Alphabet(GOOGL.US)旗下自动驾驶公司Waymo周二表示，将更新用于操作其自动驾驶汽车的软件，并 改进其紧急响应协议。上周六，其无人驾驶出租车因大规模停电在旧金山部分地区停滞，导致交通混乱 和城市部分区域堵塞。 当地时间12月20日晚间，太平洋燃气电力公司一个变电站发生火灾，导致旧金山约三分之一地区停电， 影响了约13万居民，并迫使一些企业暂时关闭。Waymo随后暂停了服务。 社交媒体上发布的多个视频显示，由于停电导致交通信号灯停止工作，Waymo的无人驾驶出租车滞留 在十字路口，危险警示灯闪烁。 Waymo表示，其自动驾驶汽车的设计能够处理信号灯熄灭的十字路口的状况，但车辆可能偶尔会请求 进行确认检查。 Waymo表示，这些确认协议在早期部署阶段是合理的，但公司目前正在完善它们，以适应其当前的运 营规模。Waymo正在实施全车队的软件更新，为车辆提供"特定的停电场景信息，使其能够更果断地导 航"。 Waymo还表示，将结合此次事件的教训改进其紧急响应协议。 Waymo在旧金山湾区、洛杉矶、亚利桑那州凤凰城都市区、得克萨斯州奥斯汀和佐治亚州亚特兰大运 营着超过2500辆汽车。该公司表示，已于周日 ...

Core Insights - The article discusses advancements in Vision-Language-Action (VLA) models for autonomous driving, highlighting a shift from traditional supervised learning methods to reinforcement learning (RL) approaches to enhance model generalization and reasoning capabilities [2]. Summary by Sections VLA + RL Research Overview - The article summarizes recent works in the VLA + RL domain, indicating a trend towards using RL to address limitations in previous models, particularly in terms of hallucination issues and the efficiency of continuous action space exploration [2]. Key Papers and Contributions - **MindDrive**: Introduces a framework that transforms action space into a discrete language decision space, achieving a driving score of 78.04 and a success rate of 55.09% on the Bench2Drive benchmark using a lightweight model [6]. - **WAM-Diff**: Proposes an end-to-end VLA framework that utilizes masked diffusion for trajectory optimization, achieving superior performance on the NAVSIM benchmark [7]. - **LCDrive**: Addresses temporal expression and latency issues in text chain reasoning by employing a latent chain-of-thought mechanism, demonstrating improved reasoning efficiency and trajectory quality [12]. - **Reasoning-VLA**: Develops a framework that enhances parallel trajectory generation through learnable action queries, achieving high performance across multiple datasets [13]. - **Alpamayo-R1**: Bridges reasoning and action prediction through a modular architecture and multi-stage training, improving generalization in long-tail scenarios [18]. - **AdaThinkDrive**: Introduces a dual-mode mechanism to balance decision accuracy and reasoning efficiency, achieving a PDMS score of 90.3 on the Navsim benchmark [20]. - **AutoDrive-R²**: Combines supervised fine-tuning and RL to enhance trajectory planning accuracy, achieving state-of-the-art performance with a significant reduction in error rates [25]. - **IRL-VLA**: Proposes a framework that avoids reliance on simulators by using a reward world model, achieving state-of-the-art performance on the NAVSIM v2 benchmark [31]. - **DriveAgent-R1**: Integrates active perception with hybrid thinking, achieving significant improvements in decision reliability and efficiency [32]. - **Drive-R1**: Connects reasoning and planning in VLMs, providing effective methods for integrating reasoning with motion planning [37]. - **ReCogDrive**: Merges cognitive reasoning with diffusion planners, achieving state-of-the-art performance while addressing the limitations of imitation learning [38].

Core Insights - Momenta is an autonomous driving company aiming to provide AI drivers for every household and user, with three significant "decadal visions": saving one million lives in ten years, freeing up 100% of time in ten years, and doubling logistics and travel efficiency in ten years [1][4]. Group 1 - The CEO defines the company as an "icebreaker," emphasizing the need for courage to explore uncharted territories in AI, new product forms, business models, and technologies [3][6]. - The company office names are categorized into two types: one representing new islands discovered during the Age of Exploration, symbolizing courage and innovation, and the other named after scientists, indicating that the company's discoveries and innovations are based on scientific methods [3][6]. - A crucial aspect of the company's culture is the "low-cost, short-cycle" hypothesis testing, acknowledging the limited time and funds available to a startup [3][6].

全球汽车行业正站在一个巨大的转折点上。据大摩21日发布的《机器人年鉴第六卷：自动驾驶汽车》（The Robot Almanac Vol. 6: Autonomous Vehicles），分析师Adam Jonas团队指出，尽管人类目前每年累计在汽车内花费的时间相当于8200万年，但这一传统的驾驶模式即将被打破，自动驾驶 技术正处于从"0到1"爆发的前夜。 在这一轮全球技术竞赛中，大摩明确指出中国已经处于领先地位。据追风交易台消息，报告数据显示，中国目前拥有全球约60%的L2+级自动驾驶汽车 份额，这不仅得益于中国在电动汽车（EV）领域的胜利，更在于其通过海量电动车队构建的"数据探针"优势。 与此同时，美国市场的竞争格局也正发生深刻变化，Waymo与特斯拉之间的路线之争愈演愈烈。Waymo正在从凤凰城、旧金山等地向洛杉矶、奥斯汀 及亚特兰大等全美主要城市快速扩张，其运营里程预计将在未来几年实现爆发式增长；而特斯拉采取的纯视觉方案则被视为"刚刚起步"，虽然硬件成本 极低，但旨在通过庞大的车队数据实现超越。报告将当前的传感器技术路线之争比作19世纪末的"电流之战"（War of the Currents），认为不同 ...

Core Viewpoint - The company Xiaoma Zhixing has launched an accessibility feature for its autonomous driving vehicles, aiming to provide a seamless experience for visually impaired users from booking to the end of the ride [1]. Group 1: Accessibility Feature Overview - The accessibility feature is designed based on extensive communication and testing with visually impaired user groups [1]. - It includes a full-process voice interaction system that allows users to confirm vehicle selection and receive real-time updates [3]. Group 2: User Interaction and Technology - Users can activate the feature by enabling the phone's screen reader or reading function, and the app will announce vehicle information to assist in selection [3]. - Upon arrival at the pickup point, users can press a button in the app to hear a voice prompt indicating the vehicle's location [3]. - The feature incorporates Bluetooth automatic unlocking technology, allowing users to unlock the vehicle by simply approaching it with their phone [3]. Group 3: Voice Command Functionality - After entering the vehicle and fastening their seatbelt, users can start the journey by saying "start the trip," eliminating the need for manual operation [3]. - The system can be activated with the wake word "Hello POPO," enabling users to adjust air conditioning and play music, with plans for additional voice control features in the future [3].

Group 1: Starlink User Growth - Starlink, a satellite internet service by Elon Musk, has surpassed 9 million active users globally, providing high-speed internet in 155 countries and regions [2] - The user growth rate has accelerated significantly, with the latest addition of 1 million users occurring in just 47 days [2] Group 2: AI Model Developments - Alibaba Tongyi has released a new open-source voice interaction model, Fun-Audio-Chat-8B, suitable for various applications including voice chat and customer service [3] - Zhiyu AI has launched and open-sourced its flagship model GLM-4.7, which enhances programming capabilities and task planning [4] Group 3: Security Incidents and Responses - WeChat has refuted rumors regarding account theft linked to a live streaming platform, confirming that their security measures are intact [5] - JD.com has responded to reports of a theft at its Paris warehouse, disputing the reported loss of over €37 million and stating that operations have returned to normal [6] Group 4: Mergers and Acquisitions - Harman International, a subsidiary of Samsung Electronics, has agreed to acquire ZF Group's Advanced Driver Assistance Systems (ADAS) business for €1.5 billion (approximately $1.8 billion) [7] Group 5: Legal Challenges in AI - A group of authors has filed a copyright infringement lawsuit against six AI companies, including OpenAI and Google, claiming their works have been used without compensation [8] Group 6: Automotive Developments - Zoox, an autonomous driving company owned by Amazon, is recalling 332 vehicles due to software errors that increase collision risks [9] - Beijing has issued the first L3-level autonomous driving vehicle license plates, marking a significant milestone in the commercialization of autonomous vehicles in China [13] Group 7: Robotics and AI Investments - Galaxy General has signed a strategic cooperation agreement with Baida Precision to deploy over 1,000 intelligent robots in industrial manufacturing [10] - EVE Energy has initiated the construction of a sodium battery headquarters and an AI robotics center, with a total investment of approximately 1 billion yuan [10] Group 8: Semiconductor Industry Updates - SMIC has implemented a price increase of about 10% on certain capacities, driven by rising demand for semiconductor products due to mobile and AI applications [15] - Chip and Semiconductor has completed its IPO counseling, indicating a clear direction for its capital market activities [19]

Core Viewpoint - Tesla's market capitalization has reached $1.63 trillion, driven by multiple positive factors including the reinstatement of Elon Musk's compensation plan and the expansion of Full Self-Driving (FSD) capabilities [1][2]. Group 1: Stock Performance - On December 22, Tesla's stock price rose by 3.66% to $498.82 per share, nearing the $500 mark, before closing at $488.73, reflecting a month-to-date increase of nearly 15% [1]. - Analysts' optimistic forecasts and the recent developments regarding Musk's compensation have contributed to the stock's upward momentum [1][2]. Group 2: Musk's Compensation Plan - The Delaware Supreme Court ruled that Musk's original $56 billion compensation plan must be reinstated, overturning a lower court's decision [2]. - The value of Musk's compensation plan has increased to approximately $140 billion, reflecting investor confidence in Tesla's future [2]. - A new compensation plan, approved by over 75% of shareholders, could be worth up to $1 trillion, although it includes challenging performance targets [2]. Group 3: FSD Expansion - Tesla announced the latest version of its "FSD Supervised" feature, which allows vehicles to autonomously navigate to destinations and park themselves [4]. - Musk indicated that this feature could be available in the UAE as early as next month, with the region serving as a potential model for broader adoption [4]. - The recent power outage in San Francisco did not affect Tesla vehicles using the FSD system, highlighting its reliability compared to competitors [4]. Group 4: Robotaxi Developments - Tesla is currently piloting a Robotaxi service in Austin without a safety driver, with Musk confirming ongoing tests [5]. - Analysts predict that Tesla will capture approximately 70% of the global autonomous driving market by 2026, with autonomous driving and robotics becoming core focus areas for the company [5].

智通财经APP获悉，智驾概念股多数走高，截至发稿，浙江世宝(01057)涨8.25%，报5.12港元；佑驾创 新(02431)涨5.25%，报14.23港元；黑芝麻智能(02533)涨2.02%，报19.68港元；禾赛-W(02525)涨1.2%， 报176.7港元。 此前，工信部公布了我国首批L3级有条件自动驾驶车型准入许可，分别为长安、北汽蓝谷两款车型。 交银国际指出，辅助驾驶功能已逐渐成为中国消费者购车决策的重要考量因素，部分车企已具备L3级 智驾方案和L3级车型量产能力。L3级自动驾驶法规的正式落地将成为2026年行业发展的关键催化剂， 值得重点跟踪。 消息面上，据央广网消息，12月23日，北京首批L3级高速公路自动驾驶车辆专用号牌，由市公安交管 局正式发放给北京出行汽车服务有限公司名下的3辆北汽极狐智能网联汽车，这也是国内首批发放的L3 级高速公路自动驾驶车辆专用号牌。标志着我国自动驾驶车辆已由测试示范进展到正式量产，率先开启 L3级自动驾驶时代。 ...

盘前要闻 1、英伟达放风春节前向中国客户交付H200芯片 类型：行业 情绪影响：正面 多位知情人士透露，英伟达已告知中国客户，计划于明年2月中旬，即春节前向中国客户交付其性能排 名第二的人工智能芯片H200。还表示，英伟达计划动用库存履行首批订单，预计发货总量为5000至 10000套芯片模组，相当于约4万至8万颗H200芯片。不过能否顺利交付仍存在较大不确定性。 2、我国自动驾驶正式进入量产!北京发放首批L3级车辆专用号牌 类型：行业 情绪影响：正面 12月23日，北京首批L3级自动驾驶车辆专用号牌正式发放给三辆智能网联汽车，这是国内首批发放的 L3级自动驾驶车辆专用号牌。标志着我国自动驾驶车辆已由测试示范进展到正式量产。研究认为，随 着特斯拉Robotaxi的持续放量验证，以及中国L3车型制度化落地，高阶自动驾驶产业拐点正加速到来。 3、伦铜突破12000美元、黄金涨破4500，贵金属牛市继续扩散 类型：市场 情绪影响：正面 在地缘政治紧张局势升级、以及美联储进一步降息预期的推动下，黄金和白银再度飙升至历史新高。截 至发稿，现货黄金首次涨破4500美元/盎司关口，白银每盎司超71美元。而伦铜突破12000 ...

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 论文作者 | Zhenya Yang等 编辑 | 自动驾驶之心 来自香港大学、华为以及华中科技大学的最新工作 GenieDrive，提出了一种以 4D Occupancy 作为中间表示的自动驾驶世界模型框架，在 4D 占据预测、轨迹可控性 以及长时序视频生成能力等方面均显著优于现有自动驾驶世界模型，为该领域提供了一条"先生成 4D 占据、再生成视频"的全新研究路径。 项目简介 GenieDrive 是一种新型的以 4D Occupancy 作为中间表示的自动驾驶世界模型框架，能够实现高度可控、多视角一致且符合物理规律的自动驾驶视频生成。得益于上述 设计，GenieDrive 在仅使用 3.47M 参数的情况下实现了 41 FPS 的推理速度，并在 4D 占据预测任务上取得了 7.2% 的 mIoU 提升；同时，通过在视频生成阶段引入归 一化多视角注意力机制，在 4D Occupancy 的引导下显著提升了多视角视频生成质量，将 FVD 指标降低了 20.7% ...