内容来源：https://aibusiness.com/robotic-process-automation/humanoid-hard-sell-building-robots

内容总结：

【导语】随着人形机器人走出实验室、进入实际部署阶段，业内开始出现审慎声音：对其能力的热情可能正在超越现实。英国机器人公司Q5D首席执行官斯蒂芬·本宁顿在接受采访时指出，在工业部署中，制造商不应默认“复制人类工人”是自动化任务的最佳方式。他建议从工艺流程出发而非从机器人出发——当任务不再受人类环境束缚时，专用机器可以更快、更精准、成本也显著更低。

【正文】

人形机器人：进步与炒作并存

当前人形机器人领域热度极高。本宁顿表示，技术进步令人振奋，人形机器人能力不断提升，已开始在工业环境中试点，而非仅停留在演示阶段。但他也坦言，在许多应用场景中，炒作仍走在现实前面。

兴趣背后有其逻辑：工厂、仓库和工具都是围绕人设计的，若能造出在人类环境中活动、使用相同设备的机器人，就无需围绕机器重新设计工作场所。但本宁顿提醒，不应假设模仿人类操作员总是自动化任务的最佳路径。他以洗碗为例：可以造一台高度复杂的人形机器人站在水槽前模仿人的每个动作，也可以造一台不受人体形态限制的洗碗机，以最高效、对家庭干扰最小的方式完成任务。

制造业同理。若人形机器人需在为人设计的环境中穿行并使用现有工具，人形形态合理；但若从零设计制造流程、不受这些限制，专用机器可以更快、更精准、成本大幅降低。制造商应从工艺出发，而非从机器人出发。

现实部署的主要瓶颈

本宁顿指出，一些看似最普通的任务反而最难自动化。例如布线：人类操作员可以拿起柔性电线，几乎本能地操控、穿过复杂形状并完成连接；而对机器人而言，柔性材料每次处理时形态和行为都不同，极具挑战。

与其单纯让机器人更好地模仿人类动作，不如重新思考工艺本身——能否拆分为不同阶段？能否围绕机器擅长的能力设计每个阶段？本宁顿认为，工业机器人领域真正有趣的突破正来自这里。另一大瓶颈是编程：机器人再强，也需要有人告诉它做什么。AI正在加速传统编程，但仍需训练数据。

两条解决路径

应对挑战有两条路径：一是让机器人更具适应性，这正是人形机器人的方向——AI和海量训练数据帮助机器人理解环境、学习不同任务；二是重新设计任务，使其更易自动化。Q5D采取的是后者：围绕机器擅长之事重新设计工艺，从而在每次设计变更时大幅减少手动示教需求。本宁顿认为，这指向机器人领域更广泛的机会——灵活性不必完全来自机器人本身，也可以将专用机械与软件结合，打造高度灵活的制造系统，使产品切换更加便捷。

AI如何改变工业机器人

本宁顿表示，AI将使工业机器人更强、更具适应性，且远不止于人形机器人。从检测到预测性维护，AI进步可提升方方面面。对人形机器人而言，AI尤为重要，因其需应对的情况极为多样，需要大量智能和训练数据。但同样的进步也能提升专用机器：若制造系统已有数字信息明确说明要造什么、怎么造，AI就可应用于定义更清晰的问题。

赋予机器人更大自主权也带来挑战，尤其是安全方面。在工业环境中，需确保机器人在正常运行和突发情况下都表现可预测，与人类协同工作时这一点更为关键。因此，AI会让通用机器人更全能，也会提升专用自动化的能力，两者将并行发展，选择取决于任务——有些场景灵活性最重要，有些则优先考虑速度、精度、安全和成本。

中国经验与西方启示

中国已成为工业机器人领域的重要力量。本宁顿认为，中国尤其清晰地展示了机器人与制造规模之间的关系：庞大的工业基础既创造自动化需求，也为机器人企业提供部署、测试和改进技术的场所，形成强大反馈循环——更多制造带来更大机器人市场，更多机器人应用又提升制造效率。

对西方经济体而言，本宁顿认为存在一种倾向：关注谁拥有最先进的机器人或最惊艳的演示。但更重要的问题是，新技术能否有效进入工厂车间。这本质上是生产力问题，随着人口老龄化和可用劳动力缩减将日益重要。若西方想在相对高工资国家把更多制造带回本土、强化供应链并保持竞争力，就必须大幅提升自动化部署能力。这意味着不能只盯着技术本身：制造商需要信心和资本投入，技术也需提供足够清晰的商业回报，以证明改变既有生产流程的合理性。

展望未来

本宁顿预计，未来机器人产业将比当前人形机器人热潮所呈现的更加多元，不同类型机器人将并行发展，各自适配不同的制造问题和用户需求。最大变化之一将是工业自动化的灵活性。传统自动化适合百万次重复生产同一种产品，但在品种多、批量短、产品不断变化时历来难以证明其合理性。这需要不同的自动化思路：机器快速适应，而非为特定任务专门设计。

未来十年，本宁顿认为讨论重点将从“机器人长什么样”转向“制造系统能交付什么”。更好的软件、AI以及数字设计与物理机器之间更紧密的连接，将使自动化在产品与用户需求变化时更易调整。

【编者注】本采访经过编辑，以求清晰简洁。

中文翻译：

由谷歌云赞助
选择你的首批生成式人工智能用例
要开始使用生成式人工智能，首先应关注那些能够改善人类信息体验的领域。
在人形机器人的宣传声势盖过现实之际，为什么专用机器值得更多关注，以及西方如何将自动化引入工厂车间。
在人形机器人的宣传声势盖过现实之际，为什么专用机器值得更多关注，以及西方如何将自动化引入工厂车间。
随着人形机器人走出实验室、进入现实世界部署，一些人警告说，对其能力的热情可能正在超越现实。
英国机器人公司Q5D首席执行官斯蒂芬·本宁顿表示，在工业部署中，制造商不应假定模仿人类工人是自动化一项工作的最佳方式。他的建议是从工艺流程出发，而不是从机器人出发：当一项任务不与人类建造的环境绑定时，一台专用机器可以更快、更精准，而且成本显著更低。
在这篇问答中，本宁顿讨论了人形机器人技术目前所处的位置、人工智能正在如何改变工业自动化，以及西方制造商可以从中国学到什么。
目前人们对人形机器人抱有极大的热情。这项技术真正处于什么阶段，宣传又在哪些地方跑在了现实前面？
斯蒂芬·本宁顿：进展令人振奋。人形机器人正变得越来越有能力，我们也开始看到它们在工业环境中的试验，而不仅仅是令人印象深刻的演示。但在许多应用中，我确实认为炒作仍然领先于现实。
当然，这种关注有一个合乎逻辑的原因。我们的工厂、仓库和工具都是围绕人来设计的。所以，如果你能造出一台可以在这些环境中移动、并与相同设备交互的机器人，你就有可能避免围绕机器重新设计工作场所。
然而，我会稍微更为谨慎的地方在于这样一种假设：模仿人类操作员始终是自动化一项工作的最佳方式。
拿洗碗这样平凡的事情来说。你可以造出一台非常精密的人形机器人，站在水槽边模仿人做的每一个动作。或者，你也可以造出一台洗碗机，它不受人类形态的限制，因此专注于确保以最高效的方式完成任务，同时尽量减少对家庭生活的干扰。
同样的原则也适用于制造业。如果一台人形机器人需要利用现有工具穿过为人类建造的环境，那么采用人形形态就是合理的。但如果你是从零开始设计一个制造流程，或者不受这些限制约束，那么解决方案就可以是一台更快、更精准且成本显著更低的专用机器。
制造商不应从机器人出发，而应从工艺流程出发——仔细审视他们试图自动化的是什么，以及实现它的最有效方式是什么。
在现实应用中部署这些机器的主要瓶颈是什么？
本宁顿：有些最难自动化的任务，实际上恰恰是人来做时看起来非常普通的任务。
布线就是一个很好的例子。人类操作员可以拿起一根柔性电线，对其进行操控，让它沿着复杂形状走线，并几乎凭本能完成连接。对机器人来说，柔性材料很难处理，因为每次操作时它们的移动方式和表现都不同。
与其只是试图让机器人更好地复制这些人类动作，你可以重新思考工艺流程本身。你能把它拆分成不同阶段吗？你能围绕机器真正擅长的东西来设计每个阶段吗？我认为，工业机器人领域一些真正有趣的进步将来自这里。
另一个主要瓶颈是编程。你可以拥有一台极其强大的机器人，但仍然得有人告诉它该做什么。人工智能正在通过加速传统编程来改变这一过程，但首先你仍然需要训练数据来训练机器。
有没有什么特别的方法来应对这些挑战？
本宁顿：解决这个问题实际上有两种方式。一种是让机器人更具适应性。这就是我们在人形机器人身上看到的情况，人工智能和大量训练数据正在帮助机器人理解其环境并学习不同任务。
另一种是重新设计任务，使其更容易自动化。这就是我们在Q5D采取的方法：我们围绕机器擅长的事情重新设计流程，从而在很大程度上消除了每次设计变化时都需要手动教机器人做什么的需要。
我认为这指向了机器人技术中一个更广泛的机会。灵活性不必完全来自机器人本身。你也可以通过将专用机械与软件结合起来，创建一套高度灵活的制造系统，使从一种产品切换到另一种产品变得容易得多。
人工智能如何改变了工业机器人的能力？
本宁顿：人工智能将使工业机器人更有能力、更具适应性，而这远不止适用于人形机器人。人工智能的进步可以改善从检测到预测性维护的一切。
对于人形机器人来说，人工智能尤其重要，因为它们需要处理的情况种类极其繁多。如果你想让一台机器人执行许多不同工作，它就必须解读自己所看到的东西，理解周围环境，并想出如何回应。这需要大量智能和大量训练数据。
但同样的进步也可以改善专用机器。如果一套制造系统已经有数字信息来规定需要制造什么以及如何制造，那么人工智能就可以应用于一个定义得更清晰得多的问题。
赋予机器人更大自主性也会带来挑战，尤其是在安全方面。在工业环境中，你需要确信机器人不仅能在正常运行期间表现可预测，在发生意外情况时也能如此。当机器人与人类并肩工作时，这一点就更为重要。
所以，人工智能会让通用机器人更加多才多艺，但它也会提升专用自动化的能力。两者将并肩发展，正确选择取决于任务。在某些情况下，多才多艺最重要；在其他情况下，制造商则会优先考虑速度、精度、安全和成本。
中国已成为工业机器人领域的一支重要力量。是什么在推动这一点，西方制造商又该从中学习什么？
本宁顿：中国尤其清楚地展示了机器人技术与制造规模之间的关系。
中国拥有庞大的工业基础，这既创造了自动化需求，也为机器人公司提供了部署、测试和改进其技术的地方。这形成了一个强大的反馈循环：更多制造为机器人创造了更大市场，而机器人更广泛的应用又帮助制造业变得更具生产力。
对西方经济体来说，我认为有一种诱惑，就是把注意力集中在该谁拥有最先进的机器人或最令人印象深刻的演示上。更重要的问题是，新技术究竟能以多高的效率真正进入工厂车间。
归根结底，这是一个生产力问题，而且随着人口老龄化和可用劳动力萎缩，这个问题将变得越来越重要。如果西方经济体想把更多制造业带回本土、强化供应链，同时在工资相对较高的国家保持竞争力，我们就必须大幅提升部署自动化的能力。
这意味着要超越技术本身去思考。制造商需要信心和资本去投资，但技术也必须提供足够清晰的商业回报，才能证明改变既有生产流程是合理的。
展望未来，随着用户需求和技术进步，你预计机器人产业将如何演变？
本宁顿：我怀疑，未来会比当前围绕人形机器人的热情有时所暗示的更加多样。我们很可能会看到不同类型的机器人技术并肩发展，各自适合不同的制造问题和不断演变的用户需求。
可以说，最大的变化之一将是工业自动化的灵活性会达到何种程度。
传统自动化在你想要制造数百万次相同产品时非常出色，但从历史上看，当你有很多变体、更短的生产批次或不断变化的产品时，它就更难被证明是合理的。这需要一种不同的自动化方式，在这种方式下，机器能够快速适应，而不是为某一特定任务而专门设计。
在未来十年里，我认为讨论的重点将不再是机器人看起来像什么，而更多是制造系统实际上能够交付什么。更好的软件、人工智能，以及数字设计与实体机器之间更紧密的联系，将使自动化在产品 and 用户需求变化时更容易调整。
编者注：本采访经过编辑，以提高清晰度和简洁性。

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
Where humanoid publicity outpaces reality, why purpose-built machines deserve more attention and how the West can get automation onto the factory floor.
Where humanoid publicity outpaces reality, why purpose-built machines deserve more attention and how the West can get automation onto the factory floor.
As humanoids move out of the lab and into real-world deployment, some caution that excitement about their capabilities may be outpacing reality.
Stephen Bennington, CEO of British robotics company Q5D, says that with industry deployments, manufacturers shouldn't assume copying the human worker is the best way to automate a job. His advice is to start with the process rather than the robot: when a task isn't tied to human-built environments, a purpose-built machine can be faster, more accurate and significantly cheaper.
In this Q&A, Bennington discusses where humanoid robotics stands, how AI is changing industrial automation and what Western manufacturers can learn from China.
There is a huge amount of excitement about humanoid robots at the moment. Where does the technology genuinely stand, and where is the publicity getting ahead of reality?
Stephen Bennington: The progress is exciting. Humanoid robots are becoming more capable, and we are starting to see trials in industrial environments rather than just impressive demos. But for many applications, I do think the hype is still ahead of reality.
There is, of course, a logical reason for the interest. Our factories, warehouses and tools were designed around people. So, if you can build a robot that can move through those environments and interact with the same equipment, you potentially avoid having to redesign the workplace around the machine.
However, where I would be slightly more cautious is the assumption that copying the human operator is always the best way to automate a job.
Take something as mundane as washing dishes. You could build a very sophisticated humanoid robot that stands at a sink and copies every movement a person makes. Or you can build a dishwasher that isn't constrained by the human form and is therefore focused on ensuring the task is completed as efficiently as possible with minimal disruption to the household.
The same principle applies in manufacturing. If a humanoid is required to move through an environment built for humans using existing tools, then being in a human form makes sense. But if you're designing a manufacturing process from scratch or aren't constrained by those limitations, the solution can be a purpose-built machine that is faster, more accurate and significantly cheaper.
Rather than starting with the robot, manufacturers should start with the process -- looking closely at what they are trying to automate and the most effective way of doing it.
What are the main bottlenecks to deploying these machines in real-world applications?
Bennington: Some of the hardest tasks to automate are actually the ones that look very ordinary when a person does them.
Wiring is a good example. A human operator can pick up a flexible wire, manipulate it, route it through a complicated shape and make connections almost instinctively. For a robot, flexible materials are difficult because they move and behave differently every time you handle them.
Rather than simply trying to make a robot better at copying those human movements, you can rethink the process itself. Can you break it into different stages? Can you design each stage around what machines are actually good at? That is where I think some of the really interesting advances in industrial robotics will come from.
Another major bottleneck is programming. You can have an extremely capable robot, but someone still has to tell it what to do. AI is changing how that happens by speeding up traditional programming, but you still need training data to train the machine in the first place.
Are there any particular approaches to address these challenges?
Bennington: There are really two ways of tackling the problem. One is to make the robot more adaptable. That is what we're seeing with humanoids, where AI and large amounts of training data are helping robots understand their environment and learn different tasks.
The other is to redesign the task to make it easier to automate. That's the approach we have taken at Q5D, where we redesign processes around what a machine can do well, thus removing much of the need to manually teach the robot what to do every time the design changes.
I think that points to a broader opportunity in robotics. Flexibility does not have to come entirely from the robot itself. You can also create a highly flexible manufacturing system by combining purpose-built machinery with software that makes it much easier to move from one product to another.
How has AI changed industrial robots’ capabilities?
Bennington: AI is going to make industrial robots more capable and adaptable, and that applies to far more than just humanoids. Advances in AI can improve everything from inspection to predictive maintenance.
For humanoids, AI is particularly important because of the sheer variety of situations they need to deal with. If you want a robot to perform lots of different jobs, it has to interpret what it sees, understand its surroundings and work out how to respond. That requires a lot of intelligence and a lot of training data.
But those same advances can also improve specialist machines. If a manufacturing system already has digital information specifying what needs to be made and how, AI can be applied to a much more clearly defined problem.
There are also challenges that come with giving robots greater autonomy, particularly around safety. In an industrial environment, you need to be confident that a robot will behave predictably not only during normal operation, but also when something unexpected happens. That becomes even more important when robots are working alongside people.
So, AI will make general-purpose robots more versatile, but it will also raise the capabilities of purpose-built automation. The two will develop alongside each other, with the right choice depending on the task. In some situations, versatility will matter most; in others, manufacturers will prioritize speed, precision, safety and cost.
China has become a major force in industrial robotics. What is driving that, and what should manufacturers in the West learn from it?
Bennington: What China demonstrates particularly well is the relationship between robotics and manufacturing scale.
It has an enormous industrial base, which creates demand for automation but also provides robotics companies with places to deploy, test and improve their technology. That creates a powerful feedback loop, as more manufacturing creates a bigger market for robotics, while greater adoption of robotics helps make that manufacturing more productive.
For Western economies, I think there is a temptation to focus on who has the most sophisticated robot or the most impressive demo. The more important question is how effectively new technology actually makes its way onto the factory floor.
Ultimately, this is a productivity question, and one that will become increasingly important as populations age and the available workforce shrinks. If Western economies want to bring more manufacturing closer to home and strengthen supply chains while remaining competitive in relatively high-wage countries, we have to become much better at deploying automation.
That means thinking beyond the technology itself. Manufacturers need the confidence and capital to invest, but the technology also needs to offer a clear enough commercial return to justify changing established production processes.
Looking ahead, how do you expect the robotics industry to evolve as user demands and technology advance?
Bennington: I suspect the future will be much more varied than the current excitement around humanoids sometimes suggests. We are likely to see different types of robotics developing alongside one another, each suited to different manufacturing problems and evolving user demands.
One of the biggest changes will arguably be in how flexible industrial automation becomes.
Traditional automation is brilliant when you want to make the same thing millions of times, but historically it has been harder to justify when you have lots of variants, shorter production runs or products that are constantly changing. That requires a different approach to automation, in which machines adapt quickly rather than being engineered for a specific task.
Over the next decade, I think the conversation will become less focused on what the robot looks like and more focused on what the manufacturing system can actually deliver. Better software, AI and closer links between digital design and the physical machine will make automation much easier to adapt as products and user requirements change.
Editor’s note: This interview was edited for clarity and conciseness.