Core Insights - The rapid commercialization of Vision-Language-Action (VLA) models is redefining the technical threshold for advanced intelligent driving [1] - The competition surrounding VLA will significantly influence the future competitive landscape of Chinese automotive companies and may lead to a reshuffling of the entire intelligent driving industry [7] Group 1: VLA Model Development - Li Auto has launched its VLA driver model with the flagship electric vehicle i8, while Yuanrong Qixing released its self-developed VLA model, DeepRoute IO 2.0, on August 26, covering approximately 200,000 vehicles [2] - XPeng Motors introduced a new generation VLA architecture at the P7 launch on August 27, claiming a latency of less than 100 ms and a planning frame rate of 20 Hz, setting a new benchmark for mass production [3] - The VLA model's ability to abstract and categorize real-world scenarios through language enhances its generalization capabilities compared to traditional end-to-end models [8][12] Group 2: Technical and Economic Challenges - The VLA model requires significant computational power, with Li Auto and XPeng utilizing cloud clusters of 13 EFLOPS and 8 EFLOPS, respectively, while many smaller companies are limited to 0.2-0.6 EFLOPS [14] - The data requirements for VLA are substantial, necessitating the collection and annotation of visual-language-action triplets, with 90% of the training data sourced from 2.93 billion kilometers of real vehicle logs [13] - The cost of training a VLA model can reach 12-15 million RMB per session, which is a significant portion of the annual R&D budget for smaller companies [15] Group 3: Industry Restructuring - The high costs associated with VLA models create survival challenges for smaller automotive companies, which may struggle to compete against larger firms with established technological advantages [19][29] - The transition from rule-based algorithms to VLA models requires a gradual and systematic approach, making it difficult for many second-tier brands to replicate the success of leading companies [21][23] - The VLA model's emergence may lead to a significant industry reshuffle, with many mid-tier companies potentially becoming "outsourcing providers" or low-end manufacturers [24][30] Group 4: Competitive Landscape - The VLA model's introduction is expected to alter the competitive dynamics, with companies like Huawei and Momenta currently holding a dominant market share in intelligent driving [45] - The VLA model's multi-modal learning and reasoning capabilities allow companies like Li Auto and XPeng to achieve performance levels close to those of larger competitors in long-tail scenarios [48] - The year 2025 could mark a pivotal moment for both leading companies and VLA practitioners, potentially leading to a reversal of fortunes in the intelligent driving market [51]

Core Viewpoint - The interview with Dr. Luo Jianlan emphasizes the importance of real-world data in the development of embodied intelligence, highlighting the challenges and strategies in data collection, model training, and application deployment. Data Discussion - The company collaborates with multiple sensor suppliers focusing on the joint development of visual, tactile, and high-density sensors, while building a cross-platform data collection API for standardized data input [2] - Achieving a high performance rate of 95% for robots in real-world applications remains a significant challenge, particularly in household tasks [2] - The company uses 100% real machine data for training multimodal large models, agreeing with the notion that simulation environments have scalability limitations [2][3] - The cost of collecting real-world data is not the main issue; rather, the lack of standardized mechanisms for data collection is a core challenge [6] - The company acknowledges the data scarcity and performance optimization difficulties in both autonomous driving and robotics, emphasizing the need for high success rates in open environments [7] Evaluation of Embodied Large Models - There is currently no universal benchmark for evaluating embodied intelligence models due to significant differences in software and hardware environments across companies [9] - The evaluation of different large models is primarily based on their technical routes and the challenges they face in the current landscape [9][10] - The company aims to establish a unified real-machine testing platform to facilitate model evaluation across different scenarios [9] Embodied Intelligence Applications and Implementation - The deployment process for robots involves four steps: task modeling, scene migration, scene adaptation, and safety verification, emphasizing the importance of hardware-software collaboration [18] - High success rates are crucial, but challenges in generalization, robustness, and real-time performance must also be addressed [20] - Industrial environments are seen as the most promising for the initial large-scale deployment of embodied intelligence due to their structured nature and clear commercial demands [21] Future Outlook for Embodied Intelligence - The company aims for a "DeepSeek moment," focusing on achieving near 100% success rates and high-speed execution capabilities in future models [24] - The transition to a data-driven paradigm is recognized as a significant shift in the field, moving away from traditional hypothesis-driven approaches [25] - The potential of brain-like architectures is acknowledged, with ongoing exploration to combine computation with physical capabilities for future intelligent systems [26]

以下文章来源于智能涌现 ，作者邱晓芬 苏建勋 智能涌现 . 直击AI新时代下涌现的产业革命。36氪旗下账号。 "同时做大脑和本体，看起来可能会非常难，但对我来说，因为我都能做，所以这是一个自然选择。" 文 ｜ 邱晓芬 苏建勋 编辑 ｜ 苏建勋 来源｜ 智能涌现（ID：AIEmergence） 封面来源 ｜ 企业官方 "外界对我们的认知，和我们实际的业务状况，确实存在一定差距。" 在"星动纪元"的北京办公室中，创始人陈建宇对"智能涌现"表示。 "星动纪元"成立于2023年8月，由清华大学交叉信息研究院助理教授陈建宇创办。2025年7月7日，"星动纪元"宣布完成近5亿元A轮融资，由鼎晖CGV资本和 海尔资本联合领投，厚雪资本、华映资本、襄禾资本、丰立智能等跟投，老股东清流资本、清控基金等继续追加投资。 尽管成立至今不过两年，在机器人硬件业务上，"星动纪元"接连发布了灵巧手、轮式、全尺寸人形等产品，这些动向， 让不少人误将星动纪元视作一家机 器人本体公司，甚至"觉得我们是一家灵巧手公司 "。 这不是陈建宇希望公司被贴上的标签。 做一款通用、智能的机器人，是陈建宇在近十年前看到AlphaGo时就定下的目标，这意味着机 ...