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 端到端和VLA涉及的技术栈实在是太多了，今天就从小白入门学习的角度和大家聊聊端到端和VLA的发展路线。 首先看一下大语言模型的近五年的关键时间线： 聊大模型，离不开Transformer，为了方便后续理解，我们进行一个通俗的概括。 进一步展开Token化、BPE、位置编码等等~ Transformer: Attention is all you need 3. 合并频次最高的两个非结束字符组成一个新 字符,并重新统计所有字符频次(新字符会分 走部分原高频字符的频次 ) 4. 重复2-3直至字符数量达标or迭代轮次达标 $$P E_{(p o s,2i)}=s i n(p o s/10000^{2i/d_{\mathrm{model}}})$$ PE(pos,2i+1) = COS(pos/1000022/dmodel 7 x D 向量 "这是一段文字" Tokenizer + Positional 231 34 462 4758 762 38 7 x D 向量 Encoding [EQgmbedding 7 ...

不过值得注意的是，与一季度的14.9%的毛利率相比，半年度毛利率有所下降，零跑汽车在二季度的实 际毛利率不足14%。 在智能辅助驾驶技术落地方面，零跑汽车基于端到端算法的城市通勤领航辅助功能首次实现量产并上 车，并计划在下半年加大端到端和VLA技术的研发资源，在今年年底前实现城市NOA组合辅助驾驶能 力。 这家新势力接下来的月均销售量目标达6万辆。 8月18日，零跑汽车（09863.HK）发布利润转正的中期业绩，并提出年度盈利5亿~10亿元的净利润指 引，同步上调全年销售目标至58万辆~65万辆。 2025年上半年，零跑汽车实现净利润为人民币0.3亿元，撇除作为雇员福利开支一部分的以股份为基础 的付款，经调整净利润（非国际财务报告准则）为人民币3.3亿元。继理想汽车后，零跑汽车成为中国 造车新势力中第二家实现半年度盈利的企业。 交付数量方面，2025年上半年，零跑汽车交付新车22.17万辆，成为中国新势力品牌销冠，较2024年同 期交付量增长155.7%；实现营业收入242.5亿元，较2024年同期增加174%；毛利率为14.1%；同时，零 跑在手资金达295.8亿元。 零跑汽车预期销售规模在下半年大幅增加， ...

Core Viewpoint - The article emphasizes the evolving landscape of autonomous driving, particularly the integration of traditional planning and control (PnC) with end-to-end systems, highlighting the necessity for professionals to adapt to these changes in order to remain competitive in the job market [2][4][29]. Group 1: Industry Trends - The shift towards end-to-end and VLA (Vision-Language Alignment) systems is impacting traditional PnC roles, which are now required to incorporate more advanced algorithms and frameworks [2][4]. - As of 2025, end-to-end systems are expected to become more prevalent, yet traditional PnC methods will still play a crucial role, especially in safety-critical applications like Level 4 autonomous driving [4][29]. - The article discusses the importance of understanding both traditional and modern approaches to planning and control, as they are increasingly being integrated in practical applications [4][29]. Group 2: Educational Offerings - The company has launched specialized courses aimed at bridging the gap between theoretical knowledge and practical application in the field of autonomous driving, focusing on real-world challenges and interview preparation [5][7]. - The courses are designed to provide hands-on experience with current industry practices, including classic and innovative solutions in PnC, and are tailored for individuals with some background in the field [8][12]. - The curriculum includes modules on foundational algorithms, decision-making frameworks, and advanced topics such as contingency planning and interactive planning, which are critical for modern autonomous driving systems [20][21][24][26][29]. Group 3: Career Development - The courses not only focus on technical skills but also offer support in job application processes, including resume reviews and mock interviews, to enhance employability [9][10][31]. - Previous participants have successfully secured positions at major companies in the autonomous driving sector, indicating the effectiveness of the training provided [10][12]. - The program aims to equip participants with the skills necessary to construct decision-making systems and address real-world challenges in autonomous driving, thereby enhancing their career prospects [13][29].

Core Viewpoint - The article discusses the impact of end-to-end autonomous driving technology on traditional rule-based control (PNC) methods, highlighting the shift towards data-driven approaches and the complementary relationship between the two systems [2][6]. Summary by Sections Differences Between Approaches - Traditional PNC relies on manually coded rules and logic for vehicle planning and control, utilizing algorithms like PID, LQR, and various path planning methods. Its advantages include clear algorithms and strong interpretability, suitable for stable applications [4]. - End-to-end algorithms aim to directly map raw sensor data to control commands, reducing system complexity and enabling the model to learn human driving behavior through large-scale data training. This approach allows for joint optimization of the entire driving process [4]. Advantages and Disadvantages - **End-to-End Approach**: - Advantages include reduced system complexity, natural driving style emulation, and minimized information loss between modules [4]. - Disadvantages involve challenges in traceability of decision processes, high data scale requirements, and the need for rule-based fallback in extreme scenarios [4]. - **PNC Approach**: - Advantages include clear module functions, ease of debugging, and stable performance in known scenarios, making it suitable for high safety requirements [5]. - Disadvantages consist of high development costs and potential difficulties in handling complex scenarios without suitable rules [5]. Complementary Relationship - The analysis indicates that end-to-end systems require PNC for certain scenarios, while PNC can benefit from the efficiencies of end-to-end approaches. This suggests a complementary rather than adversarial relationship between the two methodologies [6]. Job Opportunities - The article highlights job openings in both end-to-end and traditional PNC roles, indicating a demand for skilled professionals in these areas with competitive salaries ranging from 30k to 100k per month depending on the position and location [8][10][12][14].

Core Insights - The article discusses the results of the 2025 Waymo Open Dataset End-to-End Driving Challenge, highlighting the advancements in end-to-end autonomous driving systems and the shift towards using large-scale public datasets for training models [2][18]. Group 1: Competition Results - The champion of the competition was the EPFL team, which utilized the DiffusionDrive model, nuPlan data, and an ensembling strategy [1]. - The runner-up was a collaboration between Nvidia and Tubingen teams, which also referenced DiffusionDrive and SmartRefine, employing multiple datasets to demonstrate the importance of training data quality [1][22]. - The third place was secured by Hanyang University from South Korea, which focused on a simplified structure using only front-view input and vehicle state [1][3]. Group 2: Methodology - The UniPlan framework was introduced, leveraging large-scale public driving datasets to enhance generalization in rare long-tail scenarios, achieving competitive results without relying on expensive multimodal large language models [3][18]. - The model architecture is based on DiffusionDrive, which employs a truncated diffusion strategy for efficient and diverse trajectory generation [4][6]. - The diffusion decoder utilizes a cross-attention mechanism to refine trajectory predictions based on scene context [5][6]. Group 3: Data Processing - The nuPlan dataset was processed to create a diverse training set, resulting in 90,000 samples by applying a sliding window approach [7]. - A similar filtering strategy was used for the WOD-E2E dataset, generating 35,000 training samples and 10,000 validation samples [8]. - The model was trained on a powerful computing setup with four H100 GPUs, achieving significant training efficiency [10]. Group 4: Experimental Results - The performance was evaluated using Rater Feedback Score (RFS) and Average Displacement Error (ADE), with various configurations tested [12][17]. - The results indicated that the combined training of WOD-E2E and nuPlan datasets led to slight improvements in average RFS, particularly in long-tail categories [23]. - The analysis showed that while additional datasets generally provide benefits, the quality of the data sources is more critical than quantity [39]. Group 5: Conclusion - The article emphasizes the potential of data-centric approaches in enhancing the robustness of autonomous driving systems, as demonstrated by the competitive results achieved with the UniPlan framework [18][39].

Investment Rating - The report assigns a "Buy" rating for Horizon Robotics (9660.HK) [3] Core Insights - Horizon Robotics is positioned as a leader in the new generation of automotive intelligent chips and a world-class AI algorithm company, focusing on software-defined principles and exploring new boundaries in intelligent driving [5][14] - The intelligent driving market is expected to grow significantly, with the AD market projected to take over from ADAS as the main growth driver, achieving a market size of 407 billion yuan by 2030 [12][37] - The company has a leading market share in the intelligent driving computing solutions market, with a 28.65% share in the first half of 2024, and is expected to further increase its share in the OEM ADAS and AD markets [11][57] Summary by Sections Company Overview - Horizon Robotics focuses on intelligent driving chip platforms, full-scene intelligent driving solutions, and supporting toolchains, establishing itself as a comprehensive supplier in the industry [5][14] - The company has launched several intelligent driving chips, including J2, J3, J5, and J6, and has developed a self-adaptive BPU computing unit that maximizes computational efficiency [14] Market Growth - The AD+ADAS market in China has seen a compound annual growth rate (CAGR) of 57.8% from 2019 to 2023, with the AD market growing at a CAGR of 144.2% [12][37] - By 2030, the AD market is expected to reach a size of 407 billion yuan, with a CAGR of 48.8% from 2025 to 2030 [12][37] Competitive Position - Horizon Robotics has a steadily increasing market share, with 41% in the ADAS market and over 30% in the AD market among Chinese OEMs by the end of 2024 [12][57] - The company has established partnerships with major OEMs, including BYD, Geely, and Chery, to support their intelligent driving strategies [61][69] Financial Projections - Revenue projections for Horizon Robotics are expected to reach 36.10 billion yuan in 2025, 56.97 billion yuan in 2026, and 80.53 billion yuan in 2027, with corresponding growth rates of 51%, 58%, and 41% respectively [6] - The company anticipates a return to profitability by 2027, with a projected net profit of 668 million yuan [6] Customer Base and Partnerships - Horizon Robotics has a broad customer base, covering major domestic automakers and new energy vehicle manufacturers, which positions it well for future growth as the demand for intelligent driving solutions increases [69]

Core Viewpoint - The report from Shenwan Hongyuan indicates that SUTENG JUCHUANG (02498) is expected to experience significant revenue growth from 2025 to 2027, with projected revenues of 2.62 billion, 3.66 billion, and 4.70 billion yuan respectively, while the net profit is forecasted to be -238 million, 106 million, and 320 million yuan respectively. The company is currently not profitable, leading to a PS valuation method being employed for its assessment [1]. Group 1 - The company is rapidly leading the global LiDAR industry, focusing on providing quality solutions in the field of embodied intelligence. The sales of LiDAR products have seen a non-linear high growth, confirming the explosive demand from automotive companies for LiDAR configurations under the trend of increasing intelligence [2]. - In 2024, the total sales of LiDAR products are expected to reach approximately 544,000 units, representing a significant year-on-year increase of 109.6%. The sales of LiDAR products for ADAS applications are projected to be around 520,000 units. The company is expected to maintain a leading market share of 26% in 2024, ranking first globally [2]. - The product matrix of the company is comprehensive, covering various technical paths including mechanical, semi-solid, and solid-state LiDAR, with performance ranging from short to ultra-long distances and low to high beam configurations. This allows the company to meet a wide range of demands across different price segments [2]. Group 2 - The first driving force is the end-to-end vehicle integration and equalization of intelligent driving. The previous debate over LiDAR configurations in vehicles has been influenced by Tesla's insistence on a pure vision and neural network approach. With advancements in computing power and the maturity of end-to-end algorithms, the integration of multi-sensor fusion with pure vision is becoming more feasible [3]. - The LiDAR industry is expected to enter the "thousand-yuan machine era" by 2025, with prices dropping to the range of 25,000 to 30,000 yuan. This price reduction is anticipated to significantly increase the configuration rate of LiDAR as an "invisible safety airbag" for autonomous driving [3]. - The global market for LiDAR in passenger vehicles is estimated to reach approximately 7 billion yuan by 2025, with the Chinese market accounting for about 6.3 billion yuan. The overseas market is expected to gradually open up and grow rapidly, representing an important direction for LiDAR's incremental growth [3]. Group 3 - The second driving force is the strategic positioning of the robotics technology platform. The company focuses on the development of incremental components such as robotic vision and dexterous hands, launching solutions based on hand-eye coordination for upper body operations and lower body mobility [4]. - The year 2025 is viewed as the year of mass production for humanoid robots, with companies like Tesla aiming to produce 5,000 units of Optimus this year, and domestic companies like Zhiyuan Robotics achieving deliveries in the thousands [4]. - In the niche market of robotic lawn mowers, the demand for LiDAR products is projected to exceed 400,000 units by 2025 and is expected to surpass 900,000 units by 2028 [4].