智驾时代新篇》 ——20250309 事项 公司公告： 根据地平线官微，地平线征程6系列全方位覆盖智能驾驶全阶应用，征程6系列共推出六个版本，包括征程 6B、征程6L、征程6E、征程6M、征程6H、征程6P，面向不同智能驾驶场景进行了计算方案的灵活配置，均能提供兼 顾性能与成本的最优解， 1）面向低阶智驾市场，征程6B以极致性价比重塑主动安全新体验，致力于打造业界最强性 价比主动安全一体机方案；2）面向中阶智驾市场，地平线推出普惠城区性价比方案最优解——征程6M，以及极致体 验高速NOA最优解——征程6E；3）面向高阶智驾市场，地平线推出了专为新一代全场景智能驾驶而生的征程6P，单 颗征程6旗舰可支持感知、规划决策、控制等全栈计算任务，将重新定义全场景NOA计算效率。 截止2025年1月，地平 线征程家族车载智能计算方案累计已出货超700万套。征程6系列已获超20家车企及品牌的平台化定点，自20 ...