11月5日，电厂获悉，小鹏汽车在"2025小鹏科技日"上发布了第二代VLA——区别于传统VLA模型的"视觉-语言-动 作"架构，小鹏的第二代VLA无需语言转译，实现从视觉信号到动作指令的端到端直接输出。 第二代VLA也是小鹏首个量产物理世界大模型，它既是动作生成模型，也是理解和推演的物理世界模型，能够为AI汽 车、人形机器人、 飞行汽车提供了统一的智能基座。小鹏汽车成功在2250TOPS的Ultra版车型上搭载数十亿级参数规 模的第二代VLA。同时，用于第二代VLA的数据训练量接近1亿clips。 基于第二代VLA，小鹏发布了"小路 NGP"， 把复杂小路的平均接管里程提升13倍；无导航自动辅助驾驶Super LCC和 人机共驾可以不依赖导航在全球范围开启，展现模型的泛化学习与智能水平。 小鹏第二代VLA将于 2025 年 12 月启动先锋共创体验，2026 年一季度随 Ultra 车型全量推送。何小鹏还宣布大众汽车 成为小鹏第二代VLA首发战略合作伙伴。同时，小鹏图灵AI芯片已获得大众汽车定点。 在第二代VLA的支持下，小鹏汽车还计划将于2026年推出三款Robotaxi车型、具备L4级体验的有人驾驶Rob ...

活动上，小鹏汽车董事长、CEO何小鹏展示了全新的"汇天A868飞行汽车"。 图片来源：直播截图 每经编辑｜金冥羽 据悉，该款飞行汽车设计目标续航里程超过500公里，这使其具备了执行城市间航线的潜力，而不仅仅是市内短途交通。设计巡航速度超过360公里/小 时，远超地面交通，能极大缩短出行时间。 11月5日，小鹏科技日（2025 AI DAY）正式开幕。 据透露，这一代IRON是预研的最后一个版本，量产版本将不高于170cm。据介绍，新一代人形机器人IRON首发应用全固态电池，并搭载3颗图灵芯片和 物理世界大模型。 何小鹏表示，小鹏汇天陆地航母全球订单目前已经7000单，并有望创造载人飞行器年销量世界纪录。 何小鹏还提到，小鹏飞行汽车会执行"高管先飞计划"，汇天管理团队亲身验证飞行安全量产交付前，管理者飞行里程必须超过5000公里。 图片来源：直播截图 此外，小鹏汽车宣布，将在2026年推出3款Robotaxi（无人驾驶出租车）车型，并于2026年开启试运营。 图片来源：直播截图 何小鹏表示，整车厂自己投入做Robotaxi，更有利于推动该产业的规模化与商业化，同时小鹏将自研软件、自制硬件、开放SDK等。据规划， ...

Core Viewpoint - The article discusses the recent funding and technological advancements of Hangzhou Yingshen Intelligent Technology Co., Ltd., focusing on its development of embodied intelligence and robotics, aiming to provide low-cost, high-reliability solutions for various industries [4][9]. Group 1: Company Overview - Yingshen Intelligent was established in 2024 and specializes in the research and application of embodied intelligence technology, utilizing its self-developed spatial large model and industrial scene robots [4]. - The company has recently completed several rounds of seed financing, amounting to several million yuan, which will be used for the development and commercialization of its robotic technology [4]. - The founder and CEO, Min Wei, has a background in leading Alibaba's robotics team and has experience in deploying robots in various scenarios such as buildings, hospitals, and hotels [4]. Group 2: Technological Advancements - The advancement of embodied intelligence is crucial for robots to perform complex tasks in dynamic environments, requiring enhanced perception, decision-making, and execution capabilities [5]. - Yingshen Intelligent has developed a spatiotemporal intelligent large model that constructs a four-dimensional model of the real world, enabling robots to understand and map the physical world effectively [6]. - The company aims to reduce reliance on human language for robot instruction by directly modeling video data, which enhances the accuracy and efficiency of the robots [6]. Group 3: Data Utilization and Training - Yingshen Intelligent leverages vast amounts of domestic video data to keep training costs low, using strategically placed cameras to capture worker movements for 3D spatial modeling and action generation [7]. - The training process involves capturing joint positions and simulating worker perspectives, which simplifies the training workflow and minimizes disruption to normal production [7]. Group 4: Market Strategy and Future Plans - The company has launched the "Yingshen" series of industrial robots, which are designed to operate stably across various working conditions and are expected to deliver over a hundred robots by 2025 [8]. - Yingshen Intelligent is focusing on developing the robotic brain to enhance the robots' understanding of the external world and task execution capabilities, aiming for the widespread adoption of Level 4 embodied intelligence in daily production and life [8].