宝玉@dotey · 6月17日66

NVIDIA GEAR 实验室发布了 ENPIRE，一套让 AI 编程 agent 自主操控真实机器人做实验的系统。是在真实物理世界里跑，让 AI 自己重置场景、跑实验、判断成败、查论文、改代码、再来一轮，人类只需要早上起来看报告就行。 ENPIRE 的全称是 Agentic Robot Policy Self-Improvement in the Real World，由 NVIDIA GEAR 实验室联合 CMU、UC Berkeley 共同开发。Jim Fan 是 GEAR 实验室的联合负责人，他把这个方向叫 AutoResearch，意思是让 AI 来做科研。之前这个概念只在纯数字环境里跑通过，ENPIRE 是第一次把它搬到了物理世界。 系统的运作方式是这样的：给一组 AI 编程 agent 分配一队机器人、一批 GPU 和足够的 token 预算，然后设定一个目标，比如“尽快完成这个任务，让机器人保持忙碌但别出事”。接下来人就可以走了。 agent 会自动做四件事： 1. 搭建环境：包括自动重置场景和自动判断成败 2. 改进操控策略：从启发式学习到行为克隆到强化学习都会尝试 3. 在真实机器人上跑实验收集数据 4. 分析失败原因并迭代改进代码 整个循环完全自主，不需要人参与。 ENPIRE 测试了几个相当刁钻的操作任务：把细小的 pin 脚精确插入盒子、穿扎带、用剪刀剪扎带、把 GPU 插到主板上。这些都是需要高精度灵巧操作的任务，不是简单的抓取搬运。最终 agent 自主训练出的策略达到了 99% 的成功率（pass@8 标准）。 项目同时测试了三款编程 agent： 1. OpenAI 的 Codex（搭配 GPT-5.5） 2. Anthropic 的 Claude Code（搭配 Opus 4.7） 3. 月之暗面的 Kimi Code（搭配 Kimi K2.6）。 三家的 agent 都能跑通整个流程，但在不同任务上的研究进展速度有差异。 一个比较有意思的发现是他们所谓的“物理 scaling law”：8 台机器人并行探索，比 1 台或 4 台的研究推进速度快得多。 更多机器人意味着更多并行实验，agent 之间还能互相参考对方的成功策略，类似于多个研究者同时攻关同一个问题。不过扩大规模也有代价，token 消耗会显著增加，而且 agent 花在总结同伴进展上的时间会变多，机器人的实际利用率反而下降。 所有代码会开源，让你在家也能搭一个自运行的机器人实验室，但你至少得有一队机器人臂和一批 GPU。开源本身的意义在于，其他机器人实验室可以直接复用这套 agent 驱动的实验框架，省掉大量人工盯实验的时间。

译NVIDIA GEAR 实验室联合 CMU、UC Berkeley 发布 ENPIRE，让 AI 编程 agent 自主操控真实机器人做实验。系统自动搭建环境、重置场景、改进操控策略（从启发式到强化学习）、收集数据、分析失败并迭代代码，全程无需人类参与。高精度测试任务（细针脚插入、穿剪扎带、安装 GPU）中自主训练策略达 99% 成功率（pass@8）。测试了 Codex、Claude Code、Kimi Code 三款 agent，发现“物理 scaling law”：8 台机器人并行探索比 1 台或 4 台推进快得多。所有代码将开源。

jason@jxnlco · 6月17日65

codex you say????

译NVIDIA GEAR实验室推出ENPIRE项目，首次实现物理世界自动研究。8个Codex智能体被授权控制机器人舰队、GPU及token预算，自主完成绑扎带、整理精密插针、安装GPU等高精度任务。系统无需人类干预，agent可自学视觉线索、复位场景、调参、翻阅论文、辩论反思，并直接硬件试错。团队发现“物理规模扩展”效应：8台机器人并行探索效率显著高于少数台。部分实验室已实现夜间自主改进，每天早上看报告。ENPIRE将全部开源，支持在家搭建自运行机器人实验室。

AYi@AYi_AInotes · 6月17日28

来看看中国宇树科技的机器人跳难度超高的古风舞，这舞我是跳不出来的，感觉几年后没有人类啥事了😂

译仿人机器人无需像人。Eno 是 Genesis 的首款通用机器人——不是假装成人的机器，而是赋予身体的智能。它冷静、能干、乐于助人，将于今年 Q4 面市。

AK@_akhaliq · 6月17日34

μ_0 A Scalable 3D Interaction-Trace World Model

译μ_0 一个可扩展的3D交互追踪世界模型

Jim Fan@DrJimFan · 6月17日64

Today, we enable AutoResearch in the physical world for the first time! Introducing ENPIRE: we give 8 Codex agents a fleet of robots, an allocation of GPUs, and generous token budget. We set them free with a simple goal: solve the task as quickly as possible, keep the robots busy but stay safe, don't waste precious compute. Make no mistake. Then humans step aside and our watch begins. The robot fleet starts to come alive: they learn to look for visual clues, reset the scene, practice novel skills, tinker with control stack, read papers online, debate, reflect, get stuck, and try again directly on the hardware. All we did is to give Codex an API to the world of atoms, and the rest is emergence. ENPIRE is able to solve high-precision tasks like tying zip-ties, organizing fine pins, and installing GPUs all by itself. We also discovered a new type of "physical scaling": 8 robots exploring in parallel improves significantly faster than fewer ones. A part of our NVIDIA GEAR lab now self-improves tirelessly over night. We just read the reports in the morning. /goal: we all take a holiday and Jensen wouldn't even notice ;) We will be open-sourcing everything, so you can host your self-running robot lab at home too! Deep dive in the thread:

译NVIDIA GEAR 实验室首次在物理世界启用 AutoResearch，推出 ENPIRE 项目。给 8 个 Codex 智能体分配机器人舰队、GPU 和 token 预算，目标快速安全完成任务。人类退出后，机器人舰队自主学会寻找视觉线索、重置场景、练习新技能、调整控制栈、阅读论文、辩论反思。ENPIRE 能高精度完成扎带、整理细针、安装 GPU 等任务。发现物理扩展：8 机器人并行探索比少机器人效率显著提升。实验室部分可整夜自我改进，早上读取报告。所有内容将开源。

Rohan Paul@rohanpaul_ai · 6月17日55

Language had a strange advantage robotics does not: Text is already a compressed, shared interface for human thought, while physical action is split across bodies, sensors, surfaces, speeds, and failure modes. $5B + is already betting on world models, $18B has gone into robotics, and yet the field still has no widely trusted shared benchmark, no architecture convergence, and a 100,000-year data gap between robot experience and the data scale behind modern AI. World models are promising because they try to predict what will happen before a robot acts, but prediction alone does not solve data collection, evaluation, real-time control, or deployment reliability. The serious startup opportunities sit in those bottlenecks. Whoever builds the data loops, eval systems, memory layers, inference stack, or vertical deployment engines may shape embodied AI more than the teams arguing over model labels today A great piece from Charlotte Xia (@xia_char)

译Rohan Paul引用Charlotte Xia的博客，讨论Jim Fan的“Great Parallel”论点：具身AI将像LLM一样扩展。与语言不同，文本是压缩共享接口，物理行动分散于不同实体。尽管已有$5B+投资世界模型、$18B投入机器人，领域仍缺乏共享基准、架构收敛，且存在10万年的数据差距。世界模型能预测行动结果，但无法解决数据收集、评估、实时控制和部署可靠性。真正的创业机会在于数据循环、评估系统、记忆层、推理栈和垂直部署引擎等瓶颈。

AYi@AYi_AInotes · 6月17日53

可能要被杠，但我真觉得再过几年普通人的羽毛球水平，连四足机器人都打不过了🐶

Alibaba Cloud@alibaba_cloud · 6月16日70

📣 Introducing the Qwen-Robot Suite — Qwen-RobotNav, Qwen-RobotManip, Qwen-RobotWorld, three foundation models, a full stack for embodied intelligence. 🧭 Qwen-RobotNav — the gateway to mobility. • Unifies 5 navigation tasks in one model: instruction following, point-goal, object-goal, target tracking, autonomous driving • Controllable observation protocol • Tool interface for agentic systems 🤖 Qwen-RobotManip — the foundation of interaction. • Unified state-action space across heterogeneous robots • Camera-frame delta poses for coherent cross-embodiment training • Pretrained on a 38,100+ hour open-source corpus 🌍 Qwen-RobotWorld — infinite worlds for physical agents. • Single world model, 20+ embodiments • Natural-language action interface • Predicts physically grounded futures across manipulation, driving, and navigation Each model is independently useful, and could be composed as physical-world tools.Together, they form the low-level toolkit for general-purpose agentic systems that don't just see the world, but act in it. 📷 Blog: https://qwen.ai/blog?id=qwen-robotsuite 📖 Report： Qwen-RobotNav: https://qianwen-res.oss-accelerate.aliyuncs.com/qwenrobot/papers/Qwen_RobotNav.pdf Qwen-RobotManip: https://qianwen-res.oss-accelerate.aliyuncs.com/qwenrobot/papers/Qwen_RobotManip.pdf Qwen-RobotWorld： https://qianwen-res.oss-accelerate.aliyuncs.com/qwenrobot/papers/Qwen_RobotWorld.pdf

译阿里云推出 Qwen-Robot 套件，包含三个基础模型：Qwen-RobotNav 统一指令跟随、点目标、对象目标、目标追踪和自动驾驶 5 种导航任务，提供可控观测协议和智能体工具接口；Qwen-RobotManip 统一异构机器人状态-动作空间，基于 38,100+ 小时开源语料预训练；Qwen-RobotWorld 单个世界模型支持 20+ 种具身形态，通过自然语言动作接口预测操控、驾驶、导航等物理可行未来。三者可独立使用，也可组合为通用智能体系统的底层物理世界工具。

Qwen@Alibaba_Qwen · 6月16日72

📣 Introducing the Qwen-Robot Suite — Qwen-RobotNav, Qwen-RobotManip, Qwen-RobotWorld, three foundation models, a full stack for embodied intelligence. 🧭 Qwen-RobotNav — the gateway to mobility. • Unifies 5 navigation tasks in one model: instruction following, point-goal, object-goal, target tracking, autonomous driving • Controllable observation protocol • Tool interface for agentic systems 🤖 Qwen-RobotManip — the foundation of interaction. • Unified state-action space across heterogeneous robots • Camera-frame delta poses for coherent cross-embodiment training • Pretrained on a 38,100+ hour open-source corpus 🌍 Qwen-RobotWorld — infinite worlds for physical agents. • Single world model, 20+ embodiments • Natural-language action interface • Predicts physically grounded futures across manipulation, driving, and navigation Each model is independently useful, and could be composed as physical-world tools.Together, they form the low-level toolkit for general-purpose agentic systems that don't just see the world, but act in it. 📷 Blog: https://qwen.ai/blog?id=qwen-robotsuite 📖 Report： Qwen-RobotNav: https://qianwen-res.oss-accelerate.aliyuncs.com/qwenrobot/papers/Qwen_RobotNav.pdf Qwen-RobotManip: https://qianwen-res.oss-accelerate.aliyuncs.com/qwenrobot/papers/Qwen_RobotManip.pdf Qwen-RobotWorld： https://qianwen-res.oss-accelerate.aliyuncs.com/qwenrobot/papers/Qwen_RobotWorld.pdf

译通义千问推出Qwen-Robot Suite，包含三个基础模型：Qwen-RobotNav统一5种导航任务（指令跟随、点目标、物体目标、目标追踪、自动驾驶），具备可控观测协议和智能体工具接口；Qwen-RobotManip实现异构机器人统一状态-动作空间，基于38,100+小时开源语料预训练；Qwen-RobotWorld是单一世界模型，支持20+具身形态，通过自然语言动作接口预测物理世界未来（涵盖操作、驾驶、导航）。三个模型可独立使用或组合，构成通用智能体的底层工具包。

Berryxia.AI@berryxia · 6月16日67

NVIDIA直接开源了一个骨架能适应所有体型的3D人体模型，把机器人和物理AI的建模门槛又往下砸了一层。 SOMA-X v0.2用单一骨架就能表达各种体型，关节扭转自动修正让变形自然，骨骼还能自动缩放，还带高级姿态反转和超轻量数据。 Apache 2.0开源，机器人训练、物理仿真、动作迁移都直接能用。 以前做具身智能最麻烦的就是不同机器人体型不统一，动作数据很难复用。 现在一个通用骨架就把这个问题解决了，同一个动作数据可以更轻松地迁移到不同硬件上，训练效率直接起飞。 最关键的是这不是玩具级别的东西，而是专门为机器人和物理AI设计的。 自然变形、自动缩放、轻量数据，这些细节堆在一起，让数据飞轮转得更快。 以前大家觉得物理世界AI慢是因为硬件和仿真跟不上，现在连模型层面的基础积木都开始被开源社区快速迭代了。 项目地址见评论区👇🏻

译NVIDIA 推出开源 3D 人体模型 SOMA-X v0.2，采用单一骨架即可表达各种体型，专为机器人和物理 AI 设计。模型支持关节扭转自动修正、骨骼自动缩放、高级姿态反转及超轻量数据，以 Apache 2.0 协议开源。该模型解决了不同硬件之间体型不统一导致动作数据难以复用的问题，通过通用骨架实现同一动作数据跨硬件迁移，显著提升机器人训练、物理仿真与动作迁移的效率。

AYi@AYi_AInotes · 6月15日41

Damn，在AI的加持下， 农业打药居然进化到只打杂草了， 36 台相机每秒扫 2100 平方英尺， 药剂直接省三分之二。 现在约翰迪尔的 See & Spray， 只对准杂草精准下药, 36 台相机沿喷杆排布， 每秒扫描 2100 平方英尺， AI 用 200 毫秒区分作物与杂草， 对应喷头单独触发， 除草效果和传统方式持平， 非残留除草剂用量直接砍掉三分之二John Deere， 很多人以为农业自动化就是没人开车， 但其实真正有价值的升级， 是把过去靠经验的粗放作业， 变成可感知、可识别、可精准控制的系统， 省下的不只是药剂成本， 还有土壤和食品里的化学残留， 这才是真正的科技让生活更美好， 让地球更美好啊🥹

译John Deere的See & Spray系统搭载36台相机，每秒扫描2100平方英尺，AI用200毫秒区分作物与杂草，对应喷头单独触发，非残留除草剂用量减少三分之二，效果与传统持平。此外，TRIC Robotics的Luna机器人在夜间用紫外线破坏害虫DNA，减少30-70%化学农药，已在加州草莓田部署上千英亩，按服务付费，成本接近传统喷药。农业正从粗放经验转向精准可控系统，降低化学残留。

Rohan Paul@rohanpaul_ai · 6月14日37

In China, Jiangxi Province, smart parking-assistant robots have been deployed!

译在中国江西省，智能停车辅助机器人已经部署！

Rohan Paul@rohanpaul_ai · 6月14日23

Robots everywhere, but not like movies. Small, specialized machines quietly slip into streets and daily ops, doing 1 narrow job and improving through software updates.

译机器人无处不在，但不像电影中那样。 小型、专门的机器悄悄潜入街道和日常运营，做一项狭窄的工作，并通过软件更新不断改进。

Rohan Paul@rohanpaul_ai · 6月13日52

Sony AI’s Ace robot defeats pro Miyuu Kihara under official ITTF rules Nature paper - "Outplaying elite table tennis players with an autonomous robot"

译Sony AI 的 Ace 机器人在官方 ITTF 规则下击败了专业选手 Miyuu Kihara Nature 论文——“用自主机器人超越精英乒乓球选手”

Rohan Paul@rohanpaul_ai · 6月13日68

Nvidia's Cosmos 3: 1 model that can understand, simulate, and act across many physical AI tasks. It treats action as a first-class language of the world. Most AI models look at reality from the outside: images become captions, videos become descriptions, and motion becomes something to label after the fact. Cosmos 3 tries to collapse that distance by putting language, image, video, audio, and action into one shared system, so a robot can connect what it sees with what might happen next and what it should do. A home robot cannot simply recognize a plate, a table, and a human instruction, because the useful question is what changes when it moves, grasps, slips, bumps, or waits. That is why the paper’s action-token design matters: it turns movement into something the model can condition on, infer from video, or generate alongside a future scene. ---- Link – arxiv. org/abs/2606.02800 Title: "Cosmos 3: Omnimodal World Models for Physical AI"

译Nvidia发布Cosmos 3——一种全模态世界模型，将语言、图像、视频、音频和动作整合到同一系统，使物理AI能跨越“理解、模拟、行动”三大任务。它把动作视为世界的第一类语言，通过动作token设计，让模型可基于视频推断动作，或同时生成未来场景及对应运动。这使机器人从“识别物体”升级为预测“移动、抓取、滑动”等交互后果。相关论文《Cosmos 3: Omnimodal World Models for Physical AI》已发布于arXiv。

Berryxia.AI@berryxia · 6月13日60

AI有些地方真的还是“啥也不是的层面”！ 空白和进步空间巨大！ AI现在连抓个杯子都抓不对，手还没真碰到，杯子自己就飞起来了。 极客公园这期对谈里，Aether AI创始人黄碧薇教授举了这个例子：今天的视频生成模型学的是“手靠近杯子，杯子常常会动”这种相关性，而不是“为什么动、我这一抓到底会发生什么”这种因果。 聊天里说错话改改就行，可一旦进入物理世界——机器人、自动驾驶、生物医药。 一个变量算错，后果就是真的。幻觉在这里可没那么好玩。 所以下一代AI的分野，不是把世界预测得更像，是真正理解世界为什么这样运行。 这就是因果世界模型想干的事：让AI不只看表象，更看懂机制。 黄教授团队的benchmark显示，因果结构能让机器人成功率提升25-50%，样本需求降5-10倍。 同一堆数据，换个结构，经济性直接变了。 以前大家觉得规模化利用相关性就能一路走到黑，现在物理世界把这套玩法直接打脸了。 真正的智能，得从“知道是什么”进化到“知道为什么”。

译当前视频生成模型仅学到“手靠近→杯子动”的相关性，而非因果机制，导致抓杯子时杯子提前飞起。Aether AI 创始人黄碧薇教授提出因果世界模型（Causal World Model），旨在让 AI 理解物理运行机制而非仅预测表象。其 benchmark 显示，引入因果结构可使机器人成功率提升 25-50%，样本需求降低 5-10 倍。这标志着下一代 AI 需从“知道是什么”进化到“知道为什么”，尤其在机器人、自动驾驶等真实物理场景中。

Emad@EMostaque · 6月12日55

Unitree whose robots you have all seen only shipped 5,500 last year. The robotics market is literally going to be thousands of times bigger and likely end up bigger than the entire human labor market. Especially when they spread to space.

译人形机器人初创EngineAI秘密提交港股IPO，成为继宇树、优必选后第三家冲刺公开市场的中国公司。该公司2023年成立，最新估值超100亿元，苹果代工厂立讯精密为股东。2025年出货约400台，目标2026年4000-5000台。对比之下，宇树2025年出货5500台，智元5200台，优必选全尺寸1079台。EngineAI PM01定价2.55万美元起。创始人赵同阳原为小鹏机器人团队负责人。主推文观点认为机器人市场将增长数千倍，最终超过整个人类劳动力市场。

Google DeepMind@GoogleDeepMind · 6月12日67

Our Robotics Accelerator has launched with 15 startups helping shape the future of physical AI in Europe. 🤖 This three-month program will connect them with access to our AI stack, Gemini Robotics models and hands-on support from our teams. Meet the companies → https://goo.gle/4oeEk2K

译我们的机器人加速器已启动，15家初创公司助力塑造欧洲的物理AI未来。🤖 这个为期三个月的项目将为他们提供我们的AI堆栈、Gemini Robotics模型以及我们团队的实际支持。 了解这些公司 → https://goo.gle/4oeEk2K

AK@_akhaliq · 6月12日62

CHORUS Decentralized Multi-Embodiment Collaboration with One VLA Policy

译CHORUS 去中心化多本体协作，基于单一VLA策略。

Rohan Paul@rohanpaul_ai · 6月12日83

Jeff Bezos on CNBC explains revealed what Prometheus is building. Today his new company Prometheus announced a $12B funding round at a valuation of $41B . Prometheus trying to build an artificial general engineer that can help design and manufacture physical products like engines, medical devices, and electronics. So the target areas are hard physical products like jet engines, chips, bridges, medical devices, consumer electronics, aerospace systems, vehicles, and drug design, where design cycles can take years because every idea has to survive physics, materials, cost, testing, and factory limits. Bezos’ jet-engine example explains it well: asking for the same engine with 10% more thrust can become a 10-year engineering program, and Prometheus wants to shrink that “dream-build” cycle by 10x or more. The $6.2B launch funding gave Prometheus a massive starting base, and the new raise says the company likely needs far more compute, talent, and industrial data before it can prove the product. Their $41B valuation shows that frontier AI is becoming less a software race than a compute procurement race. A company with no broadly shipped product can raise $12 billion at a $41 billion valuation because investors are not only funding a model, they are prepaying for the machines that might make the model possible. The scarce asset is no longer just talent or algorithms, but clustered GPUs, power contracts, cooling, networking, and the operational skill to keep expensive silicon busy. They are proof that demand is arriving faster than infrastructure can be built, and that every frontier funding round quietly turns into a future claim on power, racks, GPUs, and uptime.

译Jeff Bezos 在 CNBC 披露其新公司 Prometheus 的愿景：构建人工通用工程师，设计制造喷气发动机、芯片、医疗设备等硬物理产品，将传统数年设计周期缩短 10 倍以上。公司宣布完成 120 亿美元融资，估值 410 亿美元。初始启动资金 62 亿美元，新一轮融资表明公司需要更多算力、人才和工业数据才能验证产品。410 亿美元估值表明，前沿 AI 已从软件竞赛变为计算采购竞赛——投资者实质在为可能实现模型所需的机器预付费。

Rohan Paul@rohanpaul_ai · 6月12日29

Guangzhou Auto Show in China hyper-realistic robotic suit that many initially mistook for XPeng’s IRON humanoid robot.🙂

译中国广州车展 一个超逼真的机器人服装，许多人最初误以为是小鹏的IRON人形机器人。🙂

Chubby♨️@kimmonismus · 6月10日51

TARS' DexHand Highlights a New Era of Hand-Brain Integration At ICRA 2026 in Vienna, TARS Robotics is showing their 21-DoF DexHand with tactile sensors detecting textures down to 0.05mm, built as a true 1:1 copy of human bone and joint structure, precise enough that it reproduces 26 standard sign-language gestures and real-time mirror-control interaction with attendees, paired with their AWE 3.0 model trained on Human-centric data for full "hand-brain integration." Live demo: their A1 robot autonomously packs a backpack and performs sub-millimeter wire harness assembly with real-time error correction. The underlying research, TacForeSight - a newly released technology developed by TARS in collaboration with global universities and institutions, allows the robot to predict contact events before they fully occur, which is critical for high-frequency, contact-rich manipulation tasks.

译TARS Robotics 在 ICRA 2026（维也纳）展出 21-DoF DexHand，触觉传感器可检测 0.05mm 纹理，精确复制人体骨骼关节结构，能重现 26 种标准手语手势并实现实时镜像控制交互。配合 AWE 3.0 模型实现“手脑整合”。现场演示 A1 机器人自主打包背包及亚毫米线束装配（实时纠错）。新发布 TacForeSight 技术可预测接触事件发生前的状态，适用于高频接触操作任务。

Huawei Cloud@HuaweiCloud1 · 6月10日69

Unveiling the world’s first end-to-end embodied AI development platform! Huawei Cloud CloudRobo streamlines the entire embodied AI development lifecycle—from data and models to deployment and integration—backed by a secure, trusted PB-scale data foundation. At #INSPIRE2026, the National and Local Co-built Humanoid Robotics Innovation Center, Yijiahe Technology, and Shanghai Jiao Tong University highlighted CloudRobo’s core strengths: • Dual evaluation system for data & models • Fast assembly of active force control models • Robots connected to cloud in hours, models deployed in minutes Ready to shape the embodied AI future? Join the Embodied AI Zone today. 👉 Learn more: https://tinyurl.com/3zda3sr3 #HuaweiCloud #CloudRobo #EmbodiedAI #Robotics

译华为云推出全球首个端到端具身AI开发平台CloudRobo，覆盖从数据、模型到部署、集成的全生命周期，基于PB级可信数据底座。在INSPIRE2026上，国家地方共建人形机器人创新中心、Yijiahe Technology、上海交通大学展示了其核心能力：数据与模型双评估系统、主动力控模型快速组装、机器人小时级上云、模型分钟级部署。

Rohan Paul@rohanpaul_ai · 6月10日49

A Unitree Robotics humanoid performed a solo sequence of traditional Chinese martial arts outdoors fluid kicks, wide stances, and precise arm movements with strong balance and coordination.

译宇树机器人人形机器人在户外单独表演了一套中国传统武术套路 流畅的踢腿、宽大的步法以及精准的手臂动作，展现出强大的平衡与协调能力。

Rohan Paul@rohanpaul_ai · 6月10日47

Those eyes moved so naturally AheadForm’s Bionic face, pushing against the uncanny valley. Imagine this face on XPeng’s fashion-model-like smooth walker. That would go so hard.

译那些眼睛动得如此自然 AheadForm的仿生面部，正在突破恐怖谷。 想象这张脸出现在小鹏那模特般流畅的步行机器人上。那将非常震撼。

Rohan Paul@rohanpaul_ai · 6月9日68

Robotics is slow because every change needs physical setup, people, space, and repeated field runs. Physical AI needs the kind of testing system software teams already relies on. Antioch just introduced Antioch Agent, a browser-based robotics simulator. Antioch runs your existing robot software inside simulation, connects it to virtual sensors and actuators, and lets you test robot behavior without spending every test cycle on physical hardware.

译Rohan Paul 介绍 Antioch 推出的 Antioch Agent，一款基于浏览器的机器人仿真器。它允许现有机器人软件在仿真环境中运行，连接虚拟传感器与执行器，无需物理硬件即可反复测试。Antioch 宣称首次实现完整物理 AI 堆栈的浏览器端闭环智能体模拟，将数周实地测试缩短至数小时，加速实体自主系统的开发进程。

Chubby♨️@kimmonismus · 6月9日49

It was only a matter of time. UBTECH has released its first bionic humanoid robots. They don’t just look human - they feel human, too.

译这只是时间问题。 UBTECH 发布了其首批仿生人形机器人。 它们不仅外表像人——触感也像人。

Elon Musk@elonmusk · 6月9日42

SpaceX AI Satellites

译SpaceX AI 卫星

Rohan Paul@rohanpaul_ai · 6月9日50

A new US bill could ban some Chinese robots from America. The GUARD Act would force security agencies to review robots from China and other adversary countries, then place risky systems on the FCC’s Covered List, the same kind of restriction used against companies such as Huawei and ZTE. The fear is not only that a robot has cameras, microphones, sensors, maps, motors, and wireless links, but that the whole machine becomes a moving computer inside factories, labs, homes, and police departments. A separate Schumer-Cotton bill would stop federal agencies from buying or using Chinese humanoid robots, with exceptions for controlled military or law-enforcement research.

译美国新提出的GUARD法案要求安全机构审查来自中国及其他敌对国家生产的机器人，并将高风险系统列入FCC“覆盖清单”，类比此前对华为、中兴的禁令。立法者担忧机器人不仅配备摄像头、麦克风、传感器、地图、马达和无线链路，更会成为可在工厂、实验室、家庭和警局内移动的计算机。另一项Schumer-Cotton法案单独禁止联邦机构购买或使用中国人形机器人，但允许受控的军事或执法研究例外。

Rohan Paul@rohanpaul_ai · 6月9日49

AheadForm Elf V1 humanoid robot is slowly revealing its body design: its bionic skin uses magnetic attachment, making it easy to take off. This exposes the servo engines that power its facial expressions with 30 artificial facial muscles.

译AheadForm Elf V1 人形机器人正在逐步展示

SemiAnalysis@SemiAnalysis_ · 6月9日63

China's Unitree Will Dominate Global Robotics The Fastest Iteration Cycle In Next-Gen Robotics Should See Unprecedented Acceleration https://newsletter.semianalysis.com/p/chinas-unitree-will-dominate-global

译中国宇树将主导全球机器人 下一代机器人最快的迭代周期将迎来前所未有的加速

X.PIN@thexpin · 6月8日56

The price of humanoids in China just fell off a cliff! Unitree is facing a massive margin squeeze. Their avg. selling price keeps dropping: 🔹2023: $81.6k 🔹2024: $35.8k 🔹2025 (Q1-Q3): $23k Meanwhile, their unit cost only fell from $10k to $8.5k. Bumi, a fierce competitor in the humanoid market, just launched their cosumer-level robot at $1.4k after subsidies. The robot stands at 3′1″, aimed at education and companionship. Nowadays, core components like servo motors and reducers are mainly domestically manufactured. Robots are borrowing heavily from EV and smartphone supply chains, with batteries and vision sensors all repurposed from mature production lines.

译中国人形机器人价格断崖式下跌。宇树科技（Unitree）均价从2023年$81.6k降至2024年$35.8k，2025年前三季度进一步跌至$23k，而单位成本仅从$10k降至$8.5k，利润空间被急剧压缩。竞争对手Bumi推出消费级人形机器人，补贴后售价仅$1.4k，身高3英尺1英寸，面向教育和陪伴场景。核心部件如伺服电机、减速器已实现国产化，机器人供应链大量借鉴电动汽车和智能手机成熟产线，电池、视觉传感器均从现有产线复用。

AYi@AYi_AInotes · 6月8日53

好家伙，赛博女友这么快就来了？！ 不对，是赛博伴侣，因为是男女两款sku，男款身高183，女款身高168， 全系标配88个运动关节，据说动作比真人还自然，支持WiFi联网，但续航时间不是很长，只有2-4个小时， 魔幻的是优必选连发布会都没开，直接上线了这款全球首个超仿生机器人，兄弟们注意，是全仿生！ 搭载养成系情感大模型，本地加密存储记忆，支持多维度外观定制，而且还开启了IP合作，damn！ 尽管要交¥3000定金，目前已经有2000多人直接盲订了这款赛博伴侣机器人🤖

译优必选直接上线全球首个超仿生机器人“赛博伴侣”，分男女两款：男款身高183cm，女款168cm，全系标配88个运动关节，动作逼真。支持WiFi联网，续航仅2-4小时。搭载养成系情感大模型，本地加密存储记忆，支持多维度外观定制并开启IP合作。虽未召开发布会，已开启盲订，需支付3000元定金，目前超2000人下单。

Berryxia.AI@berryxia · 6月8日65

波士顿机器人直接要替代搬运工了吗？ 以后抗楼战神的活儿要被抢走了😄

译波士顿动力展示了其新Atlas人形机器人，它举起并搬运了一个50磅的冰箱。 机器人利用全身平衡并移动重物，动作自然且类人。 在现实世界中执行之前，Atlas在虚拟仿真中练习了数百万小时。 网友调侃：“以后抗楼战神的活儿要被抢走了😄”

Rohan Paul@rohanpaul_ai · 6月8日42

Robot's locomotion and recovery under unexpected force in real time. The recovery phase was something 😀

译机器人在实时突发外力下的移动与恢复。 恢复阶段有点意思 😀

Emad@EMostaque · 6月7日27

Only one solution

译据《纽约邮报》报道，一台机器人足球运动员踢球力量大到将墙壁击穿，可能会使人类变得过时。 Emad Mostaque 评论道：“唯一的解决方案。”

Rohan Paul@rohanpaul_ai · 6月7日36

Nice tease from @UBTECHRobotics . the faces of its “emotional” humanoid robot couple before their June 30 debut. This is their U1 series, a ready-to-use consumer-oriented product. Marketed primarily for emotional companionship in home settings.

译来自 @UBTECHRobotics 的有趣预告。其“情感”人形机器人情侣在 6 月 30 日首秀前展示面部。 这是他们的 U1 系列，一款即用型面向消费者的产品。主要面向家庭场景的情感陪伴。

Elon Musk@elonmusk · 6月7日24

Join Tesla if you want to solve real-world AI

译真实世界的AI在于解决边缘案例，而非平均案例。马斯克号召：想解决真实世界AI，加入特斯拉。

Rohan Paul@rohanpaul_ai · 6月7日36

Masayoshi Son says AI could be 50x bigger than dot-com and the next trillion-dollar company will come from robotics. "So even dot-com there was a bubble burst, but then right after that, the peak of the dot-com bubble, the year 2000 was not really a peak. It was like a small hill. It went down, but then it went much, much bigger with a big trough bubble and a free cash flow. So AI will be, this is the beginning, and the future of the profit and the growth opportunities is tremendous." ~ Masayoshi Son, the founder and CEO of SoftBank --- From 'CNBC International Live' YT channel (link in comment)

译软银创始人兼CEO孙正义表示，AI的市场规模可能比互联网泡沫大50倍，下一个万亿美元公司将来自机器人领域。他指出，互联网泡沫在2000年破裂后实际增长远超当时顶峰，而AI目前才刚刚起步，未来的利润和增长机会极为巨大。

Rohan Paul@rohanpaul_ai · 6月6日34

This is useful stubbornness. Recovery is a first-class robotics skill. The floor is the eval.

译这是有用的固执。 恢复是头等机器人技能。 地板就是评估。