YQ
universal-ai-employee-builder is a reusable Agent Skill for turning a job, role, or recurring business mission into a deployable AI employee specification, which was built from two primary inputs: 1. Source concept — @BrianRoemmele AI bot / AI employee idea 2. Skill-building methodology — Feynman Skill Builder The concept was researched, decomposed, tested, and compiled into this reusable Agent Skill using the feynman-skill-builder methodology.
中文: 通用-人工智能-员工-构建器是一种可重复使用的代理技能,用于将工作、角色或循环业务任务转化为可部署的人工智能员工规范,该规范基于以下两个主要输入构建而成: 1。来源概念 — @BrianRoemmele 人工智能机器人/人工智能员工创意 2。技能建设方法——费曼技能建设者,该概念采用feynman-Skill-builder方法,被研究、分解、测试并汇编成这种可重复使用的Agent Skill。
YQ
universal-ai-employee-builder is a reusable Agent Skill for turning a job, role, or recurring business mission into a deployable AI employee specification, which was built from two primary inputs: 1. Source concept — @BrianRoemmele AI bot / AI employee idea 2. Skill-building methodology — Feynman Skill Builder The concept was researched, decomposed, tested, and compiled into this reusable Agent Skill using the feynman-skill-builder methodology.
中文: 通用-人工智能-员工-构建器是一种可重复使用的代理技能,用于将工作、角色或循环业务任务转化为可部署的人工智能员工规范,该规范基于以下两个主要输入构建而成: 1。来源概念 — @BrianRoemmele 人工智能机器人/人工智能员工创意 2。技能建设方法——费曼技能建设者 该概念经过研究、分解、测试,并采用feynman-skill-builder方法,将其编译为这种可重复使用的Agent Skill。
YQ
@elonmusk @bot I build a Universal AI Employee Builder SKILL based on the article for Grok @bot. @BrianRoemmele provides the core AI-employee concept, while my Skill Builder provides the method used to turn that concept into an evidence-grounded, portable skill. https://github.com/yongqianme/ai-bot-employee
中文: @elonmusk @bot 我根据 Grok @bot 的文章构建了一款通用的人工智能员工构建技能。@BrianRoemmele 提供核心的人工智能员工概念,而我的技能构建者则提供将这一概念转化为循证、便携技能的方法。
YQ
I ran the LinkedIn-market optimizer against @adcock_brett Adcock’s public profile, using the project method of treating the profile as evidence first and a marketing asset second—so the recommendations below avoid inventing achievements or personal technical expertise. https://chatgpt.com/s/t_6ac307ae90d481919c127d28f0cf6fd3
中文: 我使用项目方法将领英市场优化器与@adcock_brett Adcock的公开个人资料进行对比,将个人资料作为证据,并先将营销资源作为营销资源,因此以下建议避免了创造成就或个人技术专长。
YQ
有很多事自己动手,带来的安全感和成就感都很不一样,比如修车比如做饭。我自己做饭,也去外面吃,自己修车也去外面修,也算是一种业余时间的乐趣和消遣。 https://twitter.com/yongqianme/status/2106218822302896341/photo/1
YQ
《兰香如故》里有一段很有意思:大奶奶心爱的银狐裘被烧坏,兰香提出用紫貂滚边重新修补。大奶奶听完觉得主意很好,却告诉她:做好了有赏,做不好就要罚。 从创业者,到后来给创业者打工,我站过这张桌子的两边。创业本来就是在没有答案的时候不断下注,如果每一次创新都是“成功有赏、失败受罚”,最后留下的往往不是最敢尝试的人,而是最懂得不犯错的人。
YQ
Really a cool hand !
中文: 真是一只很酷的手!
YQ
YQ
兼职导师|AI × 时尚,时尚 × 机器人 很高兴分享,我接下来作为@马兰戈尼Marangoni 的客座讲师,研究时尚在具身智能里怎么落地,也研究具身智能对版型、材料、身体和穿着场景提出了什么要求,AI 在融合过程中的新作用。 各位DeepFashion以前的伙伴朋友们,具身人型的朋友们可以聊起来
YQ
I’ll be a guest lecturer (AI × fashion, fashion × robotics ) at Istituto Marangoni, working on how fashion lands in embodied intelligence, what those systems require of pattern, material, the body, and wear, and where AI actually helps. Let’s talk about the topics. https://twitter.com/yongqianme/status/2105633894108467396/photo/1
中文: 我将担任伊斯蒂图托·马朗戈尼的客座讲师(AI × 时尚、时尚 × 机器人),致力于研究时尚如何体现智能,这些系统对图案、材质、身体和穿着的需求,以及人工智能究竟在哪些方面具有帮助。 让我们来谈谈这些话题。
YQ
I’m happy to share that I’m starting a new position as Part-Time Tutor, AI × Fashion and Fashion × Robotics at Istituto Marangoni!
中文: 我很高兴地分享,我将在Istituto Marangoni担任兼职导师,担任AI × 时尚与时尚 × 机器人新职位!
YQ
A small career update. Over the past few months, I’ve been working as a contract Product Manager on humanoid robot projects. Humanoid robotics is still very early. There is a huge distance between a good demo and a product someone is willing to pay for and actually deploy. Still learning. Still building.
中文: 一次小小的职业更新。 过去几个月里,我一直在担任人形机器人项目的产品经理合同。 人形机器人技术还处于早期阶段。一个好的演示与一个愿意付费并实际部署的产品之间有着巨大的距离。 仍在学习。还在建造。
YQ
一个小小的职业近况。 过去几个月,我一直以兼职产品经理的身份,参与一个人形机器人项目。 这份工作让我重新回到了离20年前职业起点很近的地方:真实的机器、真实的问题,以及真实的客户需求。 人形机器人还处在非常早期的阶段。从一个看起来不错的 Demo,到真正有人愿意付钱、愿意部署到实际场景里的产品,中间还有很长的路要走。
YQ
A small career update. Over the past few months, I’ve been working as a contract Product Manager on humanoid robot projects. It has brought me back very close to where my career started 20 years ago: real machines, real engineering problems, and real customer requirements. Humanoid robotics is still very early. There is a huge distance between a good demo and a product someone is willing to pay for and actually deploy. Still learning. Still building.
中文: 一次小小的职业更新。 过去几个月里,我一直在担任人形机器人项目的产品经理合同。 这让我回到了20年前职业生涯起步的非常近的地方:真正的机器、真正的工程问题以及真正的客户需求。 人形机器人技术还处于早期阶段。一个好的演示与一个愿意付费并实际部署的产品之间有着巨大的距离。 仍在学习。还在建造。
YQ
上身按普通人尺度做,是为了数据,不是为了像。设计的时候有思考,为什么要这么设计而不是受限于技术就妥协。 他们也有参考别人的地方,但也有很多自己的思考。国内有很多创业公司有能力做很好的产品,特别是机器人,但是资本管理层甚至技术团队没有承担风险的勇气,抄是一个不错的决定,于是就抄最火的。
YQ
Taku is not another walking humanoid. It is a semi-humanoid physical agent built for a laundry-room shift: load, transfer, unload, fold, stack, and interrupt the current job when a machine finishes. Form follows the work. Human-shaped above the waist, a folding column below, on a wide, low four-wheel base. Two 7-DoF arms; about 21 DoF in the upper body. The goal is a human work envelope, not a human gait. The job is hard floors and stations a few meters apart: reach into a dryer drum, fish a leftover towel from a washer, crouch to the bottom shelf, reach one shelf above the head. Wheels move between stations. The folding lower body covers height. The wide base buys stability and long unattended runs. Stairs, curbs, and crowded aisles are out of scope. The human-scale torso is a data choice, not a likeness choice. Dyna’s URR uses wrist, elbow, chest, and footprint poses as one language that both people and the robot can emit. A model pretrained on a million hours of human video transfers only if those landmarks line up. The arms use **low-ratio planetary actuators** for speed and acceleration; a 100 Hz whole-body RL controller is supposed to absorb the missing stiffness. The hands are parallel jaws, not five fingers: enough for cloth, a smaller action space, with drops written into the workflow—pick up, mark dirty, wash again. Control runs on three clocks: 100 Hz tracking of joints and wheels, Dyna-2 at about 5 Hz proposing the next few seconds of motion, and a slower vision-language orchestrator choosing the next step and keeping memory. Joint-level detail stays in the controller, so a hardware change means retraining that controller in simulation and old data still works. Morphology is a data interface. The honesty is the strength: optimized for long indoor jobs on flat floors, not for bipeds or dexterous hands. The limits are equally clear: not a generalist, the base takes floor space, the gripper has a ceiling, and payload and mechanism specs are still unpublished. Compounding reliability is the product. Nearly 80 steps in a row only work if each one is extremely steady. Fewer dropped towels in hardware matter as much as better sequencing in software.
中文: 塔库不是另一个行走的人形。它是一种用于洗衣房换班的半人形物理设备:负载、转移、卸载、折叠、堆叠,以及当机器完成时中断当前的工作。 形式遵循工作。腰部上方呈人形,下方有一列折叠柱,位于宽阔、低矮的四轮底座上。两具7-DoF手臂;上半身约21个DoF。目标是一个人类工作的信封,而不是人类的步态。工作是硬质地面,距离几米远:伸手入干火桶,从洗衣机里捞出一条剩毛巾,蜷缩在底架上,到达头顶上方的一个架子。车轮在车站之间移动。折叠的下半身遮盖了高度。宽底座具有稳定性和长时间无人看管的运行。楼梯、路障和拥挤的通道都不在范围之内。 人尺度躯干是一种数据选择,而不是类似的选择。Dyna的URR使用手腕、肘部、胸部和脚印作为人和机器人都能发出的一种语言。一个模型预训练了一百万小时的人类视频传输,前提是这些地标会排起来。该武器采用低比例行星执行器,以实现速度和加速;100 Hz全身RL控制器应能够吸收缺失的刚性。双手是平行的颌骨,而不是五根手指:足够布料,动作空间更小,工作流程中写着滴落——捡起来,标记脏污,再洗一次。 控制运行三个时钟:对关节和车轮进行100赫兹的跟踪,在约5赫兹处运行的Dyna-2,建议接下来几秒钟的移动,以及一个较慢的视觉语言编排器,选择下一步并保留记忆。关节级细节仍停留在控制器中,因此硬件变更意味着该控制器在仿真和旧数据中仍有效。形态学是一种数据接口。 诚实是优势:专为平地室内长工作而优化,不适合带有�bedes或灵巧的双手。界限同样明确:不是普通主义者,基地需要地面空间,抓地器有天花板,有效载荷和机制规格至今仍未发表。复合可靠性是产品。连续近80步只有在每个步骤都非常稳定的情况下才能发挥作用。硬件物质中掉落的毛巾减少,以及软件中更好的测序功能。
YQ
Taku is not another walking humanoid. It is a **semi-humanoid physical agent** built for a laundry-room shift: load, transfer, unload, fold, stack, and interrupt the current job when a machine finishes. Form follows the work. Human-shaped above the waist, a folding column below, on a wide, low four-wheel base. Two 7-DoF arms; about 21 DoF in the upper body. The goal is a human work envelope, not a human gait. The job is hard floors and stations a few meters apart: reach into a dryer drum, fish a leftover towel from a washer, crouch to the bottom shelf, reach one shelf above the head. Wheels move between stations. The folding lower body covers height. The wide base buys stability and long unattended runs. Stairs, curbs, and crowded aisles are out of scope. The human-scale torso is a data choice, not a likeness choice. Dyna’s URR uses wrist, elbow, chest, and footprint poses as one language that both people and the robot can emit. A model pretrained on a million hours of human video transfers only if those landmarks line up. The arms use **low-ratio planetary actuators** for speed and acceleration; a 100 Hz whole-body RL controller is supposed to absorb the missing stiffness. The hands are parallel jaws, not five fingers: enough for cloth, a smaller action space, with drops written into the workflow—pick up, mark dirty, wash again. Control runs on three clocks: 100 Hz tracking of joints and wheels, Dyna-2 at about 5 Hz proposing the next few seconds of motion, and a slower vision-language orchestrator choosing the next step and keeping memory. Joint-level detail stays in the controller, so a hardware change means retraining that controller in simulation and old data still works. Morphology is a data interface. The honesty is the strength: optimized for long indoor jobs on flat floors, not for bipeds or dexterous hands. The limits are equally clear: not a generalist, the base takes floor space, the gripper has a ceiling, and payload and mechanism specs are still unpublished. Compounding reliability is the product. Nearly 80 steps in a row only work if each one is extremely steady. Fewer dropped towels in hardware matter as much as better sequencing in software.
YQ
Is the wheel so big so it can fit a bigger battery? @JasonMa2020
中文: 轮子这么大,能装上更大的电池吗?@JasonMa2020
YQ
其实国内有很多创业公司有能力做很好的产品,特别是机器人,但是资本甚至管理层没有承担风险的勇气,抄是一个不错的决定,于是就抄最火的。
YQ
The bot looks cool !
中文: 机器人看起来很酷!
YQ
My Book; The Complete Guide to Former Founder Employment is now available on @AppleBooks . https://books.apple.com/us/book/the-complete-guide-to-former-founder-employment-success/id6817230261
中文: 我的书:《前创始人就业完整指南》现已在@AppleBooks上提供。
YQ
YQ
YQ
具身智能家用,就要考虑这个问题剩女这个问题, 我们现在天天吵的“剩女”“不婚”“低生育”“天价彩礼”,可能一开始就放错地方了。它们看起来像婚恋问题。再往下看一层,更像经济转型切出来的一块截面。 一个女人现在可以读完大学,进公司,自己挣钱,也可以自己租房、买房、开车出门。父母养老和生活开销,越来越多人也得自己扛。这些事一旦能做成,婚姻就很难再按上一代人的办法运转。 男人这边也变了。不是谁突然更现实,是原来撑起婚姻的那套经济安排,开始松了。 https://mp.weixin.qq.com/s/NoaF9nokdlAstUZU887Aew
YQ
The future work from home concept !
中文: 未来在家工作的理念!
YQ
具身智能摸着协作机器人的老石头在河里 行业换新,不换内核,一边是不要35+的老登,一边继续摸着老登摸过的石头在河里重走老登的路。 具身智能(Embodied AI)没有逃过物理世界的引力法则。那些在展会上令人惊艳的“人形机器人做咖啡”、“药房分拣”Demo,早在十年前协作机器人(Cobot)爆发时就被验证为商业死胡同。协作机器人最终靠着退守工业场景(如螺纹锁付、焊接、码垛)才完成了商业闭环;而今天的具身智能创业者们,正捧着天价估值,在商业化试错的暗礁区重走这条老路。 https://mp.weixin.qq.com/s/8zsl-qju_SB1O3dAX0PGVg
YQ
百亿公司的 PR,已经不是宣传部门 前几天,我看到一家估值已经进入百亿人民币区间的科技公司,出现在一家全球主流媒体上。 报道其实传播得不错。 原始文章发布后,很快被不同国家的财经、科技和地方媒体转载。仅公开可追踪的媒体数据库里,同一篇稿件就出现了数十个 syndicated copies。中文互联网也迅速跟进。 从传统 PR 的报表来看,这大概是一张很好看的成绩单: 国际顶级媒体,拿到了。 海外覆盖,拿到了。 几十家转载,拿到了。 搜索结果,铺开了。 如果月底要做一份 PPT,这一页甚至可能值得放在最前面。 但我读完报道后,却有一点说不出的可惜。 https://mp.weixin.qq.com/s/xi2byDEBCh_l31XYhqqdJw
YQ
The new @DynaRobotics is probably designed and build in their Shanghai office .
中文: 全新的@DynaRobotics 可能是其上海办事处的设计和建造。
YQ
The Road Is Yours is now available on Apple Books. https://books.apple.com/us/book/the-road-is-yours/id6816007365
中文: 《The Road Is Yours》现已在Apple Books上推出。
YQ
好的创始人和好的CEO,关于狼的故事 很多年前在德国工作时,我无聊,给德国同事讲过一个关于销售和技术的故事。 现在回头看,它更像是在讲 Founder 和 CEO。 A 和 B 开车在黑森林迷路了。 两个人都很饿。 A 说,你守着车等救援,我去找点吃的。 很久以后,B 看见 A 从森林里狂奔出来,身后追着一头狼。 A 冲进车里,狼也跟着扑了进去。 就在狼进车的一瞬间,A 跳出去,关上车门。 车里只剩下 B 和狼。 A 隔着车窗说: “这头你搞定,我再去弄一头来。” 年轻时,我觉得这个故事讲的是销售和技术。 销售敢把订单拿回来,技术负责把承诺做出来。 创业以后再想,它其实更像 Founder 和 CEO。 Founder 最重要的能力之一,是把“狼”带回来。 客户、订单、资本、人才、合作机会,很多时候都不是摆在桌上的晚餐。它们在森林里,会跑,会咬人,甚至可能反过来吃掉你。 有人必须敢走出去,把它引回来。 但把狼带回来,并不等于公司有了肉。 CEO 要面对的是另一种危险: 怎么杀掉这头狼。 一个大客户可能意味着不可能的 deadline;一笔融资可能意味着更高的增长要求;一个漂亮的产品承诺,背后可能是半年研发、供应链、现金流和几十个人的压力。 机会进入公司以后,才真正开始变贵。 所以我现在觉得,好的 Founder 和 CEO 关系,最有意思的地方甚至不是分工。 而是信任。 A 敢把狼关进车里,因为他相信 B 真能搞定。 B 愿意面对这头狼,因为他相信 A 不会什么东西都往车里塞。 创业公司最怕的,也许不是森林里没有狼。 而是 Founder 不敢去找狼; 或者 Founder 不停往车里塞狼,却从来没有想过,车里的人还能不能活着出来。
YQ
YQ
人形机器人为什么一定要像人?Figure 给出了一个新的答案。 2025年在北京上班的时候,参加了一个Google 朋友组织的聚会,期间有一位在场的创业者问了我一个问题,为什么具身机器人要像人?我的回答是:这个世界,本来就是按照人的身体设计的。 但 Figure 最近公布 Helix 2.5 的结果之后,我开始觉得,这可能只是人形结构比较表层的优势。 https://mp.weixin.qq.com/s/g6FLKwQfLW2e8FtL5GnTLQ
YQ
最近在给一台新设计的人形机器人做衣服 一开始,这件事听起来并不复杂。机器人已经设计完成,身高、肩宽、胸围、腿长都有明确尺寸。找服装设计师量一下,选一种弹性足够好的面料,打版、裁剪、缝制,似乎就可以了。 真正开始做以后,很快会发现事情不是这样。 人穿衣服的时候,衣服服务于人体。给机器人做衣服的时候,织物必须同时服务于机械结构、关节运动、散热、传感器、安全、维修,以及最后才轮到的外观。 一件在人身上很普通的衣服,到了机器人身上,可能拉住肩关节,可能在深蹲时卡进髋部,也可能遮住摄像头和散热口。机器人连续运动几十次以后,原本很漂亮的衣服还可能慢慢向一侧迁移,最后整件衣服都是歪的。 所以做到后来,我越来越觉得,我们其实不是在给机器人做衣服。 我们正在设计一个新的机器人子系统:Soft Goods,软外饰系统。 而这个看起来不起眼的变化,可能也是人形机器人从 prototype 走向真正产品时,必然要补上的一块。 https://mp.weixin.qq.com/s/qnxrp20xa2yh2vHPBhf5XQ
YQ
人型机器人创业,抄的是上一家刚公开的答案 我们到底在造机器人,还是在追美国人的 BOM? 这几年看人形机器人,我越来越经常看到一种熟悉的产品决策方式。 看到美国公司用 RealSense,我们也用 RealSense。 Tesla 做五指灵巧手,于是大家研究五指、腱绳、空心杯电机。 Figure 进入汽车工厂,工业场景很快成为行业热点。 2025 年 Sunday Robotics 发布 Memo。轮式底盘、升降躯干、双臂,以及为了操作和数据采集设计的简化机械手出现后,"三指""简化手""家庭数据"又迅速成为新的讨论方向。 单独看任何一件事,都不能叫抄袭。 RealSense 本来就是成熟的机器人视觉方案;三指机械手也远早于 Sunday Robotics,Robotiq、Barrett 很多年前就在做。工程师面对相似的成本、可靠性和开发周期约束,本来就可能得到相似的答案。 真正让我担心的,其实不是 Copy BOM。 而是另一件事: Copy Product Definition。 https://mp.weixin.qq.com/s/Ah3Z6X2nPgUXzdLaKpgdTA
YQ
小红书说,我的深度思考被人看见需要花钱 小红书最近给我发了两条消息。 大意是:你的文章不错,有几千阅读,互动率也挺高,说明内容击中了创业者的痛点。所以,要不要花35块钱,让更多人看到? 看到这里我有点恍惚。 原来一篇文章有没有价值,和它能不能被更多人看到,是两回事。 你可以花几个小时查资料、思考、写作,把二十年的经历压缩成两千字;读者也可以认真看完、点赞、讨论。但算法最后拍拍你的肩膀说:写得不错,数据也不错——现在请付钱。 短视频可以靠情绪获得流量,美女可以靠颜值获得流量,争议可以靠吵架获得流量。 而一个普通人认真写下来的深度思考,想再往前走一步,先扫码。 突然觉得这个商业模式也挺有哲学意味: 思想免费,传播收费。 深度不值钱,让深度被看见才值钱。 最有意思的是,我写的其中一篇,恰好叫: 《百亿公司的PR,已经不是宣传部门》 现在看来,我可能还得先给自己的文章做个 PR。
YQ
YQ
2020年1月20日疫情前四天,我飞德国和德国以前公司的CEO在Osnabruck 的一家酒店喝咖啡,Mr. Bayer 告诉我他要带妻子女儿去南美洲游玩,他也说,人生30年学习30年赚钱30年享受,这是他60年左右的见解。 他告诉了我一个人生哲理:生活也许很长。属于我们的时间,并没有那么长。 https://mp.weixin.qq.com/s/cJUQpr2Fjic917nNM2UGBw
YQ
我的小说是用了大量的AI,和我定制的小说skill,定制的内容边界,发布前已经给了很多人看了,没有什么AI 味道,大家可以看看给给点评。
YQ
路在你手中第二部最先赶到的人(插章第六章沈安出生) https://x.com/yongqianme/status/2103646868660199429?s=20
YQ
YQ
路在你手中第二部最先赶到的人(第三/四/五章) https://x.com/yongqianme/status/2103285672144691548?s=20
YQ
YQ
这是一部关于机器人,工程师,移民,家庭和教育以及成功责任的小说,小说里有我的影子,也有朋友伙伴的影子,也有很多创业者和从中国到世界其他国家创业者的影子,英文版已经上架Kindle, 简体中文版慢慢连载更新 https://www.amazon.com/dp/B0HKVMY1HZ
YQ
My new novel, The Road Is Yours, began with a simple question: What happens when someone who has spent his entire career solving problems encounters the parts of life that cannot be solved? https://x.com/i/article/2103010408311271425
YQ
《路在你手中》是一部长篇小说。它写机器人,也写创业;写中国、德国和美国的工厂、公路与城市;写一个工程师如何理解机器,又如何用了大半生,重新理解家庭、责任、成功和自己。 https://x.com/i/article/2102958737761959936
YQ
Freedom as a feeling evaporates. Freedom as the right to choose stays in the hand. https://x.com/i/article/2102563094522478592
中文: 自由作为一种感觉会消失。自由作为选择权,可留在手中。
YQ
Over the past month, we’ve been designing and making clothing for a newly developed humanoid robot. Once we started dressing the robot, we quickly realized: this was no longer a clothing problem. It was a robotics engineering problem. https://x.com/i/article/2102255261578878976
YQ
The AI almost starts a war ! https://x.com/i/article/2101273280414203904
中文: 人工智能几乎开始了一场战争!
YQ
When I was working in Beijing in 2025, I attended a gathering organized by a friend at Google. At one point, a founder asked me a question: Why should embodied robots look human? My answer was simple: because the world was built around the human body. https://x.com/i/article/2101114582979018752
中文: 2025年我在北京工作时,参加了一场由一位朋友在谷歌组织的聚会。有一次,一位创始人问我一个问题:为什么体能的机器人应该看起来像人类?我的回答很简单:因为世界是围绕人体构建的。
YQ
In 2026, Bay Area single-family medians have generally sat between $1.27 million and $1.45 million, with San Francisco, San Mateo, and Santa Clara Counties often much higher. Furnished or short-term whole-house rentals commonly run several thousand dollars a month. Recent median household income in the region is about $137,000; the San Jose / Santa Clara area is higher, often cited around $165,000–$167,000. In the more expensive counties, four-person area median incomes commonly fall between $160,000 and $205,000. Figure has floated a possible future home lease of about $400–$600 a month. Against Bay Area house cleaning, that is roughly the monthly cost of a mid-priced recurring cleaner. The robot only covers making beds, folding towels, and tidying living rooms, and it finishes the full job about 56% of the time, so household ROI is negative. At list performance, that is about 4.5 useful household hours a week; at 56% success, you need closer to eight hours of runtime to net the same finished work. The lease is only about 5% of median household income, so the time leverage is larger for high-wage households. The piece to watch is Index, Figure’s own data-collection platform. It is not another short-video app. It is the supply line for Helix: ordinary people film first-person labor on their phones — making beds, folding laundry, stocking shelves, cooking, cleaning — Figure pays them, the footage goes into the corpus, and that corpus pretrains the humanoid. When Figure took Index public on August 25, 2026, it reported 264,000 downloads across 108 countries, more than 44,000 weekly active creators, more than 16 million videos, $15 million already paid out, and about 35 minutes of video ingested every second — roughly 4.9 years of human work uploaded each day. Per 1,000 hours collected, the company said the set contained about 373 unique tasks, 1,146 unique objects, and 116 unique environments.
中文: 2026年,旧金山湾区独户住宅的中位数通常在127万至145万美元之间,而旧金山、圣马特奥和圣克拉拉县的中位数通常要高得多。装修或短期的全屋租赁通常每月花费数千美元。该地区近期家庭收入中位数约为13.7万美元;圣何塞/圣克拉拉地区较高,通常约为16.5万至16.7万美元。在较昂贵的县,四人区的中位收入通常在16万至20.5万美元之间。 图已提出未来可能每月约400至600美元的房屋租赁。与湾区房屋清洁相比,这大致相当于一家中等价位的定期清洁工的月度费用。机器人仅覆盖床铺、折叠毛巾和整理客厅,约56%的时间完成完整工作,因此家庭投资回报率为负值。在列表性能下,每周约4.5个有用的家庭工作时间;成功率为56%,需要接近八小时的运行时间才能完成相同的工作。租赁权仅占家庭收入中位数的约5%,因此高薪家庭的时间杠杆更大。 值得关注的是Index,这是Figures自己的数据收集平台。 它不是另一个短视频应用。这是Helix的供货线:普通人用手机拍摄第一人称劳动——制作床铺、折叠衣物、袜子架子、做饭、打扫——图片支付,录像进入语料库,该语料库预应人形。 图于2026年8月25日将Index公之于众,报告在108个国家下载量达到26.4万次,每周活跃创作者超过4.4万次,视频超过1600万次,已支付1500万美元,每秒约35分钟视频被录用——每天上传约4.9年的人类作品。每收集1000小时,该公司表示该套装包含约373项独特任务、1146件独特物品以及116种独特环境。
YQ
RT @yongqianme: In 2026, Bay Area single-family medians have generally sat between $1.27 million and $1.45 million, with San Francisco, San Mateo, and Santa Clara Counties often much higher. Furnished or short-term whole-house rentals commonly run several thousand dollars a month. Recent median household income in the region is about $137,000; the San Jose / Santa Clara area is higher, often cited around $165,000–$167,000. In the more expensive counties, four-person area median incomes commonly fall between $160,000 and $205,000. Figure has floated a possible future home lease of about $400–$600 a month. Against Bay Area house cleaning, that is roughly the monthly cost of a mid-priced recurring cleaner. The robot only covers making beds, folding towels, and tidying living rooms, and it finishes the full job about 56% of the time, so household ROI is negative. At list performance, that is about 4.5 useful household hours a week; at 56% success, you need closer to eight hours of runtime to net the same finished work. The lease is only about 5% of median household income, so the time leverage is larger for high-wage households. The piece to watch is Index, Figure’s own data-collection platform. It is not another short-video app. It is the supply line for Helix: ordinary people film first-person labor on their phones — making beds, folding laundry, stocking shelves, cooking, cleaning — Figure pays them, the footage goes into the corpus, and that corpus pretrains the humanoid. When Figure took Index public on August 25, 2026, it reported 264,000 downloads across 108 countries, more than 44,000 weekly active creators, more than 16 million videos, $15 million already paid out, and about 35 minutes of video ingested every second — roughly 4.9 years of human work uploaded each day. Per 1,000 hours collected, the company said the set contained about 373 unique tasks, 1,146 unique objects, and 116 unique environments.
YQ
RT @yongqianme: Robotics as if deployment mattered. https://x.com/i/article/2099055397147729920
中文: RT @yongqianme:机器人技术,仿佛部署很重要。
YQ
Great job !
中文: 很棒的工作!
YQ
具身智能摸着协作机器人的老石头在河里。 一边是不要35+的老登,一边继续摸着老登摸过的石头在河里重走老登的路。 具身智能(Embodied AI)没有逃过物理世界的引力法则。那些在展会上令人惊艳的“人形机器人做咖啡”、“药房分拣”Demo,早在十年前协作机器人(Cobot)爆发时就被验证为商业死胡同。协作机器人最终靠着退守工业场景(如螺纹锁付、焊接、码垛)才完成了商业闭环;而今天的具身智能创业者们,正捧着天价估值,在商业化试错的暗礁区重走这条老路。 中文版:https://mp.weixin.qq.com/s/8zsl-qju_SB1O3dAX0PGVg
YQ
具身智能摸着协作机器人的老石头在河里。 一边是不要35+的老登,一边继续摸着老登摸过的石头在河里重走老登的路。 中文版:https://mp.weixin.qq.com/s/8zsl-qju_SB1O3dAX0PGVg
YQ
The embodied AI companies are all stuck at the door. https://twitter.com/yongqianme/status/2100452592459674100/photo/1
中文: 具体的人工智能公司都陷入了困境。
YQ
Those show-floor demos — a humanoid making coffee, a robot picking drugs in a pharmacy — already failed as businesses a decade ago. https://x.com/i/article/2100393099071725568
中文: 十年前,那些展厅演示——一种制作咖啡的人形机器人,一种在药店采摘毒品的机器人——在企业领域已经失败了。
YQ
Today, humanoid and embodied-AI companies are experimenting with coffee shops, pharmacies and other consumer-facing demos again. The technology is different, especially with vision-language-action models and general-purpose manipulation, but the commercial question remains remarkably similar: Is embodied AI discovering a new market, or rediscovering why robotics eventually went back to industry?
中文: 如今,人形和化身人工智能公司正在重新尝试咖啡店、药店和其他面向消费者的演示。 技术有所不同,尤其是视觉语言动作模型和通用操作,但商业问题却非常相似: 人工智能正在发现一个新市场,还是正在重新发现机器人技术最终回归工业的原因?
YQ
Most never became large, repeatable markets. The applications that ultimately produced sustained demand were far less glamorous: machine tending, welding, palletizing, assembly, inspection and material handling inside factories and warehouses.
中文: 大多数从未变成大型、可重复的市场。最终产生持续需求的应用远不那么光鲜:机械加工、焊接、码垛、装配、检验以及工厂和仓库内的物料处理。
YQ
Embodied AI may be retracing the same path cobots took a decade ago. Coffee robots, robotic pharmacies, bartenders, retail kiosks and other service demos are not new. Cobots explored many of these applications years ago.
中文: 实体人工智能可能正在回应十年前的协作机器人路径。 咖啡机器人、机器人药店、调酒师、零售自助服务店和其他服务演示并不新鲜。协作机器人多年前就探索过许多这些应用。
YQ
Robotics as if deployment mattered. https://x.com/i/article/2099055397147729920
中文: 机器人技术,仿佛部署很重要。
YQ
中文: 会发生的事情!#羟基
YQ
Shit happens ! #hyrox
中文: 会发生的事情!#羟鐘
YQ
If you’re building in humanoid robotics or Physical AI and want an experienced sounding board to accelerate your product definition and deployment strategy, DM me! 📩 How to apply: Send a brief intro, pitch deck/website, and your biggest product bottleneck. Learn more about my background here: https://qianyong.me/ #Robotics #PhysicalAI #Startups #HumanoidRobots
YQ
What I help with: • Field-readiness & product strategy • Hardware/software/edge-to-cloud architecture • Real-world industrial deployment & GTM Who this is for: ✅ Funded startups (Seed / Series A+) ✅ Real business & tangible hardware/software (active pilots or customers) ✅ Full-time founding teams
中文: 我帮助什么: • 现场准备与产品战略 • 硬件/软件/边缘到云架构 • 真实工业部署与GTM 这是为了谁: ✅ 资助初创企业(种子/A+系列) ✅ 真实业务与有形硬件/软件(主动试点或客户) ✅ 全职创始团队
YQ
I’m offering 3 months of free product & technical advisory (10 hrs/wk, remote) for humanoid robotics startups. With 20+ years in robotics, industrial automation, and Physical AI, I want to help teams bridge the gap between lab demos and field readiness. 👇
中文: 我为人形机器人初创企业提供3个月的免费产品和功能;技术咨询(10小时/小时,远程)。 在机器人、工业自动化和物理人工智能领域工作超过20年,我希望帮助团队弥合实验室演示与现场准备之间的差距。 👇
YQ
Visited a humanoid actuator company yesterday. The founder said they have to ship 300 actuator sets a month to a single humanoid customer. That’s 300 robots a month. 3,600 a year. He put the order at about $160 million. Lies have gotten very cheap.
中文: 昨天参观了一家人形执行器公司。 创始人表示,他们每月必须向单个人形客户运送300套执行器。每月300台机器人,每年3600台。他将订单定在了约1.6亿美元。 谎言变得非常便宜。
YQ
每个时代、每个国家,都有各自的爽剧, 《基督山伯爵》就是法国那个1840年代的爽剧。 https://x.com/i/article/2097833849317404673
YQ
After money arrives, people recalculate what counts as worth it, what counts as falling behind, and what a generation is not allowed to miss. The robots are not intelligent yet. The people have already become very intelligent about the allocation. https://x.com/i/article/2097490112594714624
中文: 金钱到来后,人们会重新计算哪些是值得的,哪些东西会落后,哪些是不允许错过的。机器人还不聪明。人们已经对分配变得非常明智了。
YQ
Prediction is getting cheaper. Judgment is getting expensive. https://x.com/i/article/2096761739757174784
中文: 预测变得越来越便宜了。判断力越来越高。
YQ
100 块只是一张收据, 收据后面,是这条赛道第一次被迫用公开数字对质自己的故事。 https://x.com/i/article/2095531339344973824
YQ
The campaign is real. The analogy is only half right. America can block chips. It cannot block a production line. https://x.com/i/article/2094930940124672000
中文: 这场活动是真实的。这个类比只有一半。 美国可以屏蔽芯片。无法封锁生产线。
YQ
YQ
YQ