DogeDesigner@cb_doge · 5月3日12

The future begins with Optimus on the Moon.

译未来始于Optimus在月球上。

Rohan Paul@rohanpaul_ai · 5月2日49

Real homes in China are now data labs for robotics. A hybrid human-robot home cleaning service is already in operation. Robot handles the repetitive stuff, while the person focuses on the rest. A preview of the everyday future we will be living soon.

译中国的真实家庭如今已成为机器人的数据实验室。 一种人机协作的家庭清洁服务已投入运营。机器人处理重复性工作，而人员则专注于其余事务。 这是我们即将步入的日常未来的预览。

Chubby♨️@kimmonismus · 5月2日47

It was only a matter of time before Frontier Labs turned its attention to humanoid robotics. While AI is already making waves in the white-collar sector, the blue-collar sector is the next logical step. Google is already working directly on robotics with Gemini. Just a year ago, OpenAI had a collaboration with a robotics company, but that fell through, and Meta is now pursuing robotics.

译前沿科技实验室正将焦点转向人形机器人领域，标志着AI应用从白领工作向蓝领领域的逻辑延伸。科技巨头已展开布局：Google正利用Gemini直接研发机器人；OpenAI此前与机器人公司的合作虽已终止，但Meta正积极进军该领域。Meta近期收购了专注于机器人AI模型开发的初创公司Assured Robot Intelligence，以推进其人形技术重大计划。

TestingCatalog News 🗞@testingcatalog · 5月2日55

Meta acquired Assured Robot Intelligence, a startup developing AI models for robots. Will we see Metabots soon? 🦾🤖

译Meta收购了专注于为机器人开发AI模型的初创公司Assured Robot Intelligence。此举是Meta一项重大计划的一部分，该计划旨在构建人形机器人技术。此次收购引发了外界对Meta即将推出“Metabots”人形机器人的猜测，标志着Meta正积极将其在人工智能领域的优势扩展至实体机器人领域。

Rohan Paul@rohanpaul_ai · 5月1日43

The LongCat team just released LARYBench, a benchmark built to test whether an AI model truly learns action from video, instead of only looking good when attached to a robot policy later. It evaluates latent actions, meaning the hidden motion signals a model extracts from video, across 1.2M+ clips, 620K+ image pairs, 595K trajectories, 151 action classes, and 11 robot platforms. A latent action representation tries to store the change between frames as something like reach, pick, place, move left, or close gripper, rather than memorizing raw pixels. The key point is that robot training data is scarce, while human and robot videos are abundant, so the whole field wants a way to turn cheap video into useful action knowledge. The paper argues that older evaluations mixed too many things together, because a robot succeeding on a task depends on the policy, training recipe, environment, and controller, so you could not tell whether the action representation itself was actually good. LARYBench splits the problem into 2 cleaner tests, where one asks whether the representation knows what happened and the other asks whether it preserves enough detail for how to move. The biggest result is that general self-supervised vision models beat specialized embodied LAMs, with V-JEPA 2 reaching 76.62% average action classification accuracy, while DINOv3 gives the best overall control regression score at 0.19 MSE, far ahead of embodied models clustered around 0.87 to 0.97. The deeper point is that strong visual representations already contain a surprising amount of action knowledge, and the paper also shows that latent feature spaces map to robot control better than pixel reconstruction spaces, which helps explain why some robotics systems may be building on the wrong intermediate representation. 🧵 1.

译LongCat团队推出LARYBench基准，旨在评估AI模型是否从视频中真正学习动作，而非仅在后端机器人策略中表现良好。该基准聚焦模型从视频提取的潜在动作表示，通过超过120万视频片段等数据，将评估拆分为动作分类与控制回归两个清晰测试。关键发现是，通用自监督视觉模型（如V-JEPA 2和DINOv3）表现优于专用具身模型，表明强大视觉表示已蕴含丰富动作知识，且潜在特征空间比像素重建更利于机器人控制映射。这为利用丰富视频数据解决机器人训练数据稀缺问题提供了新方向。

Rohan Paul@rohanpaul_ai · 5月1日62

Bloomberg: 1X just opened a 58,000sqft California factory and says it will build 10,000 home humanoid robots in its first year. Plans to reach 100,000 robots by end-2027 and begin shipments by end-2026, --- bloomberg. com/news/articles/2026-04-30/humanoid-maker-1x-opens-us-factory-plans-to-make-10-000-home-robots-this-year

译Bloomberg：1X 刚刚在加利福尼亚开设了一家 58,000 平方英尺的工厂，并表示将在第一年制造 10,000 台家用仿人机器人。 计划到 2027 年底达到 100,000 台机器人，并于 2026 年底开始发货， --- bloomberg.com/news/articles/2026-04-30/humanoid-maker-1x-opens-us-factory-plans-to-make-10-000-home-robots-this-year

Berryxia.AI@berryxia · 4月30日52

🦾 想象一下： 工厂的流水线灯火通明，一个银灰色的身影悄然走来。它只有123cm高，却拥有15-31个自由度，像科幻电影里走出来的“未来伙伴”。 它叫 Unitree 双臂人形机器人。 今天，Unitree正式宣布：售价$4290！ 它不是冰冷的机器。 它有双目立体视觉，能“看”清深度； 有阵列麦克风和扬声器，能听懂你的自然语言；双臂灵活如人类，可秒换两指爪、三指爪甚至灵巧人手。 拖拽示教、底层全开放SDK，想怎么玩就怎么玩。 它轻量、紧凑，却能上流水线装配零件、实验室精准操作、甚至未来帮你做家务…… 从科研到工业，从教育到创业，它把“人形机器人”这四个字，从科幻拉进了普通人的现实。 这不是遥不可及的梦想，这是你现在就能买来、改装、一起创造未来的起点。 机器人革命，从这一刻，真正开始“平民化”。 ？ 你准备好迎接你的第一个AI搭档了吗？

译Unitree正式推出双臂人形机器人，售价为4290美元，标志着人形机器人进入“平民化”新阶段。该机器人高123厘米，具备15-31个自由度，集成了双目立体视觉、阵列麦克风和扬声器，支持自然语言交互。其双臂灵活，末端执行器可快速更换，并支持拖拽示教和底层全开放SDK，便于开发。设计轻量紧凑，目标应用场景涵盖工业装配、实验室操作乃至未来家务，旨在为科研、教育及创业等领域提供可及的人形机器人平台。

1X@1x_tech · 4月30日24

Building Your NEO

译构建你的NEO

Chubby♨️@kimmonismus · 4月30日36

$4290 for a dual-arm robot. Holy.

译4290美元买一个双臂机器人。 天啊。

Chubby♨️@kimmonismus · 4月29日51

Holy. Thousands of RobotEra L7 humanoid robots to enter service across 10+ logistics centers performing sorting tasks. And people still think their blue collar jobs are safe.

译天啊。数千台 RobotEra L7 人形机器人将进驻 10 多个物流中心，执行分拣任务。 而人们依然认为他们的蓝领工作很安全。

Rohan Paul@rohanpaul_ai · 4月29日34

On polished tiles with minimal grip, this robot dog demonstrates impressive foot placement and balance. It appears to use real-time adjustments, keeping its body level as it moves naturally despite the challenging surface.

译在抓地力极小的抛光瓷砖上，这只机器狗展现了令人印象深刻的脚步定位与平衡能力。 它似乎运用了实时调整技术，在具有挑战性的地面上自然移动时，始终保持身体水平。

宝玉@dotey · 4月29日61

港科大、新加坡国立、牛津等十余所高校联合发了一篇 88 页的综述，试图解决一个越来越尴尬的问题：“世界模型”这四个字在不同圈子里意思完全不同。 做强化学习的说的是 Dreamer 那种在脑子里想象未来再决策的系统，做视频生成的说的是 Sora 那种画面生成器，做 Web Agent 的说的是 LLM 对网页状态的预测能力。大家各说各的，论文之间根本没法比。这篇论文提出了一个“能力等级 x 领域法则”的二维框架，想把这些散落的社区拉到同一张地图上。 能力分三层： L1 预测器只管预测下一步， L2 模拟器能做多步推演且推演结果要遵守所在领域的基本规则（物理世界不能让杯子穿桌子，软件世界不能调不存在的 API）， L3 进化器能在预测出错时主动诊断原因、设计实验获取新数据、修正自身模型。 领域分四类： 物理世界（机器人、自动驾驶）、 数字世界（网页浏览器、代码、GUI）、 社会世界（多智能体交互、社会模拟）、 科学世界（天气预报、材料发现、药物设计）。 每个领域对“什么算合法的状态转换”有完全不同的约束，失败模式也不同。 综述 400 多篇工作后有几个跨领域的共性发现： 视频生成模型视觉逼真度远超物理忠实度，最好的模型在物理一致性测试上通过率只有 26.2%； LLM 驱动的社会模拟能复现舆论极化等涌现现象，但智能体有系统性的趋同偏差，跟真实人类行为分布偏离大； 目前 L3 做得最成熟的是自动化科学实验，比如 A-Lab 用机械臂在 17 天内完成 353 次闭环实验合成了 36 种化合物，失败实验被提炼为持久知识而非丢弃。 论文末尾有个观察： 从牛顿定律到麦克斯韦方程，人类历史上最成功的世界模型全是符号化的、可直接修改和组合的。 现在的神经网络把所有规则藏在权重里，L1 和 L2 阶段够用，但到了需要修改模型结构本身的 L3，这种隐式表示就成了障碍。 世界模型的终局，是越来越大的神经网络，还是终究要回到可编辑的符号规则？

译针对“世界模型”概念在强化学习、视频生成等不同社区含义割裂的问题，一篇由港科大、新加坡国立、牛津等多所高校联合发表的综述提出了“能力等级×领域法则”的二维统一框架。能力分为L1预测器、L2模拟器和L3进化器；领域涵盖物理、数字、社会和科学世界。综述发现，当前视频生成模型物理一致性差，LLM社会模拟存在行为偏差，而L3能力在自动化科学实验中最成熟。论文最后指出，神经网络隐式表示的规则在需要自我修正的L3阶段可能成为障碍，并引发了对世界模型终局形态的思考。

Rohan Paul@rohanpaul_ai · 4月29日47

Japan tests humanoid robots at Tokyo Haneda Airport to tackle labour shortage crisis These are Unitree G1 and taller UBTECH Walker E robots for tough jobs like loading cargo and moving luggage. Japan's aviation sector struggles with staff shortages due to more tourists visiting and fewer young workers available. JAL has about 4000 people doing ground handling jobs. The airline wants the robots to take on tough physical tasks to help employees. They may later use them to clean plane cabins and run other support machines.

译为应对因游客激增和年轻劳动力不足引发的航空业人力危机，日本在东京羽田机场测试人形机器人。测试机型包括Unitree G1和UBTECH Walker E，旨在承担装载货物、搬运行李等繁重体力劳动。日本航空希望借此减轻约4000名地勤人员的工作负担，并计划未来将机器人用于清洁客舱和操作其他辅助设备。

Epoch AI@EpochAIResearch · 4月29日41

Alongside our report on how fast robot production could scale up, we've also released a series of @metaculus forecasting questions on the future of robotics. If you have views on where these trends are headed, come weigh in!

译在我们关于机器人生产能多快实现规模化的报告之外，我们还发布了一系列关于机器人未来的@metaculus预测问题。 如果你对这些趋势的发展方向有见解，欢迎参与讨论！

Rohan Paul@rohanpaul_ai · 4月28日33

Close call in a robot demo: leg extension during performance passed inches from a boy's face.

译机器人演示中的惊险一刻：表演过程中的腿部伸展距离一个男孩的脸仅几英寸。

Rohan Paul@rohanpaul_ai · 4月28日39

See the handover in this Chinese robot-run convenience store. The G1 walks to the refrigerated section post-payment, uses its grippers to grasp, then returns to the counter. Early commercial pilots replacing cashier roles with robots in 24/7 retail.

译看看这家中国机器人运营便利店中的交接流程。G1在付款后走向冷藏区，用它的抓取器拿取商品，然后返回柜台。 早期商业试点正在用机器人取代24/7零售中的收银员角色。

Epoch AI@EpochAIResearch · 4月28日48

How fast could production of humanoids, quadrupeds, drones, and other robots scale up, in the event of a large demand shock?

译如果遭遇大规模需求冲击，人形机器人、四足机器人、无人机和其他机器人的生产速度能多快提升？

AK@_akhaliq · 4月28日53

Agentic World Modeling Foundations, Capabilities, Laws, and Beyond paper: https://huggingface.co/papers/2604.22748

译能动世界建模 基础、能力、法则与超越 论文: https://huggingface.co/papers/2604.22748

Chubby♨️@kimmonismus · 4月27日44

Kinetix AI has teased KAI, its new humanoid robot -engineered with 36 degrees of freedom, more than any previous model, a hybrid dexterous hand, and 18,000 sensors embedded across a soft, flexible body-positioning it as the most human-like robotic system to date. Getting closer to real humanoid robots each day.

译Kinetix AI 已预告其新款人形机器人 KAI——拥有 36 个自由度，超越以往所有型号，配备混合灵巧手，并在柔软灵活的身体中嵌入 18,000 个传感器，使其成为迄今为止最类人的机器人系统。 我们距离真正的人形机器人日益接近。

Rohan Paul@rohanpaul_ai · 4月27日34

Balancing and athletic maneuvering in robots have advanced quickly. In this OmniXtreme demo, a Unitree G1 runs at speed, drops into alternating pistol squats, then recovers with controlled acrobatic rolls.

译机器人的平衡与运动操控技术进展迅速。 在这段OmniXtreme演示中，Unitree G1快速奔跑、交替完成手枪式深蹲，随后通过受控的体操翻滚恢复姿态。

Rohan Paul@rohanpaul_ai · 4月27日44

A Chinese company made this desktop blue-eyed companion robot. Exhibits lifelike micro-expressions, eye tracking, and responsive head positioning. Multiple degrees of freedom, fluid neck motion for natural engagement, with eye-mounted cameras.

译一家中国公司制造了这款桌面蓝眼伴侣机器人。 展现出栩栩如生的微表情、眼球追踪和响应式头部定位。多自由度，流畅的颈部运动实现自然互动，并配备眼部摄像头。

小互@xiaohu · 4月27日62

兄弟们手机上现在能跑世界模型了 蚂蚁灵光 App 今天上线"体验世界模型"的功能 上传一张图，等个几秒到几十秒，就能生成一个可以拿摇杆走进去的 3D 世界。 手机的算力怎么可能支撑这件事？ 应该是云端生成，但是速度这么快，也是牛P，不过效果没有那么惊艳，但是可以玩... 我连测了四张图都进去了...

译蚂蚁灵光App推出“体验世界模型”新功能，用户上传图片后，可在几秒到几十秒内生成一个可供探索的3D世界。该功能疑似主要依赖云端算力完成生成，尽管手机本地算力有限，但其生成速度仍被评价为“牛P”。目前生成效果虽未达到“惊艳”程度，但已具备可玩性，用户实测可成功进入多个由图片生成的3D场景。

Rohan Paul@rohanpaul_ai · 4月27日38

Unitree G1’s balancing tech is on full display here. Faced with an abrupt destabilizing motion, its integrated sensors and adaptive control loop execute the necessary shifts in posture and foot placement to recover equilibrium effortlessly.

译Unitree G1的平衡技术在此得到充分展示。 面对突如其来的失衡动作，其集成传感器与自适应控制回路会执行必要的姿态调整与足部位置移动，从而轻松恢复平衡。

elvis@omarsar0 · 4月27日54

Here is a very common problem when building complex agents. Long-horizon agents (in particular) fail in two ways: the decision-maker can't decompose well, or the skill library goes stale. This new research tackles both at once. The paper introduces a co-evolution framework where an LLM decision agent and a dynamic skill bank improve each other through iterative refinement. The decision agent picks and chains skills. Performance feedback updates both the policy and the skills. New skills emerge by generalizing successful sequences instead of being hand-coded upfront. Why does it matter? Most long-horizon agent stacks treat skills and decision-making as separate optimization problems, which is why they plateau. Co-evolution gives you adaptive planning and a growing library of reusable behaviors from a single loop, which is what you actually want when task structure isn't predetermined: robotics, game agents, complex planning. Paper: https://arxiv.org/abs/2604.20987 Learn to build effective AI agents in our academy: https://academy.dair.ai/

译构建复杂智能体时，长期任务智能体常因决策者分解能力不足或技能库过时而失败。新研究提出一种协同进化框架，让LLM决策智能体与动态技能库通过迭代优化共同改进。决策智能体负责选取和串联技能，性能反馈同时更新其策略和技能库本身。新技能通过归纳成功序列自动生成，而非预先手动编码。传统方法将技能与决策作为独立问题优化，容易陷入瓶颈。协同进化则能在单一循环中实现自适应规划，并持续增长可复用行为库，这对任务结构不确定的领域（如机器人、游戏智能体、复杂规划）至关重要。

AI Notkilleveryoneism Memes ⏸️@AISafetyMemes · 4月26日51

AI can now generate novel viruses WHY THIS MATTERS: 1) Crazy people COULD use AI to make superviruses NOW, but most of them are idiots 2) Dario Amodei thinks in 6-12 months, even idiots may be able to 3) In 6-12 months, the world could shut down. You die. Your family dies. 4) This is a SUPER OBVIOUS risk unless you think AI progress is just going to magically stop soon 5) Maybe it's not 6-12 months, maybe it's 12-36 months, but... so what?? That's barely any time to prepare! 6) Yes, AI could accelerate defense faster. COULD. We shouldn't bet civilization on that. Yes, AI could make vaccines, but historically it's wayy harder to make vaccines than viruses - and you need to actually make billions of vaccines, distribute them to billions of people, etc. That takes a long time! A virus could easily infect the world before that happens! 7) This industry remains less regulated than a taco cart. Big AI has staved off regulation using the Big Tobacco playbook.

译当前AI已能生成新型病毒，斯坦福与Arc研究所实验显示语言模型成功设计出包括使用未知蛋白质的活性病毒。Anthropic CEO Dario Amodei预测6-12个月内即使非专业人士也可能具备该能力，而疫苗研发与分发速度远不及病毒传播。AI防御虽可能加速，但不应以文明存亡为赌注。该领域监管严重滞后，大型科技公司沿用烟草行业策略阻碍立法，全球性生物风险窗口期可能短至12-36个月。

阿绎 AYi@AYi_AInotes · 4月25日43

讲真，这可能是第一个我真的愿意买回家的机器人， 前斯坦福机器人博士@aaronistan Tan刚刚发布了Lume，它长得完全就是一盏普通的豪华床头灯， 但是会伸出机械臂帮你叠衣服，床上扔得乱七八糟的T恤裤子， 它会一件一件抓起来叠得整整齐齐，折完机械臂收回去，继续当你的灯， 最狠的是他们下周在Palo Alto办展示会，直接让大家自带脏衣服去测， 我之前也觉得人形机器人才是未来，看完才反应过来，这才是家用机器人该有的样子， 它不会在你家里走来走去，不会盯着你看，不会让你觉得家里多了个陌生人， 它就是一件会干活的家具，解决一个所有人都讨厌到死的问题， 这真的也很酷啊，比任何能跳舞能跑的人形机器人，都更接近我们真正想要的未来。

译前斯坦福机器人博士Aaron Tan发布的Lume机器人，外观设计为一盏豪华床头灯，但内置机械臂可自动叠放衣物。其核心理念是家用机器人应像一件会干活的家具，专注于解决叠衣服这类具体痛点，而非以人形形态在家庭中活动。团队将于下周在Palo Alto举办展示会，邀请公众自带脏衣服进行测试，强调产品解决实际问题的实用性，认为这种形态比追求拟人化功能更接近真正的家用机器人未来。

Chubby♨️@kimmonismus · 4月25日39

My last day in China. A reflection and summary. I have now spent four days visiting China (well, actually five days, because turkish airlines just cancelled my flight lol). It was a breathtaking experience. I saw and learned an incredible number of things, met wonderful people, had fascinating conversations, and, above all, got to know an entirely different culture. I was invited to China by XPENG and had the opportunity to visit both their headquarters and their factory on site. I already said this in another post, but what left a lasting impression on me, and at the same time makes me look at Europe with concern, is that China has a different mentality of creativity. They are experimenting with flying cars, working on FSD EVs at the same time, and developing humanoid robots; an entire product portfolio, all in parallel. While high technology once came from Europe, especially from Germany, that trend has shifted. Of course, China still depends on Western technology in certain areas. But first, it is trying to build independent production facilities and supply chains, and second, it is using this technology to go beyond what already exists. I think this can be said quite objectively. Does that mean China has already won the global competition? Not at all. But it has become an outstanding competitor. My subjective impressions of the infrastructure only reinforced this: a well-developed, modern road network, no potholes as far as I could tell, widespread digitalization, almost everything via QR codes, WeChat and Alipay, and a very high degree of domestically built and developed EVs. Germany, meanwhile, still largely wants to build combustion engines, and the broader zeitgeist still struggles with electromobility. Everyday reality embarrasses itself with problems that feel like they belong to the 20th century: letters and faxes instead of emails, physical visits to government offices instead of digital administration, infrastructure and highways plagued by never-ending repair work. Germany feels like an aging behemoth refusing to accept the future and move with the times. In many areas, China has already overtaken Germany, even if only through the strategic decision to become sovereign and independent in energy policy, something that, by contrast, is becoming almost disastrous for Germany and Europe. Does that mean everything in China is good? Certainly not. Still, for me as a European, it was a culture shock to see how outdated and backward-looking Europe appears. While China is expanding and developing its own chip production, such as Huawei Ascend, in order to become more independent from the US and NVIDIA, and while it is pushing forward into robotics and high tech, Europe seems delighted to congratulate itself on the next AI regulation or, as recently, on the misuse of emojis for supposedly illegal activities. Europe has lost touch. Full stop. But ultimately, these are political decisions. Europe has to choose: future or past, regression or progress, combustion engines, letters and regulation, or EVs, digitalization and data center expansion. Quo vadis, Europe? Time is running out. Disclaimer: I have not received any payment from XPENG; no money was transferred to me; my opinion is subjective and was expressed without compensation. P.S. I already had a negative view of developments in Europe and Germany before my trip. So it is not as if this trip fundamentally changed my view of Germany and Europe. It only made even clearer to me how much stagnation there is in Europe.

译作者受小鹏汽车邀请访华，对其并行研发飞行汽车、全自动驾驶电动车和人形机器人的创新格局印象深刻。中国基础设施先进，数字化全面普及，电动汽车国产化率高。相比之下，德国乃至欧洲仍沉迷于内燃机，基础设施老化，数字化进程迟缓，且热衷于监管而非技术进步。中国正通过华为昇腾等芯片研发寻求技术独立，欧洲却显得与时代脱节。作者认为欧洲必须在回归过去与拥抱电动汽车、数字化等未来产业间做出明确选择，时间已然紧迫。

Chubby♨️@kimmonismus · 4月25日31

Bolt by MirrorMe | Claims speeds of 11m/s indoors, 10.09 m/s outdoor so far (Usain Bolt's top speed is 12.42 m/s) Now robots outrun the fastest humans on earth.

译Bolt by MirrorMe | 宣称室内速度达11米/秒，室外目前达10.09米/秒（尤塞恩·博尔特最高速度为12.42米/秒） 如今机器人已能超越地球上最快的人类。

Rohan Paul@rohanpaul_ai · 4月22日

Robotics is about to reshape lives. At the Canton Fair, China, a woman with severe knee issues who could barely stand or walk tested a Chinese waist-assisted exoskeleton. Then she took steady steps without support. Her friend broke into tears.

译机器人技术即将重塑生活。 在中国 Canton Fair，一位膝盖有严重问题、几乎无法站立或行走的女性测试了一款中国腰部辅助外骨骼。 然后她在没有支撑的情况下迈出了稳健的步伐。 她的朋友泪流满面。

Rohan Paul@rohanpaul_ai · 4月22日

Toyota’s CUE7 humanoid appeared mid-game at Toyota Arena Tokyo. In front of a live crowd, it handled a pass, dribbled smoothly & executed free throws. Reinforcement learning lets it improve shooting accuracy through repeated trials & real-time feedback

译Toyota 的 CUE7 人形机器人在 Toyota Arena Tokyo 比赛期间中场亮相。在现场观众面前，它完成接球、流畅运球并执行罚球。 强化学习使其能够通过反复试验和实时反馈来提高投篮准确率。

Rohan Paul@rohanpaul_ai · 4月22日

During Beijing’s humanoid robot half-marathon, Galaxea’s R1 wheeled robot stood at a water station table. Runners approached, and the robot’s arms extended to pick up and pass plastic bottles

译在北京人形机器人半程马拉松期间，Galaxea 的 R1 轮式机器人站在水站桌子旁。跑者靠近时，机器人的手臂伸出，拿起并递出塑料瓶。

Chubby♨️@kimmonismus · 4月22日18

Flying cars feel so futuristic. What China has over Germany in automobile production is the willingness to try new things and to take creative paths.

译飞行汽车感觉如此未来感。中国在汽车生产上超越德国之处，在于愿意尝试新事物并走创新道路。

Chubby♨️@kimmonismus · 4月22日

Jeff Bezos’s secretive startup, Project Prometheus, is reportedly raising $10 billion at a $38 billion valuation, showing an ambitious push into “physical AI”,systems designed to interact with real-world industries like manufacturing and aerospace rather than just digital tasks. Led alongside Vik Bajaj, the company is quietly building talent from rivals like OpenAI and Google DeepMind, hinting at serious competitive intent. Next big competitor on the rise?

译Jeff Bezos 神秘的初创公司 Project Prometheus 据报道正以 380 亿美元估值融资 100 亿美元，展现出向 "physical AI" 进军的雄心——这类系统旨在与制造业和航空航天等现实世界行业互动，而非仅仅处理数字任务。 在 Vik Bajaj 的联合领导下，该公司正悄悄从 OpenAI 和 Google DeepMind 等竞争对手处招募人才，暗示着强烈的竞争意图。 下一个即将崛起的重要竞争者？

Rohan Paul@rohanpaul_ai · 4月22日

ROBOTIS unveiled AI Sapiens this week: a 34 kg humanoid that maintains balance even during dynamic shifts and quick leg lifts. The open-source system is meant as a shared system for physical AI work, with full hardware and software released publicly.

译ROBOTIS 本周发布了 AI Sapiens：一款 34 公斤的人形机器人，即使在动态移动和快速抬腿时也能保持平衡。该开源系统旨在作为实体 AI 工作的共享平台，完整的硬件和软件均已公开发布。

Rohan Paul@rohanpaul_ai · 4月22日

GenRobot is packaging multimodal robot data collection into wearable hardware. They just launched a 6-camera bionic wearable to capture embodied AI data, addressing common blind spots in traditional monocular setups, such as occlusion and precise hand-object timing. That matters because embodied models do not learn from pixels alone. They learn from synchronized structure: head pose, hand motion, scene layout, and action timing living on the same clock. DAS Ego utilizes six 2MP cameras to achieve a zero-distortion 270° horizontal and 150° vertical FOV. This enables mm-level trajectory reconstruction and ultra-low latency (<1ms) head-hand coordination. What GenRobot is building is a cleaner way to record natural human interactions. To prove it, they open-sourced "Gen Ego Data," a first-person, human-centric dataset covering 20+ environments and 200+ skills. By capturing authentic interactions, it helps models learn physical laws and "perception-action-outcome" causality, providing core data support for real-world embodied AI deployment.

译GenRobot推出DAS Ego六摄像头仿生可穿戴设备，以270°零畸变视场角与毫秒级头手同步解决传统单目方案的遮挡与深度盲区，实现毫米级轨迹重建与厘米级关节追踪。同步开源的Gen Ego Data数据集涵盖20余环境及200余项技能，通过第一人称视角采集帮助具身AI模型学习物理规律与"感知-动作-结果"因果关系，为真实场景部署提供核心数据支撑。

AK@_akhaliq · 4月22日47

Agent-World Scaling Real-World Environment Synthesis for Evolving General Agent Intelligence paper: https://huggingface.co/papers/2604.18292

译智能体世界 为进化通用智能体智能而扩展真实世界环境合成 论文: https://huggingface.co/papers/2604.18292

Rohan Paul@rohanpaul_ai · 4月20日

Unitree’s H1 made its debut in 2026 Beijing Humanoid Half-Marathon. It had supplemental ice backpacks to handle actuator heat during extended running, one unit tumbled on the course. Crews refreshed the cooling pack on the spot and it continued.

译Unitree H1 在 2026 Beijing Humanoid Half-Marathon 首次亮相。它配备了辅助冰背包以应对长时间奔跑时执行器产生的热量，一台机器人在赛道上摔倒。工作人员当场更换了冷却包，它继续比赛。

DogeDesigner@cb_doge · 4月19日

"The good future of AI is one of immense prosperity where there is an age of abundance, no shortage of goods and services. Everyone can have whatever they want, unless, except for things we artificially define to be scarce, like some special artwork. But anything that is of manufactured good or provided service, will, with the advent of AI plus robotics, that the cost of business services will trend to zero. I'm not saying it'd be actually zero, but it'll be everyone will be able to have anything they want." 一 Elon Musk

译Elon Musk 展望 AI 驱动的未来社会将步入极度繁荣的丰裕时代。随着 AI 与机器人技术融合，除人为定义的稀缺品（如特定艺术品）外，几乎所有制造品和服务的成本都将趋近于零。这意味着物质短缺将被消除，每个人都能获得所需的商品与服务，实现普遍的富足。尽管成本不会绝对归零，但技术将确保"人人可得"成为现实。

Chubby♨️@kimmonismus · 4月19日

This Humanoid Robot just finished a Half-Marathon in 50 minutes and 26 seconds. The point is not so much that robots are now running marathons, but that it has become a national competition. It's about the culture, the spirit, that humanoid robots can be introduced into everyday life and become an event. Robots and AI should become ubiquitous and enthusiasm for their development should be encouraged. That's why such events are organized. And this spirit is missing in Europe, for example.

译这款人形机器人刚刚以50分26秒完成了一场半程马拉松。 重点不在于机器人现在能跑马拉松，而在于这已成为一项全国性竞赛。这关乎文化，关乎精神，即人形机器人可以被引入日常生活并成为一项赛事。 机器人和AI应当变得无处不在，对其发展的热情应当受到鼓励。这正是举办此类活动的原因。而这种精神，例如在欧洲，是缺失的。

Rohan Paul@rohanpaul_ai · 4月19日

Humanoid acceleration has started. The inevitable is not asking for permission.

译人形机器人加速已经开始。 必然之事无需许可。