原文链接： https://zhuanlan.zhihu.com/p/1981658979764568316 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30个 方向 学习 路线 作者 | 高毅鹏@知乎 编辑 | 自动驾驶之心 >>自动驾驶前沿信息获取 → 自动驾驶之心知识星球 本文已获转载授权，转载请联系原文作者 百度 感知大模型 规划大模型 端到端联合训练 感知网络+决策规划网络一起进行联合训练，意味着从原始的传感器时序数据的捕捉，到最终的油门和转向输出驱动，从一头到另一个尽头一起训练的方式。这种联 合方式将确保中间态数据会以模型自己可以理解的方式（所谓隐态/隐式数据空间）从感知部分带着最佳信息丰度和准确度，进入决策规划网络，从而完成对于油门 和转向的控制。 但先于联合训练的方式，我想分模型（感知和规划两大部分）的训练和调整依然是必要的。这样可以确保即便是端到端模型，我们也可以在感知输出隐态数据时，可 以观察到这些数据是否符合人类对于感知结果的判断标准。如果只是考虑从感知到决策规划的端到端，不考虑对于中间态数据的关注，那么感知网络中的 Decoder 就 没有必要存在了（图 ...

Group 1: Core Insights - The 2025 World Artificial Intelligence Conference showcased significant advancements in humanoid robots, with over 150 humanoid robots displayed, marking the largest concentration of such technology in China to date [3][4]. - The exhibition featured a record-breaking area of over 70,000 square meters, with participation from over 800 companies and the unveiling of more than 3,000 cutting-edge exhibits, including over 60 types of intelligent robots [4][5]. Group 2: Technological Advancements - Humanoid robots are increasingly autonomous due to advancements in large model capabilities, such as the Lingxi X2, which can understand complex queries and make decisions based on its battery status [5][6]. - Companies like Tencent are developing embodied large models to enhance robot intelligence, enabling them to understand their environment and perform complex tasks [6]. - Improvements in motion capabilities have been noted, with robots like the G1 combat robot demonstrating advanced agility and balance [6][7]. - The integration of cross-modal sensing and fine manipulation has been enhanced, allowing robots to perform intricate tasks like egg carving and collaborative work with other robots [7]. Group 3: Commercialization Progress - The application scenarios for humanoid robots are expanding, with developments in industrial, medical, and commercial service sectors, such as logistics and rehabilitation [8][9]. - Several companies have announced mass production plans for humanoid robots, indicating a shift from technology validation to large-scale production [9][11]. - Significant commercial orders have been secured, including a notable 120 million yuan project for humanoid robot manufacturing, highlighting the commercial viability of these technologies [12]. Group 4: Investment Dynamics - The humanoid robot sector has seen over 80 financing cases since 2025, with total investments exceeding 10 billion yuan, primarily in early-stage funding [15][17]. - Diverse capital participation includes strategic investments from industry leaders and local government funds aimed at fostering regional industrial clusters [15][17]. - The IPO landscape is becoming more active, with companies like Yushutech initiating IPO processes, which is expected to attract further investment into the humanoid robot sector [15][17].