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 很多同学找实习秋招的时候，难免方向不对口。 尤其是自动驾驶算法岗实习或者正式岗位，要不你本身实力很强有多篇顶会，要不有直接的岗位经验，上来就能 干活。不然真的很难~现在自驾技术方向迭代很快，学生时代的技术栈不太匹配就业需求。。。 最近自动驾驶之心对接了很多秋招的小伙伴，大家普遍或多或少都有一些不匹配的问题。。。 还摸不清求职岗位的要求吗？ 分析学员画像：基于学员的简历和前期交流，评估学员的知识结构和能力模型，分析与目标岗位的具体差 距。 ...... 每年到求职季，很多童鞋总是不知道怎么写匹配岗位的简历！碰到了很多类似的问题： 这个是大多数转行的同学或新手都会遇到的问题，自动驾驶之心这两年收到了很多相关的求助。一直想做相关的 辅导，但没时间和契机开展，直至这个月，我们的求职辅导业务正式推出了。 目标人群：希望短时间成功转型智能驾驶（智驾、座舱）方向的学员，包括应届生和无目标岗位经验的职场人 士，周期2个月左右。 课程特色：以求职成功为导向，全程1v1辅导，聚焦目标岗位，迅速补足短板，短时间内具备目标岗位所需的 ...

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 论文作者 | Zeming Chen等 今天自动驾驶之心为大家分享 上海期智研究院&清华大学赵行老师团队 最新的工 作！ BEV-VAE：实现自动驾驶环视图像精准生成与操控。 如果您有相关工作需要 分享，请在文末联系我们！ 自动驾驶课程学习与技术交流群事宜，也欢迎添加小助理微信AIDriver004做进一 步咨询 >>自动驾驶前沿信息获取 → 自动驾驶之心知识星球 论文链接: https://arxiv.org/abs/2507.00707 代码仓库 (Github): https://github.com/Czm369/bev-vae 动机 编辑 | 自动驾驶之心 摘要 在自动驾驶中，多视角图像生成任务需要在不同相机视角下实现对三维场景的一致理解。然而，大多数现 有方法将其简化为二维图像集合的生成问题，缺乏对三维结构的显式建模。我们认为，对于自动驾驶场景 的生成任务，结构化表示至关重要。为此，本文提出 BEV-VAE 方法，实现具有空间一致性与可控性的多视 角图像生成。BEV-VAE 首先训练一个多视角图像 ...

Core Viewpoint - Tesla's reputation as a leader in autonomous driving technology has been severely challenged following the launch of its paid Full Self-Driving (FSD) pilot program in Austin, which showcased significant operational failures and raised questions about the company's reliance on low-cost camera systems instead of more advanced sensor technologies like LiDAR [1][2][4][5]. Group 1: Autonomous Driving Technology - Elon Musk has repeatedly stated that a significant portion of Tesla's traditional fleet will be converted into revenue-generating autonomous taxis, with expectations of "millions of self-driving Tesla cars" by 2026 [2][5]. - Tesla argues that a set of commercial cameras, trained on billions of frames, can achieve human-like vision and outperform more expensive sensor suites, but peer-reviewed literature challenges the feasibility of achieving Level 4 autonomy with cameras alone [2][4]. - A study published in June 2025 indicated that pure camera systems have a 40% higher misjudgment rate in fog and snow compared to systems equipped with LiDAR, raising concerns about safety in adverse conditions [2][4]. Group 2: Regulatory and Safety Concerns - The absence of radar exacerbates safety issues, as radar can measure relative speed and identify metal objects through rain or dust, providing a backup when cameras are obstructed [4][5]. - Recent incidents during the Austin pilot program, including a Model Y vehicle making dangerous maneuvers, have prompted investigations by the National Highway Traffic Safety Administration (NHTSA) [7][10]. - New Texas regulations effective September 1, 2025, allow the state to revoke autonomous driving permits that do not meet safety standards, highlighting the potential for increased regulatory scrutiny on Tesla's operations [5][7]. Group 3: Financial Performance and Market Reaction - Tesla's production in Q2 2025 was 410,244 vehicles, a slight increase from Q1 but a 0.2% decrease year-over-year, while deliveries fell 13.5% to 384,122 vehicles, missing market expectations [10][11]. - Following the disappointing delivery numbers, Tesla's stock price dropped 3.8%, reflecting investor concerns over the company's ability to generate revenue from its autonomous driving initiatives amid declining sales [11][12]. - Analysts are divided on Tesla's future, with some raising target prices based on potential FSD revenue, while others downgrade ratings due to rising regulatory risks and the uncertainty surrounding the FSD rollout [12][13]. Group 4: Future Outlook and Investor Sentiment - The failure of the Austin pilot program has led to increased legal liability risks, with potential collective lawsuits looming if passengers are harmed [13][17]. - Investors are advised to adjust their forecasts, anticipating no significant revenue from autonomous taxis until at least 2028, and to increase discount rates to reflect execution and legal risks [17][18]. - Despite the challenges, Tesla retains advantages such as a vast data collection capability and manufacturing efficiency, which could support future improvements in its autonomous driving technology [14][15].

Group 1 - Pony.ai has signed a strategic cooperation agreement with the Dubai Roads and Transport Authority (RTA) to promote the commercialization of Robotaxi services in Dubai [1] - The Middle East has become a core destination for Chinese companies expanding overseas, driven by abundant funding and favorable policies for autonomous driving technology [1][3] - Dubai's government aims for 25% of transportation to be handled by autonomous vehicles by 2030, with similar policies being implemented in Abu Dhabi and Saudi Arabia [1] Group 2 - WeRide has made significant progress in the Middle East's autonomous driving sector, being the first to collaborate with Dubai RTA [1] - In December 2024, WeRide and Uber launched the largest Robotaxi commercial operation fleet in the Middle East in Abu Dhabi [2] - By June 2025, WeRide, Uber, and Dubai RTA formalized their partnership to deploy Robotaxi services in Dubai, with plans for a fully unmanned operation by Q1 2026 [2] Group 3 - The competition in the Middle East's autonomous driving sector is intensifying as companies like Pony.ai accelerate their entry [3] - The combination of funding and policy support positions the Middle East as a region where the global autonomous driving ecosystem may mature rapidly [3] - WeRide has established a strong market position and local operational expertise, creating significant barriers to entry for competitors [3]

Core Viewpoint - The Chinese autopilot industry is at a critical stage of technological deepening and commercialization, showing a vigorous trend of multi-field collaborative development [1] Industry Overview - The autopilot system is designed to automatically control vehicles or aircraft using sensors, controllers, and actuators, allowing them to operate without continuous human intervention [2] - The industry has evolved through four main stages: exploration (1956-1978), initial application (1979-2014), policy-driven (2015-2019), and commercialization (2020-present) [4][5][6] Current Industry Status - In the first five months of 2025, China imported 23 autopilot systems, a year-on-year decrease of 99.53%, with an import value of 2.7065 million yuan, down 73.02%, indicating significant domestic replacement of high-end autopilot systems [10] - Exports reached 763 units, a year-on-year increase of 132.62%, with an export value of 2.7106 million yuan, up 152.65%, highlighting China's strong emergence in the global market [11] - Key drivers for this growth include cost-performance advantages, scenario adaptability, and expansion into emerging markets [11] Industry Chain - The upstream of the autopilot industry includes components like controllers, sensors, and AI chips, while the midstream focuses on R&D and production, and the downstream applications span across aircraft, vehicles, missiles, and spacecraft [8][9] Key Enterprises - Major players include Baidu, Huawei, DJI, and others, each specializing in different segments of the autopilot market, such as automotive, drone, and missile applications [18][20][22] - Baidu's Apollo platform integrates advanced sensors and algorithms for high-precision vehicle control, while DJI leads in consumer drone markets with its flight control systems [22] Industry Development Trends 1. **Technological Integration and Innovation**: The industry is experiencing a golden period of technological integration, with breakthroughs in core technologies like LiDAR and AI chips driving advancements in precision and reliability [24] 2. **Diversification of Application Scenarios**: The application of autopilot systems is expanding from single fields to diverse and cross-industry integrations, including Robotaxi services and agricultural drones [26] 3. **Policy and Standards Improvement**: Government support and policy guidance are providing a solid institutional foundation for the industry, with new standards and regulations being established to ensure safety and reliability [27]

自动驾驶岗位面试时，一份好的简历是什么样的？ 可以适当夸大，别太过分（简历上写的一定要是自己非常了解的）： 自驾行业是出了名的工资高，好多同学都想往这个方向卷！但你真的知道怎么写一份合格的简历 吗？最近好几位同学让我们帮忙改简历，但都存在各种各样的问题。 看了这么多简历，我觉得其中一位同学的蛮好，最终拿到了某新势力60k的offer，才3年经验！总结 下来，一份合格的简历是条理清晰、重点突出、细节体现、能力体现几个部分。不要乱堆项目和奖 励，要找符合项目岗位的优势点。 1）开门见山 结论先行，直接说出自己的成果和成就（可以在项目前） 举例主要成就： A公司：搭建了什么动态感知后融合，发表专利三篇； B公司：优化了静态目标的融合算法，优秀个人； 2）职责清晰 BEV 算法框架搭建：主要参与者（算法负责人） BEV 算法模型优化：负责人 3）逻辑清晰 每一个点都有目的，多用数字，条理分明，按照序号和标题进行改进（千万别段落式） 1）模型上 ，采用ohem + focal 解决长尾分布问题（经验），提升10%。改进ohem的方案（思考能 力） 2）数据上，10w数据整理，协调（综合能力和沟通能力） 3）部署和融合上 ...

Core Insights - WeRide (文远知行) is a leading autonomous driving technology company, recognized for its innovative solutions and significant market presence [1][2]. Financial Performance - As of March 31, 2025, WeRide reported total revenue of 72.437 million RMB, reflecting a year-on-year growth of 1.77% [1]. - The company experienced a net loss attributable to shareholders of 385 million RMB, which represents a year-on-year increase of 17.73% [1]. Upcoming Events - WeRide is scheduled to disclose its mid-year financial report for the fiscal year 2025 on August 20, 2023 [1]. Company Overview - WeRide, established in 2017 and registered in the Cayman Islands, operates through its domestic entity and has conducted autonomous driving research, testing, and operations in 30 cities across 7 countries [1]. - The company has been operational for over 1,600 days and holds autonomous driving licenses in China, the USA, the UAE, and Singapore, making it a unique player in the industry [1]. Product and Service Offerings - WeRide has developed a comprehensive product matrix that includes Robotaxi, Robobus, Robovan, Robosweeper, and Advanced Driving Solutions, catering to various sectors such as smart mobility, smart logistics, and smart sanitation [2]. - The company is recognized for its commercial revenue leadership among global peers in the autonomous driving sector [2]. Strategic Partnerships - WeRide has established strategic partnerships with several top global manufacturers and suppliers, including the Renault-Nissan-Mitsubishi Alliance, Yutong Group, GAC Group, and Bosch [2]. Industry Recognition - In 2023, WeRide was ranked eighth on Fortune's list of companies changing the world, making it the only Chinese company to enter the top ten [2].

今天自动驾驶之心很荣幸邀请到BezierGS工作的一作马梓培，为大家分享这篇ICCV'25中稿的新工作！今晚七点 半，自动驾驶之心直播间不见不散~ 1. 构建一个高质量街景世界，供自驾模型在其中训练、探索，减少数据采集的成本； 2. 减少对bounding box精确性的依赖，目前业界以及开源自驾数据集采集的准确性不是很高，bounding box的 标注不精确； 3. 这篇是对自驾世界的学习与探索，未来会探索一个真正的自驾世界模型，该工作只能实现轨迹内插，无法轨 迹外插。 论文链接：https://arxiv.org/abs/2506.22099 代码代码：https://github.com/fudan-zvg/BezierGS 复旦大学ICCV 2025中稿的最新工作！自动驾驶场景的真实重建对于开发闭环仿真至关重要。大多数现有方法依 赖于目标的位姿标注，使用这些位姿来重建动态目标并在渲染过程中实现动态重建。这种对高精度目标标注的依 赖限制了大规模和广泛场景的重建。为了解决这一挑战，复旦大学的团队提出了Bezier curve Gaussian splatting （BezierGS），该方法使用可学习的 ...

写在前面 & 笔者的个人理解 扩散模型在自动驾驶场景视频生成中已经展现出比较有前景的视觉生成质量。然而，现有的基于视频扩散的世界模型在灵活长度、长时序预测以及轨迹规划方面 仍存在不足。这是因为传统视频扩散模型依赖于对固定长度帧序列的全局联合分布建模，而非逐步构建每个时间下的局部化分布。本研究提出 Epona ，一种自回 归扩散世界模型，通过两项关键创新实现局部时空分布建模：1） 解耦的时空分解 ，将时间动态建模与细粒度未来世界生成分离；2） 模块化的轨迹与视频预测 ，通过端到端框架无缝整合运动规划与视觉建模。本文的架构通过引入一种新的"链式前向训练策略"（chain-of-forward training strategy），在实现高分辨率、长持 续时间生成的同时解决了自回归循环中的误差累积问题。实验结果表明，与现有方法相比，Epona在FVD指标上提升7.4%，预测时长可达数分钟。该世界模型进一 步可作为实时端到端规划器，在NAVSIM基准测试中优于现有端到端规划器。 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 今天自动驾驶之心为大家分享 地平线联合 ...

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 亲身经历：滴滴自动驾驶感知算法岗三轮技术面复盘！ 最近参加了滴滴自动驾驶部门感知算法工程师的面试，总共经历了三轮技术面，每一轮的面试官都问得非常深 入，尤其聚焦在项目细节和技术原理上。这里必须提醒大家： 面试前务必吃透简历上的每一个字！ 一旦被问 住，场面真的会很尴尬。 Q4：了解Anchor-Free检测吗？简述FCOS的核心流程。 重点考察对Anchor-Free代表算法FCOS的理解。 Q5：是否接触过端到端（End-to-End）检测算法？ 滴滴在自动驾驶领域布局多年，技术积累相当深厚。他们和广汽埃安联合成立的"广州安滴科技"专注于L4级无 人驾驶研发，投入力度很大。今年想冲击自动驾驶方向的同学，滴滴绝对是一个值得重点关注的选项。 第一轮技术面试回顾： Q1：自我介绍 面试官主要围绕我研究生期间的科研产出和研究方向进行了针对性提问。 Q2：论文核心创新点阐述？ 在讨论创新点时，面试官自然地延伸出了一些2D目标检测领域的相关问题进行探讨。 Q3：概述2D目标检测算法的主要演进脉络？ 需要梳理从传统方法到 ...