Core Insights - The article discusses various cutting-edge directions in autonomous driving research, emphasizing the importance of deep learning and traditional methods for students in related fields [2][3]. Group 1: Research Directions - Key areas of focus include VLA, end-to-end learning, reinforcement learning, 3D goal detection, and occupancy networks, which are recommended for students in computer science and automation [2][3]. - For mechanical and vehicle engineering students, traditional methods like PnC and 3DGS are suggested as they require lower computational power and are easier to start with [2]. Group 2: Guidance and Support - The article announces the launch of a paper guidance service that offers support in various research areas, including multi-sensor fusion, trajectory prediction, and semantic segmentation [3][6]. - Services provided include topic selection, full process guidance, and experimental support, aimed at enhancing the research capabilities of students [6][7]. Group 3: Publication Opportunities - The guidance service has a high acceptance rate for papers submitted to top conferences and journals, including CVPR, AAAI, and ICLR [7]. - The article highlights the availability of support for various publication levels, including CCF-A, CCF-B, and SCI indexed journals [10].

近日，主线科技(北京)股份有限公司(下称"主线科技")在港交所递交招股书，拟在主板上市，国泰海通是独家保荐人。 IPO日报注意到，自动驾驶卡车领头羊主线科技由张天雷创立(他毕业于清华大学，曾参与联合创立百度无人车团队），投资方有蔚来等知名企业和机 构，最新估值高达38.6亿元。 主线科技的解决方案包含三大核心产品：AiTruck(智能卡车)、AiBox(智能终端)、AiCloud(智能云服务)。作为所有解决方案的通用"人工智能虚拟司 机"，AiTrucker建立了统一的算法基础，利用物流枢纽的复杂数据加速公司公路物流及城市交通解决方案的鲁棒性。公司的解决方案中使用的车辆主要是自 动驾驶卡车，并辅以其他类型的商用车。主线科技将这些产品战略性地部署在三大核心商业场景：Trunk Port(物流枢纽)、Trunk Pilot(公路物流)、Trunk City(城市交通)，依托公司的通用化架构，实现从物流枢纽、公路物流到城市交通的技术复用与无缝连接。 目前，主线科技已累计交付830辆AiTruck及349套AiBox，并获得821套AiTruck及920套AiBox的意向订单。 根据咨询机构弗若斯特沙利文的资料，20 ...

出品｜搜狐财经 对于市场关心的AI泡沫问题，他认为无论从投资强度、企业财务健康度还是资本开支比例看，当前风 险都远低于2000年科网泡沫时期。"当然，2026或2027年需要验证AI应用的商业逻辑，局部领域可能存 在过热现象，但整体风险可控。" 他还强调，科技投资需要同时关注产业趋势的进展和市场交易的结构。对于如何把握细分领域的投资机 会，麻绎文认为可以关注几项指标，"特斯拉机器人产业链的订单释放、国内存储大厂的扩产进度、以 及L3级自动驾驶的政策突破时间点，将是2026年需要紧盯的三大信号。" 尽管整体看好科技板块，麻绎文也提示了需要警惕的风险点。第一，交易层面存在拥挤度风险。部分细 分领域如光模块等，2025年涨幅较大，获利盘较多，虽然估值看似合理，但交易结构可能放大波动。 第二， 产业层面需关注资本开支节奏。麻绎文指出，2026年下半年需要重点关注海外云厂商，特别是 中小型公司的现金流状况和资本开支指引。 以下为直播内容精编： 基金佳问：2025年以来，A股活跃度进一步提升，上证指数时隔十年再度站上4000点，如何看待2025年 至今A股的表现？这轮行情的主要驱动因素是什么？ 麻绎文：今年市场有两个比较 ...

出品｜搜狐财经 作者｜汪梦婷 对于市场关心的AI泡沫问题，他认为无论从投资强度、企业财务健康度还是资本开支比例看，当前风险都远低于2000年科网泡沫时期。"当然，2026或2027 年需要验证AI应用的商业逻辑，局部领域可能存在过热现象，但整体风险可控。" 他还强调，科技投资需要同时关注产业趋势的进展和市场交易的结构。对于如何把握细分领域的投资机会，麻绎文认为可以关注几项指标，"特斯拉机器人 产业链的订单释放、国内存储大厂的扩产进度、以及L3级自动驾驶的政策突破时间点，将是2026年需要紧盯的三大信号。" 尽管整体看好科技板块，麻绎文也提示了需要警惕的风险点。第一，交易层面存在拥挤度风险。部分细分领域如光模块等，2025年涨幅较大，获利盘较多， 虽然估值看似合理，但交易结构可能放大波动。 第二， 产业层面需关注资本开支节奏。麻绎文指出，2026年下半年需要重点关注海外云厂商，特别是中小型公司的现金流状况和资本开支指引。 以下为直播内容精编： 编辑｜杨锦 【编者按】2025年，A股市场迎来里程碑式发展：总市值站上100万亿元的高峰，上证指数涨破4000点创下近十年新高。站在"十五五"规划的开局之年， 2026年 ...

Group 1 - Humanoid robots are seen more as fundraising and marketing gimmicks rather than being ready for large-scale production by 2026, as they are still in the exploration phase of training data and application models [1] - Fully autonomous driving is expected to become a reality in 2026, with Tesla achieving complete driverless operation in Texas and at least one other state [1] - The low-altitude robotics market is anticipated to experience accelerated growth due to advancements in AI autonomous flight capabilities and a gap in commercial drone usage in the U.S. [1] - The U.S. federal government is predicted to expedite the rollout of autonomous driving regulations, with a potential policy window emerging in 2026 [1] - Traditional automakers are expected to fully embrace robotics, following companies like BYD, Xiaomi, and Xpeng, with more entering the robotics industry starting in 2026 [1] - A new "competitive cooperation" dynamic is expected to form between China and the U.S. in the robotics sector, with China’s advantages in advanced manufacturing and supply chains making it a key partner for U.S. robotics firms [1] Group 2 - Tesla's robotics factory is projected to become the "mother" of the next-generation robotics system, with xAI's computational power and "truth-seeking AI" significantly enhancing the value of robotic systems [2] - The first trillion-dollar unicorn in the robotics field is anticipated to emerge in the coming years, primarily focusing on the integration of embodied intelligence and high-performance computing [2] - The "Mag 7" companies, including Apple, Google, Amazon, Meta, and Microsoft, are expected to frequently mention terms like "robot," "humanoid," and "embodied" in their earnings calls over the next year [2] - Brain-computer interfaces (BCI) are projected to make significant advancements towards "superhuman capabilities," with companies like Neuralink expected to achieve major clinical breakthroughs by 2026, particularly in the video game sector [2]

▍文章来源于 康谋自动驾驶 点击下方 卡片 ，关注" 康谋自动驾驶 " 公众号 获取更多自动驾驶资讯 随着自动驾驶技术的日益升级，以UniAD、FSD V12为代表的" 端到端 "架构正重构行业格局。这一架构试图通过 单一 神经网络 直接建立从 传感器输入 到 车辆控制 的映射，从而突破传统模块化累积误差的局限。 然而 端到端模型 对数据分布的 广度 与 深度 均有着高要求，尤其是对缺乏归纳偏置的 Transformer架构 而言，" 数据 规模 "与" 场景覆盖度 "可谓直接决定了 模型上限 。 现实路测数据 面临极端的 长尾工况 数据局限，如实车采集" 采不到、标不准、测不起、太危险 "。在此背景下，" 虚拟 数据集 "成为了大家关注的热点，通过构建涵盖极端天气、复杂交互及事故场景的高保真虚拟数据，我们不仅能够以 低成本、高效率 的方式生成 海量带标签的样本 ，更能为端到端模型提供 闭环训练环境 。虚拟数据集已不再是现实数 据的简单补充，而是训练 高阶端到端模型 不可或缺的一环。 为满足自动驾驶算法对 高质量数据资产 的迫切需求，并有效应对真实路测的局限，本文将全面阐述 高 保真虚拟数据 集SimData ...

Summary of the Conference Call for Wanji Technology Industry and Company Overview - The conference call discusses the advancements in the autonomous driving industry, particularly focusing on Level 3 (L3) autonomous driving applications approved in Chongqing and Beijing, marking a significant shift from assisted driving to true autonomous driving [2][4] - Wanji Technology specializes in autonomous driving and intelligent networking, boasting the highest domestic 192-line lidar technology, validated by multiple mainstream automotive platforms [2][5] Core Insights and Arguments - The approval of L3 autonomous driving signifies a major milestone in China's conditional autonomous driving sector, with the first two models approved for production being from Changan Automobile and BAIC's Arcfox S6 [4] - The demand for lidar technology is increasing due to the commercialization of autonomous driving, with a focus on enhancing perception accuracy and computational requirements [2][6] - The industry consensus suggests that a hybrid solution combining vision and radar is likely to become the mainstream approach for future autonomous driving, providing higher precision and reliability [2][7] - Wanji Technology is actively involved in the construction of intelligent networking in cities like Hangzhou and Guangzhou, with single vehicle value ranging from thousands to tens of thousands of yuan, indicating a dynamic pricing model as applications deepen [2][10][11] Additional Important Content - The intelligent networking business is benefiting from government policies and urban development initiatives, with 20 pilot cities entering large-scale demonstration phases starting in 2024 [9] - The ETC (Electronic Toll Collection) pre-installation business is experiencing rapid growth, with monthly shipments reaching tens of thousands of units, reflecting the increasing market demand for smart driving solutions [12] - The rise in lidar shipments indicates a robust demand from the robotics industry, with both commercial and domestic robots expected to drive rapid development in the sector [3][13] - Wanji Technology's establishment of joint ventures aims to align with industry developments and strategic business growth [14] - There is speculation that Tesla's pure vision approach may evolve as lidar costs decrease and the company gains deeper insights into lidar technology [15]

Core Viewpoint - Xidi Zhijia officially listed on the Hong Kong Stock Exchange, becoming the world's first publicly traded autonomous mining truck company, highlighting the potential of the autonomous driving sector [3] Group 1: Company Development - Xidi Zhijia was founded in 2018 by experts from Hong Kong University of Science and Technology and Silicon Valley, initially requiring financial support during its early stages [6] - The company has evolved into a leading autonomous driving enterprise in China, achieving commercialization solutions in closed environments, urban roads, and intercity roads [9] Group 2: Financial Support and Partnership - SPD Bank provided crucial financial support during Xidi Zhijia's startup phase, utilizing government risk compensation funds to facilitate initial funding [6] - SPD Bank expanded credit limits through the "Specialized, Refined, and New Little Giant Loan" to support Xidi Zhijia's technology research, market expansion, and capacity enhancement [7] - As Xidi Zhijia approached its IPO, SPD Bank leveraged its comprehensive service advantages to assist in the company's entry into the international capital market [7] Group 3: Future Collaboration - Post-IPO, Xidi Zhijia plans to deepen collaboration with SPD Bank in industrial finance and capital operations, aiming to create new value together [9] - SPD Bank is committed to providing professional and warm financial services to support more technology companies in their innovative endeavors [11]

Core Viewpoint - The surge in market capitalization of domestic GPU manufacturers like Moore Threads and Muxi Co., exceeding 600 billion yuan, reflects a capital frenzy in the industry, similar to previous trends observed in the market [1]. Group 1: IPO Trends and Market Dynamics - Several companies, including the first domestic GPU stock in Hong Kong, Birun Technology, are preparing for IPOs, driven by the optimized listing regulations in Hong Kong, particularly the new Chapter 18C, which allows unprofitable "specialized and innovative" tech companies to go public [3][10]. - The Hong Kong market has seen a significant increase in IPO activity, with 102 companies listed by 2025, raising a total of 272.476 billion HKD, a year-on-year increase of 226.62%, marking a four-year high [3]. - As of December 17, there are 298 companies in the IPO hearing process in Hong Kong, with 28 new applications in just half a month of December, surpassing the 18 from the same period in November [3]. Group 2: Financial Pressures and Market Entry - Many suppliers are pursuing IPOs primarily to address "blood-making" needs, often driven by contractual obligations rather than purely for growth capital [6][14]. - Companies like Yushi Technology, which filed for an IPO on November 28, reported significant financial losses, with pre-tax losses of 250 million yuan, 213 million yuan, and 212 million yuan projected for 2022 to 2024, indicating a pressing need for capital [8]. - The flexible and inclusive nature of Hong Kong's listing requirements, especially for unprofitable companies, has attracted many firms seeking funding support [10]. Group 3: Industry Challenges and Competitive Landscape - The automotive intelligence suppliers face common challenges, including ongoing losses, funding pressures, and insufficient self-sustaining capabilities, which are critical for their survival [14][16]. - The rapid technological iteration in the automotive sector necessitates continuous high R&D investment, impacting short-term profitability and creating a competitive environment where even successful IPOs do not guarantee long-term success [16]. - The shift in Hong Kong's capital market towards a more rational and stringent review process poses additional challenges for companies seeking to enter the market, as the focus has moved from merely having a good "tech story" to demonstrating solid technological capabilities and future growth potential [10][16].

编辑丨 机器之心 点击下方 卡片 ，关注" 具身智能之心 "公众号 >> 点击进入→ 具身 智能之心 技术交流群 更多干货，欢迎加入国内首个具身智能全栈学习社区： 具身智能之心知识星球（戳我） ，这里包含所有你想要的! 随着 VLA（Vision-Language-Action）模型的兴起，端到端自动驾驶正经历从「模块化」向「大一统」的范式转移。然而，将感知、推理与规划压缩进单一模型 后，主流的自回归（Auto-regressive）生成范式逐渐显露出局限性。现有的自回归模型强制遵循「从左到右」的时序生成逻辑，这与人类驾驶员的思维直觉存在本 质差异 —— 经验丰富的驾驶员在处理复杂路况时，往往采用「以终为始」的策略，即先确立长期的驾驶意图（如切入匝道、避让行人、靠边停靠），再反推当 前的短期操控动作。此外，基于模仿学习的模型容易陷入「平均司机」陷阱，倾向于拟合数据分布的均值，导致策略平庸化，难以在激进博弈与保守避让之间灵 活切换。 针对上述痛点， 复旦大学与引望智能联合提出了 WAM-Diff 框架 。该研究创新性地将 离散掩码扩散模型 （Discrete Masked Diffusion）引入 VLA 自动 ...