10月10日，鹿明机器人宣布与面壁智能达成战略合作，双方围绕具身智能的数据采集、模型训练与场景应用展开合作，共同推动具身智能从技术验证走向规模化产业落地。双方将充分发挥各自在高效基座大模型、具身本体与产业场景落地等方面的优势。
On October 10, Luming Robotics announced a strategic partnership with ModelBest, focusing on data collection, model training, and scenario applications for embodied intelligence, jointly advancing embodied intelligence from technical validation to scaled industrial deployment. The two parties will leverage their respective advantages in efficient foundation models, embodied bodies, and industrial scenario deployment.
这个合作的分工很清晰。面壁智能做的是高效基座大模型，它的技术积累在模型架构和训练效率上。鹿明机器人做的是具身本体，也就是机器人的身体和运动能力。双方合作要解决的问题，是具身智能从技术验证走向规模化落地。前面讨论过很多具身智能公司，有的自己做全栈，比如自变量机器人从模型到数据采集方案都自己做；有的做垂直场景，比如巨蟹智能专注关节零部件。鹿明机器人和面壁智能走的是第三条路：模型公司和本体公司合作，各自做自己擅长的事。
The division of labor in this partnership is clear. ModelBest works on efficient foundation models, with technical accumulation in model architecture and training efficiency. Luming Robotics works on embodied bodies—the robot's physical form and motion capabilities. The problem they aim to solve together is embodied intelligence moving from technical validation to scaled deployment. Earlier we discussed many embodied intelligence companies: some do full-stack themselves, like AgiBot doing everything from models to data collection solutions; some focus on vertical scenarios, like Juxie Intelligent specializing in joint components. Luming Robotics and ModelBest take a third path: a model company and a body company partnering, each doing what they are good at.
这个模式的关键在于，具身智能的产业链正在分化。一两年前，一家公司做具身智能，通常是从模型到硬件全都自己做，因为市面上没有现成的、能用的基础模型和本体。现在情况变了。面壁智能这样的公司，专攻基座大模型，在高效训练和推理上有积累，可以给多家机器人公司提供模型能力。鹿明机器人这样的公司，专攻具身本体，把机器人的运动控制和硬件做好，可以接多家模型公司的模型。分工出现的前提，是每一环都有公司能做到可用的水平，别人用你的比自己从头做更划算。
The key to this model is that the embodied intelligence supply chain is differentiating. A year or two ago, a company doing embodied intelligence typically did everything from models to hardware itself, because there were no ready-to-use foundation models and bodies on the market. Now things have changed. Companies like ModelBest specialize in foundation models, with accumulation in efficient training and inference, and can provide model capabilities to multiple robotics companies. Companies like Luming Robotics specialize in embodied bodies, doing robot motion control and hardware well, and can take models from multiple model companies. The premise for division of labor appearing is that each link has companies that can reach a usable level, and using theirs is more cost-effective than building from scratch.
前面讨论过自变量机器人的WALL-B，是全球首个将视觉、语言、触觉、动作、物理预测整合进同一神经网络的具身大模型。自变量机器人选择自己做全栈，因为它有王潜、王昊这样的大模型人才，也有资本支持它烧钱做基础模型。鹿明机器人没有走这条路，而是选择和面壁智能合作。这两条路各有代价：全栈模式的技术壁垒高，但投入大、周期长；合作模式的起步快，但对合作伙伴的依赖度高，模型能力的上限受制于面壁智能。鹿明机器人选合作，说明它对自身的定位很清楚：做好本体和场景落地，模型的事交给更专业的公司。
Earlier we discussed AgiBot's WALL-B, the world's first embodied foundation model integrating vision, language, touch, action, and physical prediction into the same neural network. AgiBot chose full-stack because it has large-model talent like Wang Qian and Wang Hao, and capital support to burn money on foundation models. Luming Robotics did not take this path, choosing instead to partner with ModelBest. Each path has costs: the full-stack model has high technical barriers but large investment and long cycles; the partnership model starts fast but depends heavily on partners, with model capability ceiling constrained by ModelBest. Luming Robotics choosing partnership indicates it has a clear positioning: do the body and scenario deployment well, and leave models to more specialized companies.
这个合作的另一个看点是数据采集。双方合作的内容里明确包括了数据采集。前面讨论过机器人数据采集员的工作，也讨论过墨现科技的触觉传感器和数采手套。具身智能的模型训练需要大量真实场景的数据，而数据采集是产业链里最苦最累的一环。鹿明机器人和面壁智能合作做数据采集，说明它们意识到，模型能力的提升不只看架构，更看数据质量和数量。面壁智能提供模型，鹿明机器人提供本体和场景，双方一起采数据、训模型、落地应用。这个闭环如果跑通，会是一个可复制的合作模板，供其他模型公司和本体公司参考。
Another point of this partnership is data collection. The cooperation explicitly includes data collection. Earlier we discussed the work of robot data collectors, and Moxian Technology's tactile sensors and data collection gloves. Embodied intelligence model training requires large amounts of real-scenario data, and data collection is the hardest, most laborious link in the supply chain. Luming Robotics and ModelBest partnering on data collection indicates they recognize that model capability improvement depends not only on architecture but on data quality and quantity. ModelBest provides models, Luming Robotics provides bodies and scenarios, and together they collect data, train models, and deploy applications. If this loop works, it will be a replicable partnership template for other model companies and body companies to reference.
