作者 | qian 编辑 | 自动驾驶之心 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 >>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 视觉大语言模型 综述汇总 基础理论 原文链接： https://zhuanlan.zhihu.com/p/1922228114404143784 预训练 智能交通和自动驾驶中的 LLM：https://github.com/ge25nab/Awesome-VLM-AD-ITS AIGC 和 LLM：https://github.com/coderonion/awesome-llm-and-aigc 视觉语言模型综述：https://github.com/jingyi0000/VLM_survey 用于 CLIP 等视觉语言模型的出色提示 / 适配器学习方法：https://github.com/zhengli97/Awesome-Prompt- Adapter-Learning-for-VLMs LLM/VLM 推理论文列表，并附有代码：https://github.com/D ...

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 这几天刚和团队小伙伴沟通完后期工作建设，探讨究竟要做一个什么样的自动驾驶社区？其中一个答案比 较符合我们的思路，那就是一个能够凝聚行业人群、遇到问题能够快速响应、影响到整个行业的地方。 我们目标是未来3年内打造一个万人聚集的智能驾驶&具身智能社区，这里也非常欢迎优秀的同学加入我们 （目前已经有华为天才少年、自驾领域研究前沿的多为大佬加入）。我们和多家业内公司搭建了学术 + 产 品+ 招聘完整的桥梁和链路，同时内部在教研板块也基本形成了闭环（课程 + 硬件+问答）。社区里面既能 看到最新的行业技术动态、技术分享，也有非常多的技术讨论、入门问答，以及必不可少的行业动态及求 职分享。具身智能这么火，要不要考虑转行？自动驾驶技术的未来发展趋势如何？大模型如何预自动驾驶 &具身智能结合？这些都是我们持续关注的 星球核心目录如下： 自动驾驶视频课程及硬件、代码实战 链接：https://t.zsxq.com/9GkD5 0 内部会员独享福利视频教程（免费）：涵盖超千元的自动驾驶技术论文解读 1 自动驾驶之心原创直播课程： ...

Autonomous Trucking Technology & Development - Aurora's autonomous trucking technology has reached a point where rapid launch and iteration are possible, demonstrating significant progress in the field [7] - Aurora is one of the lead customers for Nvidia's next-generation automotive computer system on a chip, viewing them as a strategic partner for computational power [11][12] - Aurora believes it currently has the best autonomous trucking technology globally, capable of doing things that others cannot safely achieve on the road [16] - Aurora plans to expand its autonomous trucking operations to include nighttime and rainy conditions, and to operate routes from Fort Worth to El Paso, El Paso to Phoenix, and Fort Worth to Phoenix by the end of the year [6][7] Regulatory & Governmental Support - The US government, including the Secretary of Transportation and the Vice President, is showing support for automation in trucking, which is considered vitally important for the industry's transformation [13][14] - Most states are supportive of autonomous trucking, with Aurora confident in its safety and ability to operate in 44 states [20][21] - California's Department of Motor Vehicles is putting in place regulations for heavy-duty trucks, indicating a positive regulatory environment [22] Business Model & Partnerships - Aurora's long-term business model involves providing "drivers" (autonomous driving systems) to customers who purchase trucks from manufacturers like PACCAR and Volvo [8][9] - Trucking companies like FedEx, Werner, Herschbach and Schneider will buy trucks with Aurora's "driver" installed and pay Aurora to operate the trucks [9] - PACCAR requested a human be present in the truck for supervisory purposes, which Aurora agreed to in the spirit of partnership, without impacting their technology roadmap [4][5]

Core Viewpoint - Pony.ai has achieved a significant milestone with the mass production of its seventh-generation Robotaxi, moving closer to its goal of expanding its fleet to 1,000 vehicles by the end of 2025, which sets the stage for large-scale commercial deployment [1][5]. Group 1: Production and Testing - The Gen-7 Robotaxi models from Guangzhou Automobile Group (GAC) and Beijing Automotive Industry Corporation (BAIC) have entered mass production, showcasing the successful integration of Pony.ai's autonomous driving technology [2]. - Road testing for the Gen-7 Robotaxi has commenced in Guangzhou and Shenzhen, marking a transition from controlled environments to real-world traffic scenarios [3]. Group 2: Technological Advancements - The Gen-7 autonomous driving system features three key advancements: the use of 100% automotive-grade components, a 70% reduction in bill-of-materials (BOM) costs for the autonomous driving kit compared to the previous generation, and a platform-based design that allows for rapid adaptation across various vehicle models [4]. Group 3: Strategic Goals - The year 2025 is designated as the "year of mass production" for Pony.ai, with plans to build a fleet of over 1,000 vehicles by year-end, indicating a focus on scalable growth and significant commercial opportunities [5].

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年的秋招已经开始了，不少双非的同学都很迷茫。。。 目前在实验室做的工作有些落伍，现在无论是自动驾驶还是具身智能公司都需要实力比较强、背景比较好的同 学。 以下是 知识星球 内部一位双非同学的提问，非常具有代表性： 各位大佬们好，我目前是一个双非的研究生，研究方向是多传感器融合定位的，然后学过python、深度学 习、ros等等，但都学的不是很精，现在想多学一点为以后找工作用，感觉算法岗我的学历可能不太行，请 问各位大佬们我应该往哪个方向学比较好呢？学些什么知识呢？ 星主回答：你的技术栈都比较偏机器人一些，SLAM和ROS这块都可以尝试一下和机器人/具身智能打交道。这 块需求也比较大，可以做一些优化、集成类工作~ 另一方面，我们了解到大一些的公司各家的hc都不是很多，要求基本上都是端到端、大模型、VLA、强化学 习、3DGS这些比较前沿的方向。如果你做的是这块，是有机会的，很多tire 1的公司或者主机厂也正在follow前 沿的技术，大概是从无图往端到端转，差不多技术栈推迟1-2代。像LV融合、无图、 ...

Tesla (TSLA -5.03%) stock got hit with a substantial valuation pullback in Tuesday's trading. The electric vehicle (EV) leader's share price ended the daily session down 5%. Meanwhile, the S&P 500 (^GSPC -0.11%) fell 0.1%, and the Nasdaq Composite (^IXIC -0.82%) fell 0.8%.News involving Tesla played a significant role in pushing the broader market lower in Tuesday's trading. After hitting record highs in Monday's session, some investors were likely already poised to take profits -- and another ramp-up in th ...

Core Insights - Alphabet Inc. is facing increasing scrutiny and competition, particularly in its autonomous driving unit, Waymo, and its core productivity suite, Google Workspace [2][9][10] - The company reported strong financial performance in Q1 2025, with revenue of $90.24 billion and EPS of $2.81, but must navigate significant challenges to maintain its market position [13] Group 1: Waymo and Autonomous Driving - Waymo aims to create a fully autonomous driving system, with millions of miles driven on public roads and services launched in Phoenix and San Francisco, now expanding to Los Angeles and Austin [3][4] - The long-term potential for autonomous ride-hailing is substantial, with the possibility of multi-billion-dollar revenue streams, but monetization remains limited and public perception poses challenges [4][5] - Tesla's rapid rollout of its robotaxi program presents a direct threat to Waymo, with Tesla's model allowing car owners to participate in ride-hailing, potentially scaling faster and achieving profitability sooner [6][7][8] Group 2: Competition and Market Dynamics - OpenAI's plans to develop a new workspace productivity platform could challenge Google Workspace, which is crucial for Alphabet's revenue and supports its advertising ecosystem [9][10][11] - If OpenAI's platform proves to be more innovative, it could disrupt Alphabet's enterprise market share over time, impacting the company's overall ecosystem [11][12] Group 3: Financial Performance and Future Outlook - Alphabet's stock forecast indicates a potential upside of 12.71%, with a target price of $199.95 based on analyst ratings [12] - The company must defend its core businesses against emerging competitors while converting long-term investments like Waymo into growth drivers to avoid falling behind [14]

Tesla's DON'T need LiDAR for autonomous driving.Perfect example of a Tesla stopping for an object that many would think only LiDAR would be able to deal with.You use your eyes to drive, not lasers (as cool as that would be though..) https://t.co/rqFhBCsD90 ...

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 不少双非的同学都很迷茫。。。 实验室参与的工作有些落伍，现在无论是自动驾驶还是具身智能公司都需要实力比较强、背景比较好的同学。 同时呢，现在也有很多机器人的公司处于快速发展的阶段，很多初创公司都融了比较可观的钱，未来几年问题不 大，而且培养是全方面的。 工作肯定是会苦一些，但技术栈这块确实实打实的可以学习到很多，也建议你看看，像深圳、杭州我们最近也都 线下拜访了很多机器人公司，具身智能毋庸置疑是最前沿的方向了，但确实传统的机器人仍然是产品的主线。 加油~ 最后欢迎大家加入 知识星球 ，硬核资料在星球置顶： 加入后可以获取自动驾驶视频课程、硬件及代码学习资 料。业内最全的全栈学习路线图，独家业内招聘信息分享~ 我们目标是未来3年内打造一个万人聚集的智能驾驶& 具身智能 社区，这里也非常欢迎优秀的同学加入我们（目 前已经有华为天才少年、自驾领域研究前沿的多为大佬加入）。我们和多家业内公司搭建了学术 + 产品+ 招聘完 整的桥梁和链路，同时内部在教研板块也基本形成了闭环（课程 + 硬件+问答）。社区里面既能看到最新 ...

而自从端到端和大语言LLM横空出世以来，大规模无监督的预训练 + 高质量数据集做具体任务的微调， 可能也会成为量产感知算法下一阶段需要发力的方向。同时数 据的联合标注也是当下各家训练模型的实际刚需，以往分开标注的范式不再适合智能驾驶的算法发展需求。今天自动驾驶之心就和大家一起分享下4D数据的标注流 程： 最复杂的当属动态障碍物的自动标注，涉及四个大的模块： 而为了尽可能的提升3D检测的性能，业内使用最多的还是点云3D目标检测或者LV融合的方法： 得到离线单帧的3D检测结果后，需要利用跟踪把多帧结果串联起来，但当下跟踪也面临诸多的实际问题： 离线3D目标检测； 离线跟踪； 后处理优化； 传感器遮挡优化； 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 千万级4D标注方案应该怎么做？ 智能驾驶算法的开发已经到了深水区，各家都投入了大量的精力去做量产落地。其中一块最关键的就是如何高效的完成4D数据标注。无论是3D动态目标、OCC还是静 态标注。 相比于车端的感知算法，自动标注系统更像是一个不同模块组成的系统， 充分利用离线的算力和时序信息，才能得到更好的感知结果 ...