百思不得姐

一颗「脑」装进27000台真机！遍布50国_我的网站

A |     新智元报道          上海世博展览馆，WAIC 2026，H1-A530展位前围了三层人。    Ohio residents this week began voting on a constitutional amendment whose fate will ultimately determine abortion access in the state.Voters are considering two ballot initiatives; ballots began being accepted on Wednesday and go through Election Day on Nov. 7.The two measures are: Issue 1, the amendment that if approved would add abortion rights to the state constitution; and Issue 2, which would legalize recreational marijuana.The proposed abortion amendment would establish an "individual right to one's own reproductive medical treatment including but not limited to abortion" and then creates legal protections for those individuals, among other guarantees like for contraception and fertility treatment.The marijuana measure, if passed, would make Ohio the 24th state to legalize cannabis and would allow adults 21 years and older to buy and possess up to 2.5 ounces of flower and 15 grams of extract.In August, Ohio voters resoundingly rejected a contentious, Republican-backed ballot measure that would have raised the threshold for future changes to the state constitution -- an effort that would have specifically made it more difficult for the abortion amendment to pass.When the U.S. Supreme Court struck down its previous Roe v. Wade ruling, in 2022, the justices returned the question of abortion access to individual states. Since then, while some states have largely ceased abortion access, others have taken up ballot measures that put the question before voters. In those cases -- in both red and blue states so far -- voters have rejected the potential for abortion restrictions and bans or elected to expand access.On Sept. 27, the Ohio Supreme Court started hearing arguments about the state's six-week abortion ban, which was previously put on hold by a lower court.The success or failure of the abortion ballot measure could render that case moot, if state voters add abortion access to their constitution.Both sides have built up mobilization efforts over the past few weeks -- knocking on doors, canvassing and hosting rallies across the state."Whether you're pro-life or pro-choice, Issue 1 just goes too far," Ohio Gov. Mike DeWine, a Republican, said at a recent event at the Ohio GOP's headquarters.Ohio Democrats claimed last week to have been inching toward 100,000 doors knocked and more than 100,000 calls made in the time since their win in the August special election."We're not leaving anything behind as we work to protect women's rights to choose and prevent Ohio's extreme abortion ban from taking effect," Ohio Democratic Chairwoman Elizabeth Walters said during a recent press briefing.There's also been significant financial investment in the race. According to numbers from AdImpact released last week, $12.3 million had been spent or reserved in advertising around the abortion ballot measure, with anti-amendment groups outspending amendment supporters by about $1.6 million.

The two sides make their arguments

Both Ohioans United for Reproductive Rights, a coalition of groups supporting abortion rights, and allied Ohio Democrats have sounded the alarm on what they claim to be foul play by opponents of the abortion amendment.Last month, the state Supreme Court allowed the Republican-controlled Ohio Ballot Board -- which is overseen by Secretary of State Frank LaRose, who is also running for the U.S. Senate -- to use amendment language that includes the phrase "unborn child."Democrats on the board had asked for measure's language to remain what was on the petition that voters originally signed, which did not include that term.Amendment supporters have argued it was rewritten to confuse voters and is misleading about what the measure would actually do."They're running scared because they know their extreme abortion ban is wildly unpopular, and Ohio voters support a woman's right to make medical decisions for themselves without interference from politicians," Walters, with the state Democrats, said during a recent press call.Abortion opponents have also attempted to broaden the stakes of the proposal, arguing that it "goes far beyond" codifying the protections of Roe v. Wade to potentially affect parents' rights and "threaten the freedom of conscience of Ohio's healthcare professionals and preclude ... the enactment and enforcement of commonsense, protective laws," as Protect Women Ohio, one of the leading organizations fighting against the amendment, wrote in a memo."The Abortion Amendment victimizes rather than protects and empowers women and girls," they continued.LaRose, a leading voice against the measure, said in a statement to ABC News that "as a father and someone who's always stood up to protect life, I'll be working to defeat this radical measure."。         最抓眼球的是入口处的「零售区」。

B | 一个观众随手在点餐台下了单。几秒钟后，机器人「听到」了指令，头顶的「眼睛」一扫，灵巧手伸进塞满饮料的货架层里，精准抽出目标瓶子，转身送到前台。         工业区更热闹：两台人形机器人在「搭班干活」——一台搬运料筐，另一台从筐里抓取工件精准放到指定位置。         旁边展示直接挑战了机器视觉领域公认的「老大难」——透明物体识别。透明瓶子、塑料袋包装、反光饮料罐混在一起，对传统相机来说几乎是「隐形」的，但这台机器人看它们跟看积木似的，抓得又快又稳。

C |          这是2026年7月的上海，WAIC开到第九年。今年具身智能第一次与智算并列为两大核心赛道，超过200家具身智能企业到场，人形机器人几乎挤满了整个展馆。         但在这片热闹里，有一家公司做的事有点「反常」——它不造机器人。         回过头看刚才的画面：货架取货的是人形机器人，高速抓取的是工业机械臂，线束装配的又是另一种形态。

D |          形态不同，任务不同，物体更是千差万别——布偶、螺母、透明瓶子、柔性线束。         但它们表现出来的能力却是相似的：都看得清环境，都理解任务，都能规划动作、精准执行。         为什么？答案不是梅卡曼德造出了一台「万能机器人」，而是：它造机器人的「眼睛」「大脑」和「手」，然后把这一套装进不同的身体里。就像安卓不造手机，但可以让所有手机都能跑同一套系统。         智能大于形态     机器人不会只有一种身体          全行业都在问：具身智能的终极形态是什么？很多人回答「人形机器人」。

E | 但这个逻辑有问题。         工厂要的是快准稳的机械臂，人形反而笨重；仓库要的是能满场跑的轮式机器人，两条腿不如轮子效率高；便利店要在货架间穿梭的服务机器人，全尺寸人形连转身都费劲；家庭场景更不用说——橱柜下面那点空间，你让一米七的机器人蹲下去够东西？          不同环境天然需要不同的身体，用一种形态解决所有问题，就像用一把螺丝刀拆所有电器——理论上也许行，实际上一定不行。

F |          梅卡曼德创始人邵天兰在达沃斯世界经济论坛上说过：物理AI不指向某一种具体形态，而是一套通用能力。         真正决定机器人有没有产业价值的，不是它像不像人，而是能不能快速适应复杂环境、完成真实任务、用足够低的成本实现部署。         所以梅卡曼德的策略是「一脑多形」：同一套「眼脑手」，装进人形机器人、双臂轮式机器人、半身人形机器人、工业机械臂都行。         身体百花齐放没关系，脑子得是同一颗。         这颗「大脑」到底长什么样          Mech-GPT多模态大模型是整套系统的核心，不只是个听懂人话的聊天模型。         给它一句「把货架第二层左边那个蓝瓶子拿过来」，它得先弄清你要哪个瓶子，再用「眼睛」看清现场，然后把这句话拆成一串动作：移到架前、伸手、抓稳、退出、送到前台。         难的是现场随时在变。瓶子被碰歪、位置偏了、冒出新障碍，大脑就得当场改计划。         所以这颗「脑」至少干四件事：理解任务、感知环境、规划行动、监督完成——把「想做什么」「现场什么样」「下一步怎么走」「做完没有」串成一条闭环。

G |          这就是物理智能和文本智能最大的不同：答题，打完字就结束；干活，只有瓶子真拿到、工件真放准、插头真插进，才算完成。         比如，现场Mech-GPT多模态大模型的人机交互单元演示。观众走到屏幕前，用自然语言跟机器人「聊天」——你可以随口说「把积木按顺序排好」，它听得懂、想得明白、做得到。不是固定的语音指令识别，而是真正理解你在说什么、你想干什么。         播放          下一个          打开循环播放          00:00     /     00:00          倍速          3.0X     2.0X     1.5X     1.25X     1.0X     0.75X     0.5X          多音轨          AirPlay          0          静音播放中，点击 恢复音量          画中画          网页全屏          全屏          你可以 刷新 试试          视频信息     1.33.6          播放信息 上传日志 调试信息 [X]          视频ID     VID     -          播放流水     Flowid     -          播放内核     Kernel     -          显示器信息     Res     -          帧数     -          缓冲健康度     -          网络活动     net     -          视频分辨率     -          编码     Codec     -          mystery     mystery     -          按住画面移动小窗          再说「手」。光知道该做什么还不够，还得做到。世界动作模型配合新一代Mech-Hand灵巧手，把高层指令翻译成抓、捻、拿、握、敲、插接等精细动作。WAIC现场亚毫米级线束装配，就是「手」的极致体现。

H |          最后说「眼」。高精度3D视觉加上透明物体成像算法，让机器人在复杂光线、杂乱料框、狭窄货架中精准识别物体的位置、形状和材质。         传统算法面前一个玻璃杯和「什么都没有」差不多，但梅卡曼德的算法能把透明瓶子看得清清楚楚。         一句话：眼睛看清世界，大脑理解任务，手把想法变成动作。三者合在一起，不是某台机器人的专属技能包，而是一套通用智能底座。         通用性和泛化性不是口号，要靠难题来证明          「通用」和「泛化」在AI行业快被说烂了。

I | 梅卡曼德在WAIC上的回答是：你说「通用」和「泛化」，我用场景验证。         跨形态——同一套智能进入人形机器人、双臂轮式机器人、工业机械臂，形态不同，都能干活。         跨物体——不只认识规则工业零件，还认识几百种日化、食品、玩具、文具，连透明的、柔软的、反光的都能识别。

J |          在WAIC现场的分拣单元，面前堆了几百种乱七八糟的物体——日化用品、文具、水果、蔬菜、玩具、零食——十几种类别，机器人一件件拿起来，扫一眼，准确丢进对应的分类框。         跨环境——深筐、货架、流水线、狭窄空间都能应对。         跨任务——搬运、分拣、装配、取货、协同作业，不需要每换任务就从头开发。         但更关键的是硬指标。工业客户不关心动作帅不帅，就看核心指标：速度、成功率、精度等。

K | 高速抓取单件小于2.4秒，线束装配亚毫米级，双机协同全程无人干预——这说明梅卡曼德展示的不是实验室Demo，而是接近真实生产要求的系统。         比如，现场中一只机械臂在杂乱料框里高速翻找，每抓一个五角螺母不到2.4秒。你还没看清怎么定位的，零件已经稳稳落在传送带上。         真正的护城河：从1套到27000套          具身智能有一个残酷真相：让机器人成功表演一次不难，难的是换工厂、换国家、换物体后，还能快速上线稳定运行。难的不是Demo，是部署。         邵天兰说过一句很实在的话：问题不在硬件。速度、精度、负载、成本，今天的机器人都够用了。真正限制规模化的，是部署难度。         真实工业环境里，一套系统要稳定跑起来，往往得机械、电气、方案、软件、AI一大帮工程师折腾几周甚至几个月。这种「手工作坊」式交付没法复制，成本摊不下来，最终「算不过账」。         梅卡曼德的解法是用标准化「眼脑手」组件，打通「感知—决策—规划—执行—数据反馈」的完整闭环，尽量减少人工调试。         而部署量够大后，一个更强的机制开始启动——数据飞轮。

L | 机器人在汽车、物流、锂电、3C、家电的真实产线上，每天都在遇到新物体、新光线、新任务：          大规模应用产生真实数据→数据提升模型泛化性→模型降低部署难度→更快进入更多行业。         飞轮一旦转起来，后来者追赶的难度是几何级增长的——你追的是一个越跑越快的对手。

M |          而这套能力的规模，早已不只在国内。截至目前，梅卡曼德全球累计部署超过27000套产品。         能在完全不同的工厂标准、机器人品牌、集成商体系里都稳定落地，靠的不是把货卖出去，而是十年、27000套一个个磨出来的兼容性。         27000套，不是27000次演示——每一套都在真实产线上跑着。         全球化不是卖货     是一种综合能力          「全球化」不只是把产品卖到海外。真正难的是：能不能适应不同国家的工厂标准？兼容当地主流机器人品牌？跟当地集成商体系打通？提供及时的本地服务？这考验的是研发、标准化、交付和服务的综合能力。         梅卡曼德覆盖近50个国家和地区，服务100多家《财富》世界500强客户，国内细分领域连续6年市占率第一。它在美、日、韩、德都设有团队，能适配市面上几十个品牌、上千个型号的机器人，欧美日韩主流市场市占率领先。         这是用十年时间、27000套部署一个个磨出来的兼容性。         正如邵天兰所说——出海不仅仅是一项任务，更是一种能力。能在全球不同市场做到标准化交付和规模化部署，本身就是技术成熟度和产品化水平的证明。

N |          物理AI不是一场表演赛，而是一场部署能力、数据积累和产品标准化的马拉松。在这条跑道上，27000套已经跑完的路程，就是最好的简历。         编辑：大卫 所罗门。

Current article:http://6bxewzj.menxiaomoriniangbai.bond/list_brezgm/qul.html

Published on:05:52:04