X Square Robot

Meet the world at home, where life happens and bots become family 35 days ago, at our “Born to Bot, Bot to Family” launch event, we shared our vision of bringing robots into real homes. Today, we’re very happy to share that our robots are now gradually entering real families. For embodied AI, the real world is everyday life: different routines, different kitchens, and different ways of doing even the simplest tasks. This is where robots meet the world at home, where life happens and bots become family. They are still learning. They may move slowly, hesitate, and sometimes look a little clumsy. But every home they enter helps them understand the world a little better.

@servetkodu_EN Totally agree!

@XSquareRobot This is a useful framing because the demand signal is not really “sci-fi.” It is labor gaps, aging, loneliness, and household time pressure. The hard part is still whether reliability, safety, and service economics can meet that demand at a consumer price point.

Over the past month, we asked families why they would want a robot at home. The answers were not really about sci-fi. They were about only children needing company. Parents getting older. People living alone. Couples arguing over chores. Long workdays. Homes that still need care when no one has energy left. Some wanted help cleaning. Some wanted someone to check the doors, windows, gas and lights. Some wanted companionship for a child, a parent, or themselves. That is what makes home robots interesting to me. Not because they are perfect today. But because the need is already real. Would you live with one?

@GillelandKristi We’re working on making the base and overall size smaller and more flexible, so it can better fit real home layouts💪

@XSquareRobot My house isn't big enough for a robot with a base like that to get around.

感谢福爸的分享😊

Home robots are leaving stage demos and entering the only test that really matters: ordinary family life. X Square Robot is starting to move its next-gen home robot into real households. It runs on WALL-B, a world model designed to connect vision, language, touch, action, and physical prediction, which is exactly what a home robot needs when the real world refuses to stay neat. A kitchen is not a controlled environment of a factory floor. it is a moving negotiation between habits, clutter, pets, children, half-finished chores, and objects that never return to the same place twice. That is where Moravec’s paradox shows up: tasks that feel effortless to humans, like picking up clutter, avoiding pets, or judging what belongs where, are often brutally hard for robots. Would you bring a robot with daily chores?

@itseasytosayit Thank you, that means a lot. We believe robots only get truly useful when they meet the real world honestly, including all the messy, imperfect parts

@XSquareRobot your decision to advertise with humility is a welcome & refreshing change — the willingness to admit imperfection brings a deeper kind of perfection

Meet the world at home, where life happens and bots become family 35 days ago, at our “Born to Bot, Bot to Family” launch event, we shared our vision of bringing robots into real homes. Today, we’re very happy to share that our robots are now gradually entering real families. For embodied AI, the real world is everyday life: different routines, different kitchens, and different ways of doing even the simplest tasks. This is where robots meet the world at home, where life happens and bots become family. They are still learning. They may move slowly, hesitate, and sometimes look a little clumsy. But every home they enter helps them understand the world a little better.

@techniahq 😊😊

@XSquareRobot Very good work, congratulations!

@spaceandtech_ GTA 6 loading screen taking so long, robots had time to get a side hustle🤪

@XSquareRobot We got robots walking dogs before GTA 6 😭

Excited to be part of this! Looking forward to sharing WALL-WM with the Robotics & World Models community at #CVPR2026. See you there!

Great question. We don’t leave safety to the policy alone — it’s a layered system with real-time perception, task constraints, motion-level safety checks, and user stop options. When the robot is uncertain or something feels off, the robot is to pause, which is the safest way. Real homes are complex, and that’s exactly why real-world deployment matters.

@XSquareRobot Genuinely curious how do you handle safety when the robot is in a home with unpredictable situations? Is there a real-time override system, or does the policy itself learn to stop when something feels off?

@XRoboHub We’re working hard to make that future arrive faster！

@XSquareRobot This is so cool, I honestly can’t wait to have a robot help with chores at home😄

Thanks to Crystal from the @SCMPNews for trying out our robot-powered home cleaning service with 58.com! The robot may still be learning, but it’s getting faster and better with every real home service. If you’re in Shenzhen or Beijing, come try it for yourself.

China is testing AI-powered robot assistants designed to help with household chores. Developed by X Square Robot, the robots can sort shoes, organize rooms, fold clothes and pick up trash while learning from each new home they enter. The service is already being tested in Shenzhen and Beijing.

F.03 vs XSquareRobot: one shows a polished 2-minute bedroom reset, the other is already cleaning messy real homes. 🤖 In Figure’s latest video, two F.03 robots reset a bedroom in under 2 minutes. They open the door, organize clothes, close the laptop, put headphones back on the stand, throw trash into the bin, push in the chair, and make the bed together. The highlight is the bedding sequence. Two robots handling a large deformable object together is hard, and the coordination looks clean. But this is still not a full cleaning workflow: no fresh sheets, no duvet cover swap, no pillowcase change, and no real home-level mess. Now look at @XSquareRobot. This video is a compilation of its robots cleaning different homes. It is slower, messier, and less cinematic than Figure’s video. But the environment is much more real. The robot can fold clothes, organize shoes onto a rack, replace trash bags, clear clutter from sofas and dining tables, and pick up scattered toys. One detail stands out: when clearing a table, it can stack two transparent plastic cups together before throwing them into the bin. That is the kind of small domestic judgment home robots actually need. Its strongest point is human-robot collaboration. It can work with a human cleaner to fold large bedsheets, inside narrow rooms and non-standard home layouts. So yes, F.03 is a strong signal. But XSquareRobot feels closer to the first version that may actually reduce housework. The clip is messy, slow, and strangely therapeutic. Because robots solving daily chores no longer feels that far away.

We are honored to share that X Square Robot has been named to the Forbes China 2026 AI Tech Enterprises Top 50. This recognition reflects our long-term commitment to building general-purpose embodied AI for the physical world. From WALL-A and WALL-B to WALL-OSS and the QUANTA robot series, X Square Robot is developing a full-stack platform that brings together foundation models, robot hardware, real-world data systems, and commercial deployment. Embodied AI is moving from research and demonstrations into real-world applications. We are proud to be part of this shift, and we will continue working to make robots more capable, useful, and accessible across home services, industrial manufacturing, logistics, elderly care, hospitality, retail, and public-service scenarios. Thank you to Forbes China for the recognition, and to our team, partners, and supporters who are helping bring embodied AI into the real world. #XSquareRobot #EmbodiedAI #Robotics #HumanoidRobots #ArtificialIntelligence #Forbeschina

Morgan Stanley’s latest humanoid robotics report, Humanoid Horizons: Money Meets Machines, points to a clear inflection point for the industry. YTD venture funding for humanoid robots has already surpassed 2025 levels, largely driven by China. In April 2026 alone, China recorded 41 humanoid robotics financing deals, compared with 16 in April 2025 and 6 in April 2024. In the first four months of 2026, China saw 131 deals across the humanoid value chain, including robot bodies, robot brains, and integrators. But the bigger shift is not just capital. Morgan Stanley highlights that the “robot brain” stack remains unsettled, with world models and VLA models emerging as key directions. The real competitive moat is moving toward proprietary data flywheels — especially scarce real-world robot data. The next decade of robotics will be won by companies that can close the loop between data, models, hardware, and real-world deployment. We’re honored to see X Square Robot included in Morgan Stanley’s China Humanoid Value Chain map as a key player in the “Brain” category. From day one, X Square Robot has focused on world-model-based embodied intelligence, large-scale real-world robot data, long-horizon dexterous manipulation, and full-stack integration across models, data, hardware, and deployment. From WALL-A and WALL-B to WALL-OSS and the QUANTA robot series, we are building embodied AI systems that move beyond demos and create value in real environments.

A robot cleaned my apartment in Shenzhen.🤖 Booked like food delivery for ~€20, including a human cleaner, and two engineers on site. 🙋🏻‍♀️ @XSquareRobot managed to: ✅ Take out the trash ✅ Tidy scattered toys ✅ Fold laundry If you can imagine it, Shenzhen is building it. 🪩

@xiaogaifun 感谢小盖的分享！

昨天下午听了一场具身智能的分享，只有 45 分钟，非常精彩，基本把整个行业目前的情况讲清楚了。我建议对具身智能感兴趣的朋友都找回放看一下。 分享人是明星创业公司自变量机器人的 CEO 王潜和 CTO 王昊。虽然是一场新模型的发布会，但他们都很实诚，实实在在把当下行业里的问题、自家的思路都摆在了台面上。 王潜开场的话让我印象很深。他说，我们每个人都期待，一觉醒来，家里的机器人能把房间打扫的干干净净。但目前没有任何一台机器人可以在没有遥控操作的情况下搞定这事。 AI 在数字世界的进展快得惊人，但到了物理世界，离我们想要的状态还很远。 这时候肯定有人要追问：机器人都能表演武术、跳舞、后空翻了，做点家务真的有这么难吗？再说，工厂里那些机器人上个世纪就已经非常成熟，难道家里的活儿，比工厂蓝领的工作还难干？ 答案是：对，确实更难。 我们刷到的那些酷炫机器人视频，本质上都是预设动作。工程师提前把程序编好，机器人按部就班跑一遍就行。它看起来灵活聪明，其实根本不理解自己在干什么。 工厂机器人也类似，做的事情是重复。制衣厂的机械臂卡在一个环节，比如钉纽扣，每天把同一个动作重复一万次，精度是够的，但也仅此而已。 但家务活完全是另一回事。 今天餐桌的杯子，明天可能放在了卧室床头柜。昨天下午光线很亮，今天阴天光线就变了。极端一点讲，做家务的过程中，没有任何一次情况是跟上一次完全一致的。机器人必须当下识别、当下判断、当下执行。 那行业里现在是怎么解决这个问题的？ 整体的思路，叫 VLA 架构。V 是视觉，L 是语言，A 是动作。把这三样打通，机器人就具备了基础的感知和执行能力。这条路已经跑了一段时间，我们现在看到的绝大多数具身智能模型，基本都是 VLA 架构。 但 VLA 有个本质问题，它的核心范式是模仿。 具体怎么运作？举个例子。想让机器人学会叠衣服，得先喂给它大量叠衣服的数据。模型在这个过程里，把看到的画面和对应的动作一点点记下来，形成一套反应模式。再遇到叠衣服的任务，它就调出这套模式来执行。 这套方法在实验室里跑得很漂亮。只要测试环境和训练时差不多，衣服能叠得整整齐齐。但场景一换，光线变了、衣服款式不一样了、桌子高一点矮一点，它就容易当场崩溃。 问题出在哪儿？它没有真正理解环境背后的规律。它学会了拿杯子这个动作，但不知道杯子为什么会掉下来；它知道盘子要放桌上，但不知道半个盘子悬空就要摔了。它记住的是动作的样子，不是动作背后的因果。一旦环境跟训练数据对不上，它立马就懵。 话说回来，VLA 这套思路并不是走错了路，反而是目前行业能找到的最好的方向。 在 VLA 出现之前，机器人的各项能力都是分家的。VLA 第一次把视觉、语言、动作这三件事连到了一起，让机器人具备了综合意义上的感知加执行能力。这是一次很关键的跃迁，也正因为有这个突破，行业才愿意在这条路上投入这么多年。 只是跃迁归跃迁，短板还是在那儿。自变量之前发布的 WALL-A 模式，其实也是类似的 VLA 架构。 那怎么办？ 在聊自变量的新模型之前，先说一下行业里其他公司在 VLA 基础上探索新的思路。很多团队的做法，是在 VLA 之外再加一个世界模型模块，让两套系统协作。世界模型的核心能力是预测环境，就像我们人在伸手拿杯子之前，脑子里其实已经预演过一遍整个动作了。所以大家想，把世界模型加进去，机器人的理解能力不就上来了吗？ 听起来挺合理，但其实不是这么回事。这种模块化拼接的方式，还是会损失信息，本质上只是一个过渡方案。 打个比方。苹果在自研芯片出来之前，电脑里的 CPU、GPU、内存是分开的几块芯片。用过那代 Intel 芯片 Mac 的人都有体会，卡顿、发烫是家常便饭。 核心问题就出在，信息要在几个芯片之间来回倒腾，损耗很大。后来 Apple Silicon 干了什么？把这些东西统一集成，数据不用再跨模块搬运，性能和能耗立刻上了一个台阶。 这次自变量发布的 WALL-B，干的就是类似 Apple Silicon 的事情。 WALL-B 的架构叫世界统一模型，英文缩写 WUM。思路其实很简单，不再把视觉、语言、动作拆成三块分开处理，也不靠模块之间传话，而是从训练的第一天起，就把所有这些能力放进同一个模型里一起训练。相当于以前是多芯片架构，数据要在不同模块之间来回切换；现在是统一架构，所有能力在同一个系统里直接协同。 这样做带来的好处有三点。 第一是原生多模态。机器人不再是先看、再理解、再行动这种流水线式的操作，而是看的同时在理解，理解的同时在决定怎么动。整件事是同时发生的，跟我们人看到一张图不需要先翻译一遍是一个道理。 第二是对物理世界的理解。机器人不再只是识别出一个物体，而是开始理解物体背后的物理规律。比如重力、惯性、摩擦力。这些都是底层的物理规律，不管家里什么布局、桌子什么材质，规律都一样。 盘子悬在桌边会掉下来，这件事跟盘子长什么样、桌子是木头还是大理石没关系。 第三是自我进化。机器人跟真实世界交互的过程中，能自动调整。比如杯子第一次没抓紧，它能立刻反应过来，下一次就知道该用多大的力。 架构说得差不多了，还有一个绕不开的问题，就是数据。因为模型最后的天花板，很大程度上是由数据决定的。 具身智能行业的数据大致分两类。一类是实验室里的干净数据，摆拍得整整齐齐，光线灯光都安排好。另一类是家庭环境里的真实数据，乱的，复杂的，充满意外的。前一类数据不缺。后一类稀缺，而且难收集。 过去很长一段时间，行业里大家都在琢磨，能不能想办法绕开家庭真实数据这一关，比如用合成数据、用虚拟环境代替。 自变量给出的判断是，真实数据绕不开。也许最难的那条路，恰恰是最简单的路。想让机器人更快迭代，只有一个办法，尽可能多地收集真实环境的数据。 所以他们现在做的事，就是把机器人直接带进家庭，在真实的家里采集数据，用这种复杂、混乱、不可预测的环境去训练模型。按他们的时间表，下个月新机器人就会正式进入首批用户家里，长期运行，收集真实世界的反馈。 首批体验用户的招募现在已经对外开放，感兴趣的朋友可以去关注一下，说不定你就是第一批家里有机器人保姆的人。 不过具体到机器人进家庭，其实很多人的第一反应是隐私。王潜在现场也专门讲了这一点。他说自己能理解这种担心，换做他自己，也怕一个会动会听会说话的东西进到家里，偷看到晚上11点穿着睡衣坐在沙发上吃方便面的样子。 所以他们也针对性做了处理。比如机器人看到的画面，在设备里就直接打码处理了，原始图像根本不往外传，也不上传云端。要不要开机，也得主人自己按下同意键，不会偷偷启动。每台机器人只认自己的主人，察觉到奇怪的指令就会自动锁定。 最后再说两句感性的话。 我在现场看到自变量团队做的机器人发展史。在 2010 年之前的几十年里，机器人舞台上重要的创新，都是美国公司和日本公司主导。到了 2010 年之后，我才零星看到了中国公司出现。 然后在 2020 年之后，中国公司开始密集出现。当时我对着那张图看完，还是非常有冲击感，我能感觉到在这么重要的领域，中国的创业公司开始深度参与到创新的洪流当中，还是蛮兴奋。