机器人技术中的人工智能突破不会很快改变你的生活

内容总结:
人工智能机器人领域的突破短期内难以改变生活
人工智能的进步让人们得以一窥未来图景:机器人能像人类一样在世界上自如行动。但问题在于,推动AI近期取得进展的同类技术是否足以实现这一目标,还是需要开辟一条全新的路径。
本文由《麻省理工科技评论》与Aventine合作完成。Aventine是一家非营利研究基金会,致力于创作和支持关于科技与科学如何改变我们生活的内容。
一个外形酷似人类的机器人——白色机身、黑色头部和躯干——频频出现在视频中。你或许见过它跳舞、递爆米花、把垃圾扔进垃圾桶、吸尘,或者按微波炉的按钮。又或者,你看到过它在递水瓶时向后摔倒,或是艰难地熨衬衫。
这就是特斯拉的Optimus——一款由AI驱动的人形机器人。该公司CEO埃隆·马斯克认为,它“不仅是特斯拉有史以来最大的产品,而且很可能是有史以来最大的产品”,将率先进入工厂车间,随后进入我们的家庭。他在7月对股东表示,Optimus最终“将拥有人类的灵巧性,进而超越人类”。马斯克主张,Optimus机器人几乎可以自动化所有人类劳动——从搬运金属板材到叠衣服——每台成本低至2万美元。今年1月在瑞士达沃斯举行的世界经济论坛年会上,他预测这些机器人可能在2027年底前面向公众销售。
马斯克并非唯一对此充满热情的人。硅谷风投公司安德森·霍洛维茨的联合创始人兼普通合伙人马克·安德森曾表示,机器人技术可能成为“地球历史上最大的产业”。今年1月,英伟达CEO黄仁勋表示,人形机器人将在今年达到人类水平的能力。据摩根士丹利预测,到2050年,“外形和行为类似人类”的机器人数量可能达到近10亿台,创造超过5万亿美元的市场。
这些宣言在很大程度上基于这样一种想法:推动OpenAI的ChatGPT和Anthropic的Claude等工具取得突破的同类AI进展,将使新一代机器人能够模仿人类动作,就像聊天机器人模仿人类语言一样。但许多机器人研究人员对此持怀疑态度,认为这种假设低估了将建立在语言和图像之上的智能用于驾驭物理世界无限多变性的难度。“那些(建造人形机器人的)公司——没有一家——知道如何让这些机器人聪明到有用的程度,”常被称为AI教父之一的杨立昆在1月达沃斯会议期间的另一场活动上表示。
研究人员还指出,将外形酷似人类的人形机器人与能够学习和执行多种任务的所谓通用机器混为一谈,是一种误导。“造一个看起来像人的机器人非常容易,”Agility Robotics联合创始人兼首席机器人官、俄勒冈州立大学机器人学教授乔纳森·赫斯特解释说,“但要造一台在运动、行为或物理特性上像人的机器,难度要大得多。”
这些分歧——通用人形机器人究竟是近在眼前还是遥遥无期,以及当前形式的AI是否足以完善它们——正在全美各地的机器人实验室中上演。在这些实验室里,对时间表的炒作掩盖了艰苦但有意义的进展。
大约十年前,一系列突破引发了生成式AI革命,将长期设想的人工智能可能性变为现实。机器人专家们——尽管对具体时间存在分歧——相信机器人领域也可能发生同样具有变革性的革命,赋予机器长期以来难以企及的物理直觉和流畅性。随着机器人技术缓慢推进,问题在于:推动AI进步的同类方法和工具是否足以实现这一目标,还是需要一条全新的路径。
机器人遇上先进AI
要了解当今最聪明的机器人大脑之一,值得看看谷歌DeepMind能用一台名为ALOHA 2的设备做什么。ALOHA是“双臂遥操作低成本开源硬件系统”的缩写。
机器人专家长期以来一直在争论通用机器人是否需要类人形态。支持者认为这有助于机器人融入现有世界,反对者则认为不值得费这个劲。ALOHA 2体现了后一种思路。它看起来并不起眼,只是一对机械臂、一些夹爪和几个摄像头。尽管看似简单,它却是谷歌DeepMind研究人员的主力工具,他们在各实验室用它来测试其最先进的机器人AI系统Gemini Robotics。
在Gemini Robotics的控制下,ALOHA 2更像一个通用机器人,能够根据训练过的示例执行各种任务。让它装一个午餐盒,从演示视频可以看到,它能用两个钳形夹爪轻轻地把一片白面包放入Ziploc袋中,封好袋口,把一串葡萄放入特百惠容器,盖好盖子,然后小心地把这些东西放进午餐盒并拉上拉链。
这顿午餐不算丰盛。但机器人能把它组装起来,代表了相比三年前可能做到的,已经迈出了实质性的一步。
这在很大程度上归功于AI及其对所谓“机器人策略”的影响。机器人策略控制着通用机器人如何评估和理解周围环境、规划如何在其中移动,然后正确执行任务。
过去,这些策略基于工程师编写的规则,硬编码到机器人软件中——数千行代码决定机器人在数百个任务中每一毫米的动作。过去几年发生的事情——也是人们对通用机器人持乐观态度的主要原因——是机器人策略正被移交给先进的AI系统,而非硬编码到机器人软件中。这首先发生在视觉语言模型(VLM)上。VLM类似于大语言模型,但除了文字外还用图像进行训练。给VLM看一张咖啡洒了的图片,让它找工具清理,它能识别附近的抹布。这种即时的情境理解在几年前机器人策略还是硬编码时并不存在。
接下来是视觉-语言-动作模型(VLA),使机器人能够评估环境并在其中采取行动。这类模型通过添加另一个组件来实现:运动指令。VLA在一系列与执行特定任务相关的图像或视频以及机器人手臂如何运动的相关数据上进行训练。这些运动数据通常通过遥操作收集,即人类使用遥控器引导机器人完成动作。这种训练让AI学会在给定任务中指挥机器人如何移动和操作。把一个由VLA驱动的机器人放在桌前,告诉它“合上笔记本电脑”或“把耳机线缠好”,它会环顾场景、识别相关物体、规划执行请求的方式,然后挥动机械臂行动——至少如果它之前见过这个任务的话。
Gemini Robotics模型就是一个VLA,在大量人类演示中训练,涵盖各种不同动作。因此,它能执行相对复杂的任务,如用厨房夹子夹起荷兰豆、做折纸、或组装一顿简单的午餐。这令人印象深刻,但有一个明显的局限:目前,如果让VLA控制的机器人执行训练集之外的任务,极有可能失败。
“考虑到所有任务的空间,真正的通用策略应该能完成整个范围内的所有事情,”伦敦帝国理工学院机器人学教授爱德华·约翰斯说。但如今,Gemini Robotics模型只能做“这里几件事,那里几件事”。
寻找真正的通用性
那么,如何让机器人能做更多事情?通常的答案可能并不令人意外:用更多数据训练它们。
按照这种思路,更多数据等于更多示例,更多示例等于更强通用性。例如,谷歌DeepMind希望汇集“尽可能多的数据”,该公司前机器人技术负责人、目前正在从事自己的AI机器人项目的潘纳格·桑凯蒂说。但从哪里获取?大语言模型受益于海量现有文本用于训练。机器人领域没有相应的优质物理演示数据池。研究人员有几种方法来弥补,但都有缺陷。一种是雇佣大量人员创建和收集遥操作数据(成本高且耗时)。另一种是在人们执行活动的视频上训练VLA(由此产生的数据质量差)。还有一种是在现实世界中部署机器人,利用从这些经验中收集的数据进一步优化AI模型(机器人在实验室外不够安全或可靠)。桑凯蒂认为,利用从所有这些来源收集数据的“多管齐下”方法是最有可能的前进路径。
但认为仅靠训练数据就能解决问题的信念远非普遍共识。Agility的赫斯特将其描述为“一个根本上有缺陷的前提”。
问题在于,现实世界中的任务复杂度会迅速爆炸。比如你想煮咖啡,有无数变量:没有两个厨房是一样的;咖啡机工作方式不同;不同的杯子需要不同的握法;咖啡粉、热水和牛奶都需要不同的处理方式。即使这个简单任务也需要理解不断变化的可能性菜单。赫斯特认为,通过VLA实现通用性需要“对机器人可能做的所有事情实现完整数据覆盖”。换句话说,需要一个几乎无限的训练数据池。
杨立昆对整个方法不屑一顾。“那些在语言上成功的(AI)方法不适用于高维度、连续、有噪声的数据”——而这类数据在机器人领域司空见惯,他在达沃斯表示。“你必须用别的方法。”
“别的方法”的主要竞争者是世界模型——一种更多基于视频、三维扫描和传感器数据组合而非文本训练的AI形式,旨在预测现实世界中行动的结果。目标是构建拥有足够精确的现实内部表征的模型,以捕捉物理世界实际运作方式——物体如何移动、碰撞、坠落和变形。如果机器人专家能在足够忠实于真实世界物理的模拟中训练机器,开发将变得更快、更便宜、更安全,减少对真实世界测试的需求。更具变革性的是,配备世界模型的机器人可以对周围环境进行推理,而非仅仅对其作出反应,帮助它们在采取行动前预判后果。
英伟达和谷歌等公司正在研究这项技术,投资者的资金正涌入高知名度初创公司。由斯坦福AI研究员李飞飞联合创立的World Labs在2月筹集了10亿美元资金,并于9月底被AMD以82亿美元收购。由杨立昆(前Meta首席AI科学家)联合创立的AMI Labs也在3月筹集了10亿美元。然而,据他们自己承认,现在仍处于早期阶段。去年年底,李飞飞称该领域“尚处于萌芽阶段”,并补充说“基础性方法仍在建立中”。在6月的一篇Substack文章中,她描述了严峻的挑战。目前,世界模型是一个有前景的研究领域,而非通往通用机器人的直接路径,但我们开始看到它们可能实现什么的蛛丝马迹。
其中一个缩影是去年4月在旧金山一个机器人实验室发生的一次虽小但可能意义重大的飞跃。
一次突破?
在旧金山Mission District的中心地带,初创公司Physical Intelligence——或者说PI(如π),它喜欢被这样称呼——正致力于开发一个通用大脑,理论上可以将任何机器人变成通用机器人。该公司采用“什么都往里扔”的方法训练机器人AI模型,最近观察到了配备世界模型的机器人大脑可能具备的能力。
2024年,PI发布了其首个通用机器人系统π0的详细信息,声称这是“迄今为止能力最强、最灵巧的通用机器人策略”。该模型最初在1万小时的专有遥操作人类演示数据集以及几个开源机器人数据集上训练。2025年春季发布的π0.5版本在更多样化的数据集上训练,包括来自网络的有标签图像,使其更加多才多艺。2025年秋季的π0.6更新增加了强化学习。每次更新都对模型性能产生了重要改进,将其能力菜单从缓慢叠衣服扩展到在新环境中收拾物品,再到以更高成功率完成折箱子等任务。然后,在2026年4月,π0.7似乎将PI弹射到了新领域。这个版本利用了一个能力较弱的世界模型,生成执行任务所需步骤的图像。当机器人执行任务时,这个“轻量级”模型向它提供下一步该做什么的快照。
该公司声称该模型展现出组合泛化能力的最初迹象。组合泛化是指AI系统通过重新组合训练数据中学到的技能来执行从未接触过的技能。一项测试要求模型“把红薯放进空气炸锅”——这是它此前从未遇到过的任务。在演示视频中,机器摸索了一会儿,几次起步失误,最终完成了一次还算合理的尝试,尽管没有完全完成任务。
加州大学伯克利分校教授、PI联合创始人谢尔盖·莱文对这一潜力感到兴奋:“这实际上是我们第一次令人信服地看到那种组合泛化,我们可以要求模型完成我们没有专门收集数据并训练它去做的任务,而它确实能做出还算过得去的尝试。”
这一成功让团队好奇模型是如何实现这一壮举的。经过深入调查,他们发现训练材料中潜藏着相关标注遥操作数据的片段,包括两个人类操作员用机器人把空气炸锅篮子推入炸锅的示例。这些数据碎片可能足以让π0.7几乎空气炸一个红薯。
目前尚不清楚π0.7的泛化能力究竟有多令人印象深刻。不过,考虑到模型接触空气炸锅的时间如此短暂,它让我们得以一窥前沿研究目前能将机器人带到多远。
“70%的成功率等于不管用”
你可能感觉到一种脱节:机器人在实验室中迈出的断断续续的婴儿步伐——“看!它把红薯放进空气炸锅了!”——与许多演示和视频中展示的令人眼花缭乱的、栩栩如生的灵巧性形成对比。在那些视频中,机器人从在舞台上跳舞到彬彬有礼地端饮料,无所不能。这类演示往往没有清楚揭示一个关键事实:在许多情况下,是人类在控制机器人,或者已经仔细编写了机器人的动作脚本。(例如,2025年3月与英伟达CEO黄仁勋一起登台的机器人,看似在回应他的指令并跟着他走,实际上是由其制造商所称的“幕后木偶师”远程控制的。)
目前,让机器人知道该去哪里的完全自主运动规划——尤其是在建筑工地或不熟悉的家等新的混乱环境中——在很大程度上仍未解决。更大的挑战——尽管往往与之相关——在于让机器人处理更大、更模糊的工作,这些工作包含多个任务,需要决定如何做以及按什么顺序做。这就是能把盘子放进微波炉的机器人和能成功响应“做晚饭”指令、翻冰箱、切食材、开炉灶的机器人之间的区别。谷歌DeepMind在这方面最好的尝试——要求其机器人查看厨房并把所有做蘑菇烩饭的食材装进篮子——迄今以失败告终。
更大的挑战在于,实用机器人基本上必须每次都做对。对于VLA,“当结果从50%成功率提升到70%时,人们非常兴奋,”波士顿动力创始人马克·雷伯特说。“但70%的成功率等于不管用,对吧?”
少数正在真实环境中测试的人形机器人,在严格控制的环境中执行着极为有限的任务。它们远非通用机器人。据Agility称,该公司已在GXO Logistics、亚马逊和舍弗勒的设施中部署了数百台机器人进行试验。但赫斯特表示,目前这些机器人针对的是搬运箱子等简单任务。他补充说,即便如此,也花了数年时间才开发出足够安全、让物流公司敢于考虑使用的机器人。就马斯克而言,他在2025年5月声称到年底将有“数千台”Optimus机器人在特斯拉工厂工作,但今年1月他表示公司只有“一些特斯拉Optimus机器人在工厂做简单任务”。
虽然人形机器人开始踏上工厂车间,但进入家庭将更加困难。现在,如果你愿意,可以预订1X Neo家用机器人,预计今年晚些时候交付。售价2万美元,它承诺承担“家中无聊琐碎的任务”——收拾碗碟、应门、整理客厅——“让你专注于对你重要的事”。这款身高五英尺六英寸的机器人的设想是有一天能自主完成所有这些任务,但目前大多数事情需要远程人工操作员。(是的,需要有人获得许可通过机器人的摄像头窥视你的家。)当被问及完全自主的机器人何时能准备好从事家务时,赫斯特说:“如果我必须选一个数字,我会说机器人……真正在人们家中做有用的事情还需要10年。”
当那一天到来时,机器人很可能是中国制造的,因为中国在生产方面远远领先于西方。据市场情报公司Omdia和中国机器人公司宇树科技的数据,2025年约1.5万台人形机器人出货量中,近90%由中国公司制造。宇树科技去年出货的人形机器人数量超过任何其他公司,其一款型号售价不到6000美元。这样的实惠价格可能大大有助于让机器人对消费者更具吸引力,尽管该公司预计其机器将首先用于工业应用。(如果你想知道谁在购买这些中国机器人,美联社最近报道称,订单主要来自企业和学术实验室以及国有企业。)
我们以前来过这里
建造人形机器人的梦想由来已久:早在1495年,达·芬奇就草拟了一个由缆绳和滑轮控制的机械骑士的设计。整个20世纪,科幻梦想中的机器来来去去。
西屋电气七英尺高的箱形腿上机器人Elektro在1939年纽约世界博览会上亮相,还会抽烟。日本早稻田大学1973年制造的WABOT-1是第一个全尺寸可编程人形机器人。本田2000年发布的ASIMO可能是第一台证明至少有一定能力的此类机器——它至少在一定程度上能爬楼梯、识别人脸、自主穿行空间。但该机器人在2018年停产,无法在演示之外取得足够进展以变得有用。
所有这些在当时都是令人印象深刻——甚至令人瞠目结舌——的工程壮举。但没有一个准备好驾驭现实世界。今天的机器人,即使拥有先进AI的变革力量,仍然面临同样的生存挑战。
深度报道
人工智能
别被骗了——大语言模型不会推理
AlphaGo对阵围棋冠军李世石十年后,今天的AI仍未触及那场胜利背后的机制。
AI的递归自我改进可能不会那么快到来
AI智能体似乎还不够有创造力,无法开展真正创新的开放式AI研究。
别被这个夏天的AI炒作骗了
关于AGI和新能力的大言不惭经不起推敲。
这些初创公司正在追逐大语言模型的下一个大事件
来认识一下在AI巨头脚跟后面紧追不舍的新秀们。
保持联系
获取《麻省理工科技评论》的最新动态
发现特别优惠、热门故事、即将举行的活动等。
中文翻译:
AI在机器人领域的突破不会很快改变你的生活
人工智能的进步让我们得以一窥这样一个未来:机器人能像人类一样在世界上行走活动。问题在于,推动AI近期取得进展的那些技术是否足以让我们抵达那一天,还是说需要一条全新的路径。
本文由《麻省理工科技评论》与Aventine合作完成。Aventine是一家非营利研究基金会,致力于创作和支持关于科技与科学如何改变我们生活方式的內容。
一个外形像人的机器人——白色机身,黑色头部和躯干——不断出现在视频中。也许你见过它跳舞、递爆米花、把垃圾扔进垃圾桶、吸尘,或者按下微波炉的按钮。又或者你看到过它在分发水瓶时向后摔倒,或者费力地熨衬衫。
这就是特斯拉的Optimus,一款由AI驱动的人形机器人。该公司CEO埃隆·马斯克认为它“不仅是特斯拉有史以来最大的产品,而且很可能是有史以来最大的产品”,将率先进入工厂车间,之后进入我们的家庭。他在7月对股东表示,最终它“将拥有人类的灵巧性,然后超越人类的灵巧性”。马斯克认为,Optimus机器人可以以每台低至2万美元的价格,将几乎所有人类劳动自动化——从搬运金属板材到叠衣服。在今年1月于瑞士达沃斯举行的世界经济论坛年会上发言时,他预测这些机器人可能在2027年底前面向公众发售。
马斯克并非唯一为此布道的人。硅谷风投公司Andreessen Horowitz的联合创始人兼普通合伙人马克·安德森曾表示,机器人技术可能成为“地球历史上最大的产业”。今年1月,英伟达CEO黄仁勋表示,人形机器人将在今年达到人类水平的能力。据摩根士丹利预测,到2050年,“外形和行为像人类”的机器人数量可能接近10亿,创造超过5万亿美元的市场。
这些宣言在很大程度上受到这样一种想法的推动:推动OpenAI的ChatGPT和Anthropic的Claude等工具取得突破的那些AI进展,将使新一代机器人能够模仿人类动作,就像聊天机器人模仿人类语言一样。但许多机器人研究者对此持怀疑态度,他们认为这类假设低估了用建立在语言和图像之上的智能来掌握物理世界无限可变性所面临的挑战。“那些(建造人形机器人的)公司——绝对没有一家——知道如何让这些机器人聪明到有用的程度,”常被称为AI教父之一的杨立昆在1月达沃斯会议期间的另一场活动上说道。
研究人员还指出,将外形模仿人类的人形机器人与能够学习和执行多种任务的所谓通用机器混为一谈,是一种误导。“造一个看起来像人的机器人非常容易,”Agility Robotics联合创始人兼首席机器人官、俄勒冈州立大学机器人学教授乔纳森·赫斯特解释道,“造一台能像人一样动态或物理地运动或行为的机器则要困难得多。”
这些分歧——通用人形机器人究竟是近在眼前还是遥遥无期,以及当前形式的AI是否足以完善它们——正在全美各地的机器人实验室中上演。在这些实验室里,对时间表的炒作正在掩盖那些艰苦但有意义的进展。
大约十年前,一系列突破引发了生成式AI革命,将长期以来只能想象的人工智能变为现实。机器人学家——尽管他们对具体何时实现存在分歧——相信机器人领域也可能发生同样具有变革性的革命,这场革命将赋予机器长期以来无法企及的物理直觉和流畅性。随着机器人技术缓慢推进,问题在于:推动AI进步的那些方法和工具是否足以让我们抵达目标,还是需要一条全新的路径。
机器人遇上先进AI
要了解当今最聪明的机器人大脑之一如何工作,值得看看谷歌DeepMind能用一台名为ALOHA 2的设备做什么。ALOHA是“用于双臂遥操作的低成本开源硬件系统”的缩写。
机器人学家长期以来一直在争论,通用机器人是否需要人形外形。支持者认为这有助于它们融入现存世界,反对者则认为不值得为此费心。ALOHA 2反映了后一种思路。它看起来并不起眼,只是一对机械臂、一些夹爪和几个摄像头。但尽管看似简单,它却是谷歌DeepMind研究人员的主力工具,他们在各个实验室用它来测试其最先进的机器人AI系统Gemini Robotics。
当由Gemini Robotics控制时,ALOHA 2更像一个通用机器人,因为它能根据训练过的示例执行任意数量的任务。让它打包一个午餐盒,从一段演示视频可以看到,它能用两个钳形夹爪轻轻地把一片白面包放入Ziploc保鲜袋,封好袋口,把一串葡萄放入特百惠容器,盖好盖子,然后小心地把这些东西移入午餐盒并拉上拉链。
这顿午餐算不上好。但机器人能把它组装起来,这一事实本身就代表着相比哪怕三年前的可能性而言的客观进步。
这在很大程度上归功于AI及其对所谓“机器人策略”的影响。机器人策略控制着通用机器人如何评估和理解周围环境、规划如何在其中移动,然后正确执行任务。
过去,这些策略基于工程师编写的规则,他们将其硬编码到机器人软件中——数千行代码决定了机器人在数百个任务中每一毫米的动作。过去几年发生的变化——也是人们对通用机器人持乐观态度的主要原因——是机器人策略正被移交给先进的AI系统,而非硬编码到机器人软件中。这首先发生在VLM(视觉语言模型)上。它们类似于大语言模型,但训练数据既包括文字也包括图像。给VLM看一张咖啡洒了的图片,让它找工具清理,它能识别出附近的抹布。这种即时的情境理解在几年前机器人策略还是硬编码时并不存在。
接下来是视觉-语言-动作模型(VLA),它使机器人能够评估环境并在其中采取行动。这些模型通过增加另一个组件来实现这一点:运动指令。VLA在一系列与执行特定任务相关的图像或视频以及机器人手臂如何移动来执行该任务的相关数据上进行训练。这些运动数据通常通过遥操作收集,即人类使用遥控器引导机器人完成一个动作。这种训练使AI能够学习如何在给定任务中指挥机器人移动和操作。把一个由VLA驱动的机器人放在桌子前,告诉它“合上笔记本电脑”或“把耳机线缠好”,它会扫视场景、识别相关物体、规划执行请求的方式,然后挥动机械臂开始行动——至少如果它之前见过这个任务的话。
Gemini Robotics模型是一个VLA,在大量人类演示中进行了训练,涵盖了各种各样的动作。因此,它能执行相对复杂的任务,比如用厨房夹子夹起荷兰豆、做折纸,或者组装一顿简单的午餐。这令人印象深刻,但有一个明显的局限:目前,如果要求一个由VLA控制的机器人执行训练集之外的任务,它极有可能失败。
“考虑到所有任务的空间,真正的通用策略应该能够完成整个范围内的所有事情,”伦敦帝国理工学院机器人学教授爱德华·约翰斯说。但如今,Gemini Robotics模型只能做“这里几件事,那里几件事”。
寻找真正的通用性
那么,我们如何让机器人能做更多事情?通常的答案可能并不令人意外:用更多数据训练它们。
思路是,更多数据等于更多示例,更多示例等于更多通用性。例如,谷歌DeepMind希望汇集“尽可能多的数据”,该公司前机器人技术负责人、目前正在从事自己的AI机器人项目的潘纳格·桑凯蒂说。但从哪里获取?大语言模型受益于海量的现有文本用于训练。机器人训练并不存在相应的优质物理演示数据池。研究人员有几种方法来弥补,但都有缺陷。一种是雇用大量人员创建和收集遥操作数据(成本高昂且耗时)。另一种是在人们执行活动的视频上训练VLA( resulting数据质量差)。还有一种是在现实世界中部署机器人,并利用从这些经验中收集的数据进一步完善AI模型(机器人在实验室外不安全也不可靠)。桑凯蒂认为,利用从所有这些来源收集的数据的“多管齐下”方法是最有可能的前进路径。
但认为仅靠训练数据就能解决问题的信念远非普遍。Agility的赫斯特将其描述为“一个根本上有缺陷的前提”。
问题在于,现实世界中的任务会迅速变得极其复杂。如果你试图,比如说,煮咖啡,就有无数变量:没有两个厨房是完全相同的;咖啡机的工作方式各不相同;不同的杯子需要不同的握法;咖啡粉、热水和牛奶都需要不同的处理方式。即使这个简单的任务也需要理解一个不断变化的可能性菜单。赫斯特认为,通过VLA实现通用性需要“对(机器人)可能做的所有事情实现完整的数据覆盖”。换句话说,需要一个几乎无限的训练数据池。
杨立昆对整个方法不屑一顾。“在语言方面取得成功的(AI)方法不适用于高维度、连续、有噪声的数据”——而这类数据在机器人领域司空见惯,他在达沃斯说。“你必须使用别的方法。”
“别的方法”的主要竞争者是目前所谓的“世界模型”——一种更多基于视频、三维扫描和传感器数据的组合而非文本训练的AI形式,其构建目的是预测现实世界中行动的结果。目标是构建拥有足够精确的现实内部表征的模型,以捕捉物理世界实际如何运作——物体如何移动、碰撞、坠落和变形。如果机器人学家能在足够忠实于真实世界物理的模拟中训练机器,开发将变得更快、更便宜、更安全,减少对真实世界测试的需求。更具变革性的是,配备世界模型的机器人可以对其周围环境进行推理,而不仅仅是对其做出反应,帮助它们在采取行动之前预判后果。
英伟达和谷歌等公司正在研发这项技术,投资者的资金正涌入高知名度的初创公司。由斯坦福AI研究员李飞飞联合创立的World Labs在2月筹集了10亿美元资金,并于9月底被AMD以82亿美元收购。由杨立昆(前Meta首席AI科学家)联合创立的AMI Labs也在3月筹集了10亿美元。然而,据他们自己承认,现在仍处于早期阶段。去年年底,李飞飞将该领域描述为“初生的”,并补充说“基础性方法仍在建立之中”。在6月的一篇Substack文章中,她描述了令人望而生畏的挑战。目前,世界模型是一个有前景的研究领域,而非通往通用机器人的直接路径,但我们正开始看到它们可能实现什么的微光。
其中一个 glimpse 来自去年4月在旧金山一个机器人实验室发生的一次虽小但可能意义重大的飞跃。
一次突破?
在旧金山Mission区的心脏地带,初创公司Physical Intelligence——简称PI(如π)——专注于开发一个通用大脑,理论上可以将任何机器人变成通用机器人。该公司采用“什么都能塞进去”的方法来训练机器人AI模型,最近观察到了配备世界模型的机器人大脑可能具备的能力的蛛丝马迹。
2024年,PI发布了其首个通用机器人系统π0的详细信息,这是一种VLA,该公司声称它是“迄今为止最有能力和最灵巧的通用机器人策略”。该模型最初在通过遥操作收集的10,000小时专有人类演示集合以及几个开源机器人数据集上进行了训练。2025年春季发布的版本π0.5在更广泛的数据集上进行了训练,包括来自网络的有标签图像,使其更加多才多艺。2025年秋季的更新π0.6在模型中加入了强化学习。
每次更新都对模型性能带来了重要改进,将其能力菜单从缓慢叠衣服扩展到在新环境中收拾东西,再到以更高成功率完成折箱子等任务。然后,在2026年4月,π0.7似乎将PI弹射到了新领域。这个版本利用了一个能力较弱的世界模型,生成执行任务所需步骤的图像。当机器人执行任务时,这个“轻量级”模型向它提供下一步该做什么的快照。
该公司声称,该模型展现出组合式泛化的首批迹象。组合式泛化指的是AI系统通过重新组合训练数据中学到的技能来执行从未接触过的技能的能力。一项测试要求模型“把红薯放入空气炸锅”——这是它此前从未遇到过的任务。在一段演示视频中,机器摸索了一会儿,几次起步失误,最终勉强完成了一次合理的尝试,尽管没有完全完成任务。
加州大学伯克利分校教授、PI联合创始人谢尔盖·莱文对这一潜力感到兴奋:“这实际上是我们第一次令人信服地看到那种组合式泛化,我们可以 basically 要求模型完成我们没有专门收集数据并训练它去做的任务,而它实际上会做出一个过得去的尝试。”
这一成功让团队好奇模型是如何实现这一壮举的。经过一番挖掘,他们发现训练材料中潜藏着相关有标签遥操作数据的片段,包括两个人类操作员使用机器人将空气炸锅篮子推入炸锅的示例。这些数据碎片可能足以让π0.7几乎把红薯放进空气炸锅。
目前,π0.7的泛化能力究竟有多令人印象深刻仍不清楚。尽管如此,鉴于模型接触空气炸锅的时间如此短暂,它让我们得以一窥前沿研究目前能将机器人带到多远。
“70%的成功率就等于不能用”
你可能已经感觉到一种脱节:机器人在实验室中蹒跚迈出的婴儿步伐——“看!它把红薯放进了空气炸锅!”——与许多演示和视频中展示的令人眼花缭乱、栩栩如生的灵巧动作之间的落差。在那些视频中,机器人从在舞台上跳舞到礼貌地端送饮料,无所不能。这类演示往往没有清楚揭示一个关键事实:在许多情况下,是人类在控制机器人,或者精心编写了它的动作脚本。(例如,2025年3月与英伟达CEO黄仁勋一起出现在舞台上的那个机器人,看似在回应他的指令并跟着他走,实际上是由其制造商所说的“幕后的木偶师”远程控制的。)
目前,让机器人知道该往哪里去的完全自主运动规划——尤其是在建筑工地或不熟悉的家庭等新的混乱环境中——在很大程度上仍是一个未解决的挑战。更大的挑战——尽管往往与此相关——在于让机器人处理更大、更模糊的工作,这些工作包含多个任务,需要决定如何完成以及以什么顺序完成。这就是能把盘子放进微波炉的机器人和能成功响应“做晚饭”指令、通过翻冰箱、切食材、开炉灶来完成的机器人之间的区别。谷歌DeepMind对此类任务的最佳尝试——要求其机器人扫视厨房并将所有制作蘑菇烩饭的食材装入篮子——迄今为止都以失败告终。
更添挑战的是,一个实用的机器人基本上必须每次都做对。对于VLA,“当结果从50%成功率提升到70%成功率时,人们非常兴奋,”波士顿动力创始人马克·雷伯特说。“但70%的成功率就等于不能用,对吧?”
少数正在真实环境中测试的人形机器人正在高度受控的环境中执行极其有限的任务。它们远非通用机器人。据Agility称,该公司已在GXO Logistics、亚马逊和舍弗勒旗下的设施中部署了数百台机器人进行试验。但赫斯特说,目前这些机器人针对的是搬运箱子和大储物箱等简单任务。他补充说,即便如此,也花了数年时间才开发出足够安全、让物流公司甚至考虑使用的机器人。就马斯克而言,他在2025年5月声称到年底将有“数千”台Optimus机器人在特斯拉工厂工作,但今年1月他表示公司只有“部分特斯拉Optimus机器人在工厂做简单任务”。
虽然人形机器人正开始进入工厂车间,但跳入家庭将更加困难。现在,如果你愿意,可以预订1X Neo家用机器人,预计今年晚些时候可以交付。售价2万美元,它承诺承担“家中无聊琐碎的任务”——收拾碗碟、应门、整理客厅——“让你专注于对你重要的事情”。这个五英尺六英寸高的机器人的设想是有一天能自主执行所有这些任务,但目前需要远程人工操作员来完成大部分事情。(是的,需要经过许可,人才能通过机器人的摄像头窥视你的家。)当被问及完全自主的机器人还需要多久才能准备好从事家务工作时,赫斯特说:“如果我必须选一个数字,我会说还要10年,机器人才能……真正在人们家中做有用的事情。”
当那一天到来时,这些机器人很可能来自中国,因为中国在生产方面远远领先于西方。据市场情报公司Omdia和中国机器人公司宇树科技的数据,2025年出货的约15,000台人形机器人中,近90%由中国公司制造。宇树科技去年出货的人形机器人数量超过任何其他公司,其生产的一款型号售价不到6,000美元。这样的实惠价格可能大大有助于让机器人对消费者更具吸引力,尽管该公司预计其机器将首先用于工业应用。(顺便说一句,如果你想知道谁在购买所有这些中国机器人,美联社最近报道称,订单主要来自企业和学术实验室以及国有企业。)
我们以前来过这里
建造人形机器人的梦想由来已久:早在1495年,列奥纳多·达·芬奇就勾画了一个由缆绳和滑轮控制的机械骑士的设计图。整个20世纪,科幻狂梦中的机器来来去去。
西屋电气那个七英尺高的长腿盒子Elektro在1939年纽约世界博览会上亮相,还会抽烟。日本早稻田大学1973年建造的WABOT-1是第一个全尺寸、可编程的人形机器人。本田2000年发布的ASIMO可能是第一台证明自己至少有一点能力的此类机器——它至少在一定程度上能爬楼梯、识别人脸、自主穿过空间。但该机器人在2018年停产,无法在演示之外取得足够进展以变得有用。
所有这些在当时都是令人印象深刻——甚至令人瞠目结舌——的工程壮举。但没有一个准备好驾驭现实世界。今天的机器人,即使拥有先进AI的变革力量,仍然面临同样的生存挑战。
深度解读
人工智能
别被骗了——大语言模型不会推理
AlphaGo对阵围棋冠军李世石十年后,今天的AI仍然没有触及使那场胜利成为可能的机制。
AI的递归自我改进可能不会那么快到来
AI智能体似乎还不够有创造力,无法开展真正创新的开放式AI研究。
别被这个夏天的AI炒作骗了
关于AGI和新能力的激动 claims 在审视之下很快就会瓦解。
这些初创公司正在追逐大语言模型的下一个大事件
来认识一下这些紧追AI巨头的新秀。
保持联系
获取《麻省理工科技评论》的最新动态
发现特别优惠、热门报道、即将举办的活动等。
英文来源:
AI breakthroughs in robotics won’t change your life any time soon
Advances in AI offer tantalizing glimpses of a future in which robots navigate the world the way humans do. The question is whether the same techniques that fueled AI’s recent progress will be enough to get there, or if an entirely new path is required.
The story is a collaboration between MIT Technology Review and Aventine, a non-profit research foundation that creates and supports content about how technology and science are changing the way we live.
A robot shaped like a human—white with a black head and torso—has been popping up on video feeds. Perhaps you’ve seen it dance or pass popcorn, put trash in a bin, vacuum, or press the button of a microwave. Or maybe you’ve watched it fall backward while handing out water bottles or struggle to iron a shirt.
This would be Tesla’s Optimus, an AI-powered humanoid robot that Elon Musk, the company’s CEO, believes will be “not just Tesla’s biggest product ever, but probably the biggest product ever,” headed to work on factory floors and, later, in our homes. Eventually it “will have human and then superhuman dexterity,” he told shareholders in July. Optimus robots could automate almost all human labor—from hauling sheet metal to folding laundry—for as little as $20,000 each, Musk argues. Speaking at the World Economic Forum’s annual meeting in Davos, Switzerland, in January, he predicted they could be on sale to the public by the end of 2027.
Musk is not alone in his evangelism. Marc Andreessen, cofounder and general partner of the Silicon Valley venture capital firm Andreessen Horowitz, has said that robotics could become the “biggest industry in the history of the planet.” In January, Jensen Huang, CEO of Nvidia, said that humanoid robots would match human-level ability this year. According to Morgan Stanley, the number of robots that “resemble and act like humans” is likely to reach nearly 1 billion by 2050, creating a market worth over $5 trillion.
Such proclamations are in large part fueled by the idea that the same AI advances behind tools like OpenAI’s ChatGPT and Anthropic’s Claude will enable a new generation of robots to imitate human movement the way chatbots imitate human language. But many robotics researchers are skeptical, arguing that such assumptions minimize the challenges of using an intelligence built on language and images to master the infinite variability of the physical world. “None of those companies [building humanoid robots]—absolutely none of them—has any idea how to make those robots smart enough to be useful,” Yann LeCun, often referred to as one of the godfathers of AI, said at another event during the January Davos conference.
Researchers also point out that the tendency to conflate humanoid robots made to resemble people with so-called generalist machines able to learn and perform multiple tasks is misleading. ”It’s very easy to make a robot that looks like a person,” explains Jonathan Hurst, cofounder and chief robot officer of Agility Robotics and professor of robotics at Oregon State University. “It is dramatically more difficult to make a machine that moves or behaves dynamically or physically like a person.”
These tensions—over whether all-purpose humanoid robots are just around the corner or nowhere in sight, and whether current forms of AI are all that’s needed to perfect them—are playing out in robotics labs across the country, where the hype over timelines is obscuring painstaking but meaningful progress.
A decade or so ago, a series of breakthroughs led to a generative AI revolution that turned the long-imagined possibility of artificial intelligence into reality. Roboticists—though they disagree on exactly when this will happen—believe that an equally transformative revolution is possible in robotics, one that will endow machines with physical intuition and fluidity that has long been out of reach. As progress in robotics inches forward, the question is whether the same methods and tools that fueled advances in AI are enough to get there, or if an entirely new path is required.
Robots meet advanced AI
To see one of the smartest robot brains working today, it’s worth looking at what Google DeepMind can do with a piece of equipment called ALOHA 2, short for “A Low-cost Open-source Hardware System for Bimanual Teleoperation.”
Roboticists have long clashed over whether a humanlike form is necessary for generalist robots, with proponents arguing that it will help them slot into the world as it exists and detractors saying it’s not worth the trouble. ALOHA 2 reflects this second way of thinking. Not much to look at, it’s just a pair of arms, some grippers, and a couple of cameras. But despite its seeming simplicity, it is a workhorse for researchers at Google DeepMind, who use it to test their most advanced AI for robotics system, Gemini Robotics, in their various labs.
When controlled by Gemini Robotics, ALOHA 2 becomes more of a generalist robot, in the sense that it can perform any number of tasks based on examples it’s been trained on. Ask it to pack a lunchbox and, as evidenced by a video of this exercise, it can use two pincer grippers to delicately place a piece of white bread into a Ziploc bag, close it, place a bunch of grapes in a Tupperware container, secure the lid, and then carefully move the items into a lunchbox before zipping it up.
It’s not a great lunch. But the fact that the robot can put it together represents an objective step forward from what was possible even, say, three years ago.
This is in large part due to AI and its impact on what are known as robot policies, which controls how a general-purpose robot will need to assess and understand its surroundings, plan how to move within them, and then perform its task correctly.
Historically, these policies were based on rules developed by engineers who hard-coded them into the robot’s software—thousands of lines of code that would determine each millimeter of a robot’s movements in hundreds of tasks. What’s been happening for the last few years—and what is largely responsible for the optimism about generalist robots—is that robot policies are being handed over to advanced AI systems instead of being coded into the robot’s software. This first happened with VLMs, or vision-language models. These are similar to large language models, but they’re trained on images as well as words. Show a VLM a picture of a coffee spill and ask it to find a tool to clean up the mess, and it can identify a nearby cloth. This sort of immediate contextual understanding didn’t exist a couple of years ago when robot policies were hard-coded.
Next came vision-language-action models, which enable robots to assess their environment and take action within it. The models do this by adding yet another component: motion commands. VLAs are trained on a series of images or videos related to performing a given task along with associated data about how a robot arm moves to perform it. That movement data is typically collected through teleoperation, in which a human uses remote controls to lead a robot through an action. This sort of training allows the AI to learn how to command the robot to move and operate during a given task. Place a VLA-powered robot in front of a desk and tell it to “close a laptop” or “wrap up the headphone wire,” and it will survey the scene, identify the relevant object, plan a way to execute the request, and then swing its arms into action—at least if it has seen this task accomplished before.
The Gemini Robotics model is a VLA, trained on many hours of human demonstrations depicting a vast array of different actions. As a result, it can perform relatively complex tasks like picking up snow peas with kitchen tongs, doing origami, or putting together a simple lunch. It’s impressive, but there’s a glaring limitation: For now, if a robot controlled by a VLA is asked to perform a task that falls outside its training set, it’s highly likely to fail.
“Thinking about the space of all tasks, a real generalist policy would be able to do everything along that spectrum,” says Edward Johns, a robotics professor at Imperial College London. Today, though, a Gemini Robotics model can do only “a few things here and a few things there.”
The search for true generality
So how do we get robots to be able to do more things? The usual answer is probably not surprising: Train them on more data.
More data, the thinking goes, equals more examples, and more examples equals more generality. Google DeepMind, for instance, wants to pull together “as much data as possible,” says Pannag Sanketi, a former tech lead in robotics at the company who’s currently working on his own AI robotics project. But where to get it? Large language models had the benefit of oceans of existing text for training. There is no corresponding pool of high-quality physical demonstrations on which to train robots. Researchers have a few ways to make up for this, but all have flaws. One is to employ large numbers of people to create and collect teleoperation data (costly and time-consuming). Another is to train VLAs on videos of people performing activities (the resulting data quality is poor). Yet another is to deploy robots in the real world and use data collected from those experiences to further refine AI models (robots aren’t safe or reliable outside labs). Sanketi thinks a “multi-prong” approach that uses data collected from all these sources is the most likely path forward.
But the belief that training data alone is the answer is far from universal. Agility’s Hurst describes it as “a fundamentally flawed premise.”
The issue is that tasks in the real world quickly explode in complexity. If you’re trying to, say, make coffee, there are myriad variables: No two kitchens are identical; coffee machines work in different ways; different cups require different grips; coffee grounds, hot water, and milk all need to be handled differently. Even this simple task requires understanding an ever-changing menu of possibilities. Achieving generality through VLAs, Hurst argues, would require “complete data coverage of all of the things that [a robot] could ever do.” Or, in other words, an almost infinite pool of training data.
LeCun is dismissive of the whole approach. “The [AI] approaches that have been successful for language do not work for high-dimensional, continuous, noisy data”—the kind of data that is commonplace in robotics, he said in Davos. “You have to use something else.”
The leading contender for “something else” is the so-called world model—a form of AI trained less on text than on a combination of video, three-dimensional scans, and sensor data and built to predict the outcomes of actions in the real world. The aim is to build models that possess an internal representation of reality precise enough to capture how the physical world actually operates—how objects move, collide, fall, and deform. If roboticists could train machines in simulations faithful enough to real-world physics, development would become faster, cheaper, and safer, reducing the need for real-world testing. Even more transformative, robots equipped with world models could reason about their surroundings rather than merely reacting to them, helping them anticipate the consequences of an action before taking it.
Companies like Nvidia and Google are working on the technology, and investor cash is pouring into high-profile startups. World Labs, cofounded by the Stanford AI researcher Fei-Fei Li, raised $1 billion in funding in February and was acquired by AMD at the end of September for $8.2 billion. AMI Labs, cofounded by LeCun (formerly Meta’s chief AI scientist), also raised $1 billion in March. Yet by their own admission, it is still early days. Late last year Li described the field as “nascent,” adding that “foundational approaches are still being established.” In a June Substack she described daunting challenges. For now, world models are a promising area of research rather than an immediate route to general-purpose robotics, but we are beginning to see glimmers of what they could achieve.
One such glimpse came with a small but potentially significant leap forward that took place in a San Francisco robotics lab last April.
A breakthrough?
In the heart of San Francisco’s Mission District, the startup Physical Intelligence—or PI (as in π), as it likes to be known—is focused on developing a universal brain that could, theoretically, turn any robot into a generalist. Using an everything-including-the-kitchen sink approach to training AI models for robots, the company recently observed a hint of what a robotic brain equipped with a world model could be capable of.
In 2024, PI published details of its first generalist robotics system, called π0, a VLA it claimed was the “most capable and dexterous generalist robot policy to date.” The model was initially trained on a 10,000-hour proprietary collection of human demonstrations gathered through teleoperation as well as several open-source robot datasets. A version released in spring 2025, π0.5, was trained on a wider variety of datasets, including labeled images from the web, lending it more versatility. A fall 2025 update, π0.6, added reinforcement learning to the model.
Each update yielded important improvements to the model’s performance, increasing its menu of abilities from slowly folding laundry to putting things away in new environments to completing tasks like folding boxes with a higher success rate. Then, in April 2026, π0.7 seemed to catapult PI into new territory. This version makes use of a less powerful world model that generates images of steps necessary to perform a task. As the robot undertakes the job, this “lightweight” model feeds it snapshots of what to do next.
The company claims that the model exhibits the first signs of compositional generalization, a term for AI systems’ ability to perform skills they’ve never been exposed to by recombining ones learned in their training data. One test involved asking a model to “load a sweet potato into the air fryer”—a task it had never previously encountered. In a demonstration video, the machine futzes around a little, makes a few false starts, and eventually manages a reasonable effort, though it doesn’t finish the task completely.
Sergey Levine, a professor at the University of California, Berkeley, and a cofounder of PI, is excited by the potential: “It’s actually the first time that we’ve convincingly seen that kind of compositional generalization, where we can basically ask the model to do tasks that we did not specifically collect data for and train it to do, and it’ll actually make a passable attempt.”
The success led the team to wonder how the model was able to achieve such a feat. After some digging, they found snippets of relevant labeled teleoperation data lurking in the training material, including two examples of a human controller using the robot to push an air fryer basket into the fryer. Those shreds of data might have been enough to enable π0.7 to almost air-fry a sweet potato.
For now, it remains unclear just how impressive π0.7’s abilities to generalize are. Still, given how fleeting the model’s exposure to air fryers had been, it offers a glimpse into how far cutting-edge research can currently take robots.
“70% success is like it doesn’t work”
You might be sensing a disconnect between the halting baby steps robots are making in labs—“Look! It put a sweet potato into an air fryer!”—and the dazzling, lifelike nimbleness on view during many demonstrations and videos, where robots are seen doing everything from dancing on a stage to courteously serving drinks. Such demos often don’t clearly reveal a key fact: In many instances, humans are controlling the robot or have carefully scripted its actions. (The robot that appeared onstage with Nvidia CEO Jensen Huang in March 2025, for example, seemingly responding to his instructions and following him around, was remote-controlled by what its makers called “a puppeteer behind the scenes.”)
For now, fully autonomous motion planning so that a robot knows where it should go—especially in new, chaotic environments like a construction site or a unfamiliar home—remains a largely unsolved challenge. A bigger challenge still—albeit one that is often related—lies in getting robots to tackle larger, more ambiguous jobs that include multiple tasks and require decisions about how and in what order they’re done. This would be the difference between a robot that can put a plate into a microwave and one that can successfully respond to the prompt “Make dinner” by exploring the refrigerator, chopping ingredients, and firing up the stove. Google DeepMind’s best attempts at something like this—which involved asking its robot to survey a kitchen and pack all the ingredients for a mushroom risotto into a basket—have so far resulted in failure.
Adding to the challenge, a practical robot must essentially get it right every time. With VLAs, “people are very excited when their result goes from 50% success to 70% success,” says Marc Raibert, founder of Boston Dynamics. “But 70% success is like it doesn’t work, right?”
The few humanoids that are being tested in real-life settings are undertaking extremely limited tasks in tightly controlled environments. They’re far from generalists. Agility has hundreds of robots deployed across trials in facilities owned by GXO Logistics, Amazon, and Schaeffler, according to the company. But for now, Hurst says, the robots are targeting simple tasks such as moving bins and totes around. Even then, he adds, it took years to develop robots safe enough for logistics firms to even contemplate using them. For his part, Elon Musk claimed in May 2025 that “thousands” of his Optimus robots would be working at Tesla factories by the end of the year, but in January of this year he said that the company had only “some of the Tesla Optimus robots doing simple tasks in the factory.”
While humanoids are starting to venture onto the factory floor, making the jump to households will be even more difficult. Right now, should you so desire, you can preorder the 1X Neo home robot, expected to be ready for delivery sometime later this year. Yours for $20,000, it promises to take on “the boring and mundane tasks around the house”—putting away dishes, answering the door, tidying the living room—“so you can focus on what matters to you.” The idea is for this five-foot-six-inch robot to one day perform all those tasks autonomously, but for now a remote human operator is needed for it to do most things. (Yes, a person would need permission to peer into your home through the robot’s cameras.) Asked how long it will be until fully autonomous robots are ready for domestic work, Hurst said, “If I had to pick a number, I’d say it’s 10 years before robots are … actually doing useful things in people’s homes.”
When that happens, the robots might well be Chinese, as China is well ahead of the West in terms of production. Nearly 90% of the roughly 15,000 humanoid robots shipped in 2025 were made by Chinese companies, according to the market intelligence company Omdia and the Chinese robotics firm Unitree. One model produced by Unitree, which shipped more humanoid robots than any other company last year, costs less than $6,000. Such an affordable price could go a long way toward making robots more attractive to consumers, though the company expects its machines to be used in industrial applications first. (If you’re wondering who is buying all these Chinese robots, by the way, the AP recently reported that orders come predominantly from corporate and academic labs and state-owned enterprises.)
We’ve been here before
The dream of building a humanoid robot runs deep: As far back as 1495, Leonardo da Vinci sketched out designs for a mechanical knight, controlled by cables and pulleys. Through the 20th century, machines of sci-fi fever dreams have come and gone.
Westinghouse’s seven-foot-tall box on legs, Elektro, hit the New York World’s Fair in 1939, smoking a cigarette. WABOT-1, built by Waseda University in Japan in 1973, was the first full-scale, programmable humanoid robot. Honda’s ASIMO, unveiled in 2000, was probably the first such machine to prove at all competent—it could, at least to some degree, climb steps, recognize faces, and autonomously move through spaces. But the robot was discontinued in 2018, unable to advance far enough beyond what it could do in demonstrations to be useful.
All, at the time, were impressive—even jaw-dropping—feats of engineering. But none were ready to navigate the real world. Today’s robots, even with the transformative power of advanced AI, still face the same existential challenge.
Deep Dive
Artificial intelligence
Don’t be fooled—LLMs don’t reason
Ten years after AlphaGo’s match against Go champion Lee Sedol, today’s AI still isn’t tapping into the machinery that made that win possible.
AI’s recursive self-improvement might not come so quickly after all
AI agents are not yet creative enough to carry out genuinely innovative open-ended AI research, it seems.
Don’t be fooled by this summer of AI hype
Breathless claims about AGI and new capabilities fall apart pretty quickly under scrutiny.
These startups are chasing the next big thing in LLMs
Meet the new kids nipping at the heels of the AI giants.
Stay connected
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.
文章标题:机器人技术中的人工智能突破不会很快改变你的生活
文章链接:https://news.qimuai.cn/?post=5254
本站文章均为原创,未经授权请勿用于任何商业用途