Core Viewpoint - Li Auto is set to launch a dual-wheeled robot within this year, primarily for factory manufacturing scenarios, indicating the company's expansion into robotics technology [1] Group 1: Company Developments - Li Auto's stock price increased by 2.97%, reaching HKD 67.55, with a trading volume of HKD 238 million [1] - The humanoid robot team at Li Auto has been in existence since April 2025 and has been secretly developing for nearly a year [1] - The internal project, codenamed Nexus, is led by He Junpei, a hardware partner from the former robotics startup Jiuguang Intelligent, with a team of fewer than 30 people [1] Group 2: Product Information - Li Auto plans to develop two products under the humanoid robot team: a dual-wheeled robot and a bipedal robot [1] - The dual-wheeled robot is reportedly ready for launch and is expected to be released by mid-year [1]

Core Viewpoint - Li Auto-W (02015) has seen a stock price increase of nearly 3%, currently trading at 67.25 HKD with a transaction volume of 2.22 billion HKD, following the announcement of a new dual-wheeled robot set to be released within the year [1] Group 1: Company Developments - Li Auto has been secretly developing a humanoid robot team since April 2025, with the internal project code-named "Nexus" [1] - The humanoid robot team consists of fewer than 30 members and is led by He Junpei, a hardware partner from the robotics startup Jiuguang Intelligent [1] - The company plans to launch two products under the humanoid robot team: a dual-wheeled robot and a bipedal robot, with the dual-wheeled robot ready for release by mid-year [1] Group 2: Market Impact - The announcement of the dual-wheeled robot has positively influenced Li Auto's stock performance, reflecting investor interest in the company's expansion into robotics [1] - The expected application of the dual-wheeled robot is primarily in factory manufacturing scenarios, indicating a strategic move towards automation in production [1]

Core Viewpoint - Li Auto's stock has seen a nearly 3% increase, currently trading at 67.25 HKD, with a transaction volume of 2.22 billion HKD, following news of its upcoming dual-wheeled robot launch aimed at factory manufacturing scenarios [1] Group 1: Company Developments - Li Auto plans to release a dual-wheeled robot within this year, which has been in secret development for nearly a year [1] - The humanoid robot team at Li Auto, known internally as Nexus, was established in April 2025 and consists of fewer than 30 members [1] - The team is led by He Junpei, a hardware partner from the former robotics startup Jiuguang Intelligent [1] Group 2: Product Details - Li Auto is developing two products under its humanoid robot team: a dual-wheeled robot and a bipedal robot [1] - The dual-wheeled robot is reportedly ready for launch and is expected to be unveiled by mid-year [1]

本月底法定/註冊股本總額： USD 500,000 FF301 股份發行人及根據《上市規則》第十九B章上市的香港預託證券發行人的證券變動月報表 截至月份: 2026年2月28日 狀態: 新提交 致：香港交易及結算所有限公司 公司名稱: 理想汽車 呈交日期: 2026年3月5日 I. 法定/註冊股本變動 | 1. 股份分類 | 不同投票權架構公司普通股 | 股份類別 | A | | | 於香港聯交所上市 (註1) | | 是 | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 證券代號 (如上市) | 02015 | 說明 | | | | | | | | | | | 法定/註冊股份數目 | | | 面值 | | | 法定/註冊股本 | | | 上月底結存 | | | 4,500,000,000 | USD | | 0.0001 | USD | | 450,000 | | 增加 / 減少 (-) | | | | | | | USD | | | | 本月底結存 | | | 4,500,000,000 | USD | | 0.0001 ...

Core Viewpoint - The article discusses the fundamental changes in the approach to VLA (Vision-Language-Action) processing, highlighting the differences between the LinkVLA paper and previous presentations by Jia Peng, emphasizing a shift from LLM output results to native language manipulation of physical space [1][2]. Group 1: Tokenization and Action Representation - The action token in Jia Peng's version is based on high-dimensional environmental features, requiring diffusion as a translator to generate corresponding trajectories, emphasizing the model's understanding of the 3D environment [3]. - In contrast, the LinkVLA version uses discretized BEV (Bird's Eye View) space coordinates, where each action token corresponds to a unique grid coordinate, simplifying the output to a sequence of position tokens while retaining environmental understanding in the LLM's hidden layers [3][4]. Group 2: Output Mechanism and Precision - Jia Peng's version employs parallel decoding, outputting all action tokens at once and using diffusion for iterative sampling, while LinkVLA adopts a two-step tokenization process that first predicts an endpoint token and then a set of residual tokens for coordinate correction, enhancing trajectory precision and reducing latency [5]. - The tokenization approach in LinkVLA features dense grids for nearby areas and sparse grids for distant areas, addressing the precision issues of traditional uniform grids in near-field control [5]. Group 3: Alignment and Understanding - Jia Peng's VLA aligns driving preferences through RLHF (Reinforcement Learning from Human Feedback), focusing on fine-tuning the model's output [6]. - The LinkVLA version introduces a training task for action understanding, requiring the model to generate trajectories based on instructions and translate them into textual descriptions, thereby addressing the semantic gap and ensuring the model comprehends the meaning of actions [7].

Investment Recommendations - The current investment suggestion for the smart automotive sector is to maintain a strong outlook on the L4 RoboX theme for 2026, favoring B-end software over C-end hardware [2][6] - Preferred H-shares include Xpeng Motors, Horizon Robotics, Pony.ai, WeRide, Cao Cao Mobility, and Black Sesame Intelligence; A-shares include Qianli Technology, Desay SV, and Jingwei Hirain [2][6] Downstream Application Dimensions - Robotaxi perspective includes: 1. Integrated model: Tesla and Xpeng Motors 2. Technology providers + operational sharing model: Horizon, Baidu, Pony.ai, WeRide, and Qianli Technology 3. Transformation of ride-hailing/taxi services: Didi, Cao Cao Mobility, Ruqi Mobility, public transport, and Jinjiang Online [2][6] - Robovan perspective includes Desay SV and Jiushi Intelligent/New Stone Technology [2][6] - Other autonomous vehicle perspectives include mining trucks (e.g., HiDi Intelligent Driving), ports (e.g., Jingwei Hirain), sanitation vehicles (e.g., Yingfeng Environment), and buses (e.g., WeRide) [2][6] Upstream Supply Chain Dimensions - B-end autonomous vehicle OEMs include BAIC BluePark, GAC Group, Jiangling Motors, and Tongli Co. - Key upstream suppliers include: 1. Testing services (China Automotive Research and China Automotive Industry Corporation) 2. Chips (Horizon Robotics and Black Sesame Intelligence) 3. Domain controllers (Desay SV, Jingwei Hirain, Joyson Electronics, Huayang Group, and Coboda) 4. Sensors (Sunny Optical Technology, Hesai, and Suteng Juchuang) 5. Steer-by-wire chassis (Bertel and Nexperia) 6. Lighting (Xingyu Co.) 7. Glass (Fuyao Glass) [2][6] Mainstream Automakers' Autonomous Driving Strategies - A detailed comparison of major domestic automakers' autonomous driving strategies shows various approaches, including self-research and external supply partnerships, with notable collaborations with companies like Huawei, Momenta, and Horizon [4][5][20][21][30][32][39][40] - Chery's strategy includes a mixed model of multiple external algorithm suppliers and self-research platforms, with significant investments in partnerships [28][29] - Geely's integration of its autonomous driving team into Qianli Technology aims to streamline operations and enhance technological capabilities [20][22][23] BYD's Autonomous Driving Development - BYD's "Tianshen Eye" system has evolved to version 5.0, featuring advanced capabilities such as automatic emergency steering and braking, with a focus on enhancing user safety and efficiency [15][16] - The company emphasizes a dual approach of self-research and external collaboration, maintaining a significant investment in autonomous driving technology [10][14][47] Xiaomi's Strategic Investment in Autonomous Driving - Xiaomi has adopted a phased approach to its autonomous driving strategy, transitioning from strategic investments to full-scale self-research and development, with a significant increase in team size and R&D investment [47][48]

Investment Rating - The report maintains a positive outlook on the smart automotive sector, particularly emphasizing the L4 RoboX theme for 2026 [4] Core Insights - The report suggests a preference for B-end software companies over C-end hardware companies, recommending specific stocks in both H-shares and A-shares [4] - It highlights various downstream application perspectives, including Robotaxi and Robovan, and identifies key players and their business models [4] - The report also discusses upstream supply chain opportunities, including core suppliers and manufacturing partners [4] Summary by Sections Mainstream Automotive Companies' Smart Driving Technology Solutions - The report provides a comprehensive overview of the smart driving strategies of major automotive companies, detailing their partnerships and technology approaches [5][6][7][15][22][24][30][33] - Companies like BYD, Geely, Chery, and Great Wall are noted for their mixed strategies of self-research and external collaboration, with specific technology and supplier partnerships outlined [7][15][22][24][30][33] BYD's Smart Driving Strategy - BYD has shifted its smart driving approach from standard configuration to a pay-per-use model, emphasizing self-research while maintaining partnerships with algorithm companies [7][8] - The company has launched the "Tianshen Eye 5.0" system, which features advanced capabilities such as emergency steering and obstacle avoidance [12][13] Geely's Smart Driving Team Integration - Geely has completed the integration of its smart driving team under the "Qianli Zhijia" brand, focusing on enhancing its autonomous driving capabilities [15][17][19] - The company has established a structured approach to its smart driving solutions, offering multiple versions with varying hardware and software capabilities [19] Chery's Smart Driving Development - Chery has introduced the "Falcon Smart Driving" strategy, which includes multiple versions of its smart driving system, aiming for comprehensive coverage across various scenarios [22][23] - The company has also consolidated its smart driving R&D teams to enhance efficiency and innovation [22][23] Great Wall's Smart Driving Solutions - Great Wall has adopted a dual approach of self-research and external collaboration, with a focus on enhancing its computing power and algorithm capabilities [26][29] - The company has developed a tiered computing platform to support various levels of autonomous driving features [26][29] Changan's Smart Driving Framework - Changan has implemented a strategy that combines procurement from Huawei with its own smart driving research, aiming for a comprehensive autonomous driving solution [32][33] Other Companies' Strategies - The report also covers the smart driving strategies of other companies such as SAIC, GAC, and Leap Motor, highlighting their partnerships and technological advancements [33][36][38]

Core Viewpoint - The article discusses the core obstacles in the implementation of VLA (Vision-Language Action) systems, emphasizing that existing solutions treat alignment as a defect to be patched rather than a structural issue to be eliminated from the architecture [1]. Group 1: Shared Codebook - LinkVLA proposes a Shared Codebook that eliminates the need for translation between human language and vehicle action coordinates, addressing the inherent loss in translation without direct supervision [2][3]. - By transforming continuous trajectory coordinates into discrete action tokens and merging them with language tokens, LinkVLA creates a unified representation that removes the modal gap at a structural level [3]. Group 2: Action Understanding Objective - LinkVLA introduces an Action Understanding Objective that requires the model to not only generate trajectories from language commands but also to reverse-engineer language descriptions from existing trajectories, enhancing the model's reliability [4]. - The dual-task approach significantly improves performance metrics, with the average success rate increasing from 81.63% to 87.16% and lane change success rate rising from 88.49% to 97.42% [4]. Group 3: C2F Architecture - The Coarse-to-Fine (C2F) architecture in LinkVLA reduces the inference time from 361ms to 48ms by compressing the serial dependency of trajectory generation into two steps, thus enhancing real-time performance [5][6]. - This architecture not only improves efficiency but also maintains accuracy, with driving scores increasing from 90.66 to 91.01, demonstrating a simultaneous enhancement in speed and precision [6]. Group 4: Systematic Reconstruction - The contributions of Shared Codebook, Action Understanding, and C2F collectively represent a systematic reconstruction of the underlying architecture of language-action models, rather than mere local optimizations [7].

Core Insights - The article discusses a fundamental paradox in the automotive industry regarding the relationship between computing power and actual performance, questioning whether higher computing power truly translates to better efficiency [1][2][10] - It highlights the introduction of the "Soft-Hardware Collaborative Design Law" by Li Auto, which represents a significant technological breakthrough and a paradigm shift in AI foundational logic [1][4] Group 1: Computing Power Paradox - The automotive industry's past decade has been characterized by a worship of computing power, with metrics like TOPS becoming more fashionable than horsepower [2] - Li Auto's early experiences with top-tier vehicle chips revealed that the actual performance often falls short of expectations, even with high-end hardware [3][4] - Major tech companies, including NVIDIA and Apple, face similar challenges due to a traditional development model that separates hardware and software, leading to wasted computing power and inefficiencies [4] Group 2: Collaborative Design Framework - Li Auto's MindVLA team, in collaboration with the National Innovation Decision Intelligence Research Institute, developed a mathematical framework to optimize the collaboration between chips and algorithms [4][5] - This framework combines loss function expansion and Roofline performance modeling, allowing for a quantifiable and predictable approach to soft-hardware collaboration [5] - Key findings from this research indicate that optimal chip architecture is highly dependent on specific hardware parameters, emphasizing the necessity for algorithms to define chip design [5][6] Group 3: Practical Applications and Industry Impact - The launch of the new Li Auto L9, equipped with the Mach 100 chip, showcases a significant increase in effective computing power, achieving three to six times the effective performance of competitors [7][8] - Li Auto's commitment to R&D is evident, with projected investments reaching 12 billion yuan by 2025, and a cumulative R&D expenditure exceeding 46.8 billion yuan over eight years [8] - The company has published nearly 50 papers in relevant fields, contributing to a growing body of knowledge and open-source projects that foster a healthy technological ecosystem in the Chinese smart driving industry [8][9] Group 4: Open Source and Industry Collaboration - Li Auto's open-source initiative, including the Star Ring OS, aims to reduce redundant R&D costs across the automotive industry, potentially saving 10 to 20 billion yuan annually [9] - This collaborative approach reflects a strategic shift towards shared innovation, recognizing that no single company can monopolize all advancements in the complex smart vehicle sector [9] - The narrative of global AI competition is being reshaped as Chinese companies begin to contribute foundational methodologies and infrastructure, positioning themselves as co-creators of industry standards [9][10] Group 5: Conclusion and Future Implications - The article concludes that while computing power is essential, the true determinant of performance lies in the collaboration between hardware and software [10][11] - The Soft-Hardware Collaborative Design Law not only addresses current challenges in smart driving but also lays a theoretical foundation for future AI applications, such as embodied intelligence and spatial robotics [11] - The transition from follower to definitional leader in the industry signifies a profound shift in how Chinese tech companies approach innovation, showcasing their capability to contribute significantly to global AI discourse [11]

Group 1 - The Hong Kong stock market for electric vehicle companies experienced a decline in the afternoon session, with significant drops in share prices for several key players [1] - Leap Motor (09863.HK) and Xpeng Motors (09868.HK) both fell over 5%, while NIO-SW (09866.HK) decreased by 4.7% [1] - Xiaomi Group (01810.HK) and GAC Group (02238.HK) saw a decline of 3.3%, and Li Auto (02015.HK) dropped by 1.7% [1]