2026年的消费电子市场里，没有任何一类产品能像人形机器人这样矛盾。曾经动辄几十万、只出现在展会舞台上的双足人形设备，正在批量上架电商平台，万元级产品打破大众对人形机器人的价格想象。订单在增长、产能在爬坡、发布会轮番登场，厂商不断向公众描绘一个图景，人形机器人即将走入家庭，承接家务、陪护、家政劳动。但当你兴冲冲把它搬回家，指望它帮你叠衣服、收拾碗筷的时候，大概率会失望。打开电商评论区、B站测评视频、玩家社群，真实的画面截然不同。已经下单收货的尝鲜用户，很少指望它解放双手。更多人把机器人摆在家里客厅，用来跳舞、对话、编程调试、拍摄短视频。厂商们心里也清楚这一点，它们卖的不是工具，而是一张通往未来的门票。
In the 2026 consumer electronics market, no product category is as contradictory as humanoid robots. Bipedal humanoid devices that once cost hundreds of thousands and appeared only on exhibition stages are now being listed in batches on e-commerce platforms, with ten-thousand-yuan products breaking public price imaginations of humanoid robots. Orders are growing, capacity is ramping, launch events come one after another, and manufacturers continuously paint a picture for the public—humanoid robots are about to enter homes, taking on housework, companionship, and domestic labor. But when you excitedly bring one home hoping it will fold your clothes and clear the dishes, you will likely be disappointed. Opening e-commerce comment sections, Bilibili review videos, and player communities, the real picture is completely different. Early adopters who have ordered and received rarely expect it to "free their hands." More people place robots in living rooms to dance, converse, debug programming, and shoot short videos. Manufacturers know this well—they are not selling a tool but a ticket to the future.
2026年被行业定义为人形机器人量产兑现元年，这个判断已经落地成真。过去几年，无论是国内初创企业还是海外科技巨头，产品大多只用于展会展示、高校科研与企业B端试点，不面向普通消费者零售。2025年下半年起，局面开始发生根本性转变，多家国产厂商陆续发布支持现货或者预定交付的量产机型，人形机器人正式进入面向个人消费者的售卖阶段。市场数据印证了交付规模的快速扩张。调研机构Smart Analytics Global数据显示，2026上半年全球人形机器人出货量约1.91万台，同比增长近3倍，中国厂商占据97%以上份额，智元、宇树两家合计占据全球出货量四分之三。
2026 is defined by the industry as the first year of humanoid robot mass production realization, and this judgment has come true. Over the past few years, whether domestic startups or overseas tech giants, products were mostly used for exhibition displays, university research, and enterprise B-end pilots, not retail to ordinary consumers. From the second half of 2025, the situation began a fundamental shift—multiple domestic manufacturers successively released mass-produced models supporting in-stock or pre-order delivery, and humanoid robots officially entered the personal consumer sales stage. Market data confirms rapid delivery scale expansion. Research firm Smart Analytics Global data shows global humanoid robot shipments in the first half of 2026 were about 19,100 units, nearly tripling year-over-year, with Chinese manufacturers holding over 97% share, and AgiBot and Unitree together accounting for three-quarters of global shipments.
另一大变化是，价格曲线快速下行。松延动力的小布米定价9998元，首次把双足人形机器人拉进万元以内。宇树科技的Unitree R1从3.99万元降至2.99万元起，现货发售，一周后直接登顶京东人形机器人销量榜。优必选U1系列覆盖11.98万元到99万元，走的是情感陪伴路线。智元灵犀X2青春版9.8万元，加速进化Booster K1教育版6.69万元。星尘智能T1把起售价压到8.99万元。厂商集体降价的底气也来自供应链的成熟。宇树科技招股书显示，其人形机器人单价从2023年的59.34万元降至2025年的16.64万元，两年缩水约72%。核心零部件国产化率已提升至75%，无框力矩电机国产化率超过85%，谐波减速器国产化率约60%至70%。
Another major change is the rapid downward price curve. Songyan Dynamics' Xiaobumi is priced at 9,998 yuan, first bringing bipedal humanoid robots below 10,000 yuan. Unitree's Unitree R1 dropped from 39,900 to 29,900 yuan, sold in stock, topping JD.com's humanoid robot sales chart within a week. UBTECH's U1 series covers 119,800 to 990,000 yuan, taking the emotional companionship route. AgiBot Lingxi X2 Youth Edition at 98,000 yuan, Accelerated Evolution Booster K1 Education Edition at 66,900 yuan. Astribot T1 pushed its starting price to 89,900 yuan. Manufacturers' collective price cuts also draw confidence from supply chain maturity. Unitree's prospectus shows its humanoid robot unit price dropped from 593,400 yuan in 2023 to 166,400 yuan in 2025, shrinking about 72% in two years. Core component localization rate has risen to 75%, frameless torque motor localization exceeding 85%, harmonic reducer localization about 60% to 70%.
不过，价格差距的背后也是功能与硬件减配。售价2.99万元的R1 AIR入门版，相比3.99万元的标准版，整体关节自由度从26个缩减至20个，直接砍掉头部2个自由度、腰部2个旋转自由度，单臂自由度从5个降至4个，彻底失去腰部灵活转动、头部多角度俯仰转向的能力，无法完成侧身取物、低头捡拾地面杂物、多角度观察环境等基础家政动作。同时，R1 AIR取消了标准版搭载的高精度力控模块，抓取容错率降低。更关键的是，入门版机型不支持高阶二次开发，从硬件层面锁死了实用潜力。将行业价格压至万元门槛的松延动力小布米Bumi，则直接舍弃了通用灵巧手配置，仅保留拇指加简易双指极简结构，无完整五指关节，不具备任何精细抓取、操作能力。
However, behind the price gap is also feature and hardware downsizing. The 29,900-yuan R1 AIR entry version, compared with the 39,900-yuan standard version, reduces total joint degrees of freedom from 26 to 20, directly cutting 2 head DOF and 2 waist rotation DOF, reducing single-arm DOF from 5 to 4, completely losing flexible waist rotation and multi-angle head pitch, unable to complete basic housework actions like side-reaching for objects, bending down to pick up floor debris, and multi-angle environmental observation. Meanwhile, R1 AIR removes the standard version's high-precision force control module, reducing grasp fault tolerance. More critically, the entry model does not support high-level secondary development, locking down practical potential at the hardware level. Songyan Dynamics' Xiaobumi Bumi, which pushed industry prices to the 10,000-yuan threshold, directly abandoned general dexterous hand configuration, keeping only a thumb plus simple two-finger minimalist structure, without complete five-finger joints, lacking any fine grasping or manipulation capability.
结合各家厂商官方参数、钛媒体2025人形机器人消费市场调研报告、IT之家硬件实测数据，目前10万元以下面向个人销售的消费级人形机器人，普遍存在硬件规格大幅缩水的问题。整机单臂最大负载集中在1.5至2.5kg区间，仅能抬起轻质玩偶、空水杯等物品，远低于家庭日常收纳、搬运、整理所需的5kg实用负载标准。入门款机型标准续航普遍为2至3小时，部分极致低价机型续航仅1小时左右，且全系未搭载智能补电、自主回充功能。正如工信部信息通信经济专家委员会委员盘和林在接受采访时所言，本轮价格下探并非行业技术全面成熟的自然结果，而是一场以生态卡位为核心的主动布局。
Combining manufacturers' official parameters, TMTPost's 2025 humanoid robot consumer market research report, and IT Home hardware testing data, current consumer-grade humanoid robots under 100,000 yuan sold to individuals generally have significant hardware specification shrinkage. Whole-machine single-arm maximum payload is concentrated in the 1.5–2.5 kg range, only able to lift lightweight dolls, empty cups, and similar items, far below the 5 kg practical payload standard needed for daily home storage, carrying, and organizing. Entry models' standard battery life is generally 2–3 hours, with some extreme low-price models lasting only about 1 hour, and the entire series lacks smart recharging and autonomous return-to-charge functions. As Pan Helin, a member of MIIT's Information and Communication Economic Expert Committee, said in an interview, this round of price reduction is "not a natural result of comprehensive industry technology maturity, but a proactive布局 centered on ecological positioning."
买的人比预想的多。优必选U1系列预售订单达1.3361万笔，上线首日全通路订单突破1.3万台，直接超越公司2025年全年人形机器人销量的12倍以上。松延动力小布米在春晚后迅速成为京东平台人形机器人类目销量第一名。启元机器人在上海淮海中路开出全球首店，首批交付的Q1、T1定价19999元起，现场有科技博主用它给7个月大的宝宝喂辅食。但买家的心态很微妙。在小红书上，购买了宇树科技R1的网友大多提到目的是买来玩玩。在宇树科技官方旗舰店的评价区，也有不少消费者提到购买这台机器人是送给小朋友。翻看电商订单、社群访谈、B站、小红书大量实测内容，当前C端买家的画像高度集中，机器人爱好者、AI技术发烧友、高校科研团队、短视频创作者、科创培训机构。普通工薪家庭极少愿意花数万元，购买一台无法承担家务劳动的人形设备。简单来说，他们在玩，不在用。
More people bought than expected. UBTECH's U1 series pre-sale orders reached 13,361, with first-day all-channel orders exceeding 13,000 units, directly surpassing 12 times the company's full-year 2025 humanoid robot sales. Songyan Dynamics' Xiaobumi quickly became JD.com's humanoid robot category sales number one after the Spring Festival Gala. Qiyuan Robotics opened its global first store on Shanghai's Huaihai Middle Road, with first-batch Q1 and T1 priced from 19,999 yuan, where a tech blogger used it to feed complementary food to a 7-month-old baby. But buyer mentality is subtle. On Xiaohongshu, netizens who bought Unitree's R1 mostly mentioned the purpose was "buying to play." In Unitree's official flagship store review section, many consumers also mentioned buying the robot as a gift for children. Browsing e-commerce orders, community interviews, Bilibili, and Xiaohongshu大量 hands-on content, current C-end buyer profiles are highly concentrated—robot enthusiasts, AI tech fans, university research teams, short video creators, and STEM training institutions. Ordinary working families rarely spend tens of thousands on a humanoid device that cannot handle housework. Simply put, they are playing, not using.
不少测评博主拿到机器之后，给出的结论高度一致，这是一台昂贵的智能玩具，而非家政工人。大众对于人形机器人最朴素的需求，是替代保洁，拖地、擦桌子、收拾杂物、叠衣服、拿取物品。但目前量产机型普遍无法稳定完成这类任务。灵巧手是一个明显短板。家务场景里大量精细操作，拧瓶盖、叠衣物、收拾散落杂物，都依赖高精度、高可靠性的灵巧手。而当前万元到十万元区间的量产消费级人形机器人，很多并没有标配高性能灵巧手，抓取负载极低，墨茵M1手臂末端最大负载仅1.5公斤。抓取动作成功率不稳定，物体位置稍微偏移，就会抓取失败，很容易出现掉落、打滑。
Many review bloggers, after getting the machines, reached highly consistent conclusions—this is an expensive smart toy, not a domestic worker. The public's most basic demand for humanoid robots is replacing cleaning—mopping, wiping tables, organizing clutter, folding clothes, fetching items. But current mass-produced models generally cannot stably complete such tasks. Dexterous hands are an obvious shortcoming. Housework scenarios involve大量 fine operations—unscrewing bottle caps, folding clothes, organizing scattered clutter—all relying on high-precision, high-reliability dexterous hands. Current mass-produced consumer humanoid robots in the 10,000 to 100,000 yuan range often do not come standard with high-performance dexterous hands, with extremely low grasp payloads—Moyin M1's arm-end maximum payload is only 1.5 kg. Grasp success rates are unstable; slight object position shifts cause grasp failure, easily leading to drops and slips.
但最根本的障碍是泛化能力。宇树科技创始人王兴兴曾在世界机器人大会上坦言，在充分标定的固定场景内，人形机器人任务成功率可以接近100%，但一旦环境或物品发生微小改变，执行效果便会断崖式下滑，厘米甚至毫米级的物理偏差，就足以造成任务失败。而家庭恰恰是对泛化能力要求最高的场景。杯子的位置每天不一样，衣服的材质和形态随时变化，椅子可能被移动，冰箱门可能没关好，这些对人是理所当然的小事，对机器人是巨大的技术挑战。网络上大量真实用户评价，清晰暴露这种预期落差。有玩家在商品评论区写道，机器人跳舞、对话效果不错，拍视频很吸睛，但想让它帮忙递杯水，十次里成功两三次，更多时候会撞向桌子。
But the most fundamental barrier is generalization capability. Unitree founder Wang Xingxing admitted at the World Robot Conference that in fully calibrated fixed scenarios, humanoid robot task success rates can approach 100%, but once the environment or objects change slightly, execution effects plummet, with centimeter or even millimeter-level physical deviations enough to cause task failure. And homes are precisely the scenario with the highest generalization requirements. Cup positions differ daily, clothing materials and shapes change constantly, chairs may be moved, refrigerator doors may not be properly closed—these are trivial to humans but enormous technical challenges for robots. A large number of real user reviews online clearly expose this expectation gap. One player wrote in a product comment section that the robot dances and converses well and is eye-catching in videos, but asking it to hand over a cup succeeds two or three times out of ten, with more frequent table collisions.
既然现在的产品远谈不上实用，厂商为什么急着推向市场。答案是占位。当前几乎所有消费级人形机器人的商业化逻辑，都不是卖工具，而是抢用户、抢数据、抢生态位。谁先抢到用户和数据，谁就赢下一轮，这是行业里的共识。数据飞轮的逻辑很清晰，越多的用户使用，产生越多的真实场景数据，模型迭代越快，产品越成熟，然后吸引更多用户。东方财富证券预测，2026年国内表演机器人租赁市场规模将突破百亿元。租赁市场反而成了当下最清晰的商业闭环。而普华永道在《2026年智能机器人产业发展白皮书》中给出了一个更冷静的判断，人形机器人的核心价值在于场景独特性，而非对其他形态的全面替代。这句话翻译过来就是，人形机器人不是万能工具，至少在现阶段不是。
Since current products are far from practical, why are manufacturers rushing to market. The answer is positioning. Currently almost all consumer-grade humanoid robot commercialization logic is not selling tools but grabbing users, data, and ecological niches. "Whoever grabs users and data first wins the next round"—this is industry consensus. The data flywheel logic is clear—more users using it generates more real-scenario data, faster model iteration, more mature products, then attracting more users. East Money Securities predicts the domestic performance robot rental market will exceed 10 billion yuan in 2026. The rental market反而 became the clearest commercial closed loop currently. And PwC's "2026 Intelligent Robot Industry Development White Paper" gives a calmer judgment—humanoid robots' core value lies in scenario uniqueness, not comprehensive replacement of other forms. Translated, humanoid robots are not universal tools, at least not at this stage.
消费者的期望和行业的现实之间，存在一个巨大的时间差。人们想要的是能扫地的机器人，厂商给出的是能陪伴的机器人，人们期待的是帮我干活，厂商提供的是陪你聊天。这个错位并非厂商故意为之，而是技术成熟度决定的能力边界。但厂商选择在这个时间点把产品推向市场，意图很明确，先用低价拉更多人上车，让用户习惯家里有个机器人这件事，让数据飞轮转起来。等到大脑进化到能理解家务、手进化到能干活的那一天，谁手里有最多的用户和最强的数据壁垒，谁就能定义下一个时代的行业标准。当然，如果你现在花两三万买了台人形机器人，又何尝不算是拥有了一张通往未来的早鸟票。
Between consumer expectations and industry reality exists a huge time gap. People want a robot that can sweep floors; manufacturers give a robot that can accompany. People expect "help me work"; manufacturers provide "chat with you." This mismatch is not intentional but a capability boundary determined by technology maturity. But manufacturers choosing this time to market products have clear intent—first use low prices to pull more people aboard, let users get used to "having a robot at home," and let the data flywheel spin. When the day comes that the "brain" evolves to understand housework and the "hand" evolves to work, whoever has the most users and strongest data barriers can define the next era's industry standards. Of course, if you spend twenty or thirty thousand on a humanoid robot now, is it not owning an early-bird ticket to the future.
