发那科机器人变量解释
Over the years, as a child who built solar-powered boats and robot arms from science kits, then as a team captain of my high school FIRST Robotics competition, as an undergraduate majoring in a Mechanical Engineering with Robotics flex degree, as an MIT Media Lab research assistant on a big (seriously, big) robotic construction platform, and finally working on DARPA robotics research and robotic applications in industry….
多年来，作为一个孩子，他用科学工具包制造了太阳能动力船和机器人手臂，然后作为我的高中FIRST机器人比赛的队长，作为机器人挠性学位的机械工程专业的本科生，麻省理工学院的媒体在大型(非常大型)的机器人构建平台上担任实验室研究助理，并最终从事DARPA机器人研究和工业领域中的机器人应用…。
I have received many, many questions about robots.
我收到了很多有关机器人的问题。
Hopefully, this litany has outlined why I am qualified to explain this topic, the last qualification being, I can also write and explain things pretty well for an engineer, if I do say so myself. Some caveats:
希望这件事能概述我为什么有资格解释这个话题，最后一个资格是，如果我自己说的话，我也可以为工程师写和解释得很好。 一些警告：
- I am young, and relatively early in my career. I don’t have a PhD and I am not the head of a robotics company. But the field is pretty young, too. You might have reason to be suspicious of what some of those more established people might have to say — maybe they have an incentive to put a particular spin on robots, if they run a company that builds them or if robots are their life’s work. I don’t have as much of an incentive; I can still leave this field and pursue a medical device career or something (but I do love robots). 我还年轻，而且职业生涯还比较早。 我没有博士学位，也不是机器人公司的负责人。 但是这个领域也很年轻。 您可能有理由怀疑一些较老的人可能要说些什么-也许他们有动力对机器人进行特别的旋转，如果他们经营一家制造机器人的公司，或者如果机器人是他们一生的工作。 我没有太多的动力； 我仍然可以离开这个领域，从事医疗器械事业或其他工作(但我确实喜欢机器人)。
- This post will contain several Hot Takes (TM). Just roll with me ‘til the end and have fun, I don’t pretend it’s all perfect or that I know everything — that’s why I call them ‘hot takes’ and not ‘immutable truths’. I’d be more than happy to further engage and debate about the definitions and implications of technology! 这篇文章将包含几个热门影片(TM)。 只是跟我“直到最后滚来滚去，玩得开心，我不假装这一切都是完美的，或者我不了解一切。”这就是为什么我称它们为“热门”而不是“不变的真理”。 我很乐意进一步参与和辩论技术的定义和含义！
你为什么写这篇博客文章？ (Why are you writing this blog post?)
Though there are many articles on this subject (and I encourage you to read them) in my opinion, few people have done a good job explaining what robotics and automation means for you. Few people explain the technical stuff, the research stuff, and the implications for business and government and so on. People want to know what’s really going on right now, and most people are going off of viral, creepy Boston Dynamics videos, which look cool but don’t explain much about how robots actually work.
尽管我认为有很多关于该主题的文章(我鼓励您阅读它们)，但是很少有人能很好地解释机器人技术和自动化对您的意义 。 很少有人会解释技术资料，研究资料以及对企业和政府的影响等。 人们想知道到底发生了什么的，现在，病毒，令人毛骨悚然Boston Dynamics公司的视频，这看起来很酷大多数人都会关闭，但并不了解如何机器人实际工作太多解释。
Hence, why people keep asking me questions.
因此，为什么人们不断问我问题。
核心问题：您会让我失业吗？ (The Central Question: Are you gonna put me out of a job?)
All the questions that I’ve ever received seem to revolve around this one. I will address this incredibly common question in two parts. The first is social in nature, the second is technical and practical.
我收到的所有问题似乎都围绕着这个问题。 我将分两部分讨论这个令人难以置信的常见问题。 首先是社会性质，其次是技术和实践。
机器人的社会意义 (Social Implications of Robots)
Let’s look at one very big field in robotics right now: e-commerce.
现在让我们看一下机器人技术中的一个非常大的领域：电子商务。
The bulk of automation in e-commerce and manufacturing is actually motivated by the fact that there are not enough people.
电子商务和制造业中的大量自动化实际上是由于没有足够的人而造成的。
Let me say that again. There are not enough workers, so automation rushes in to fill the gaps.
让我再说一遍。 由于没有足够的工人，因此自动化工作开始填补空白。
Now, I’m not going to just give giant corporations a free pass here. Part of the reason there are not enough people are because wages are low, benefits are meh, and the jobs are very tedious. Growth-track employment opportunities are dwindling in favor of the “independent contractor” model. There’s no promise that a job in order fulfillment will ever be more than just a job. The companies around these positions have a major role to play in making these positions undesirable.
现在，我不只是给大型公司免费通行证。 人数不足的部分原因是因为工资低，福利低，工作繁琐。 支持“独立承包商”模式的增长型就业机会正在减少。 不能保证完成订单的工作将不仅仅是一份工作。 这些职位周围的公司在使这些职位不受欢迎方面可以发挥主要作用。
That said, at the end of the day, even if there is great pay and benefits, running around a warehouse all day is rarely going to be desirable. Autonomous ground vehicles, championed by Kiva Systems (now Amazon Robotics) made this job less tedious and labor intensive, by using robotic vehicles to bring the goods to the person (this is called “goods-to-person”) instead of sending people around the warehouse to find the goods.
即便如此，即使有丰厚的薪水和收益，到一天结束时仍要在仓库中四处奔跑几乎是不可取的。 由Kiva Systems(现为Amazon Robotics)提倡的自主地面车辆，通过使用机器人车辆将货物带给人(这称为“人对人”)，而不是派人到处走动，从而减轻了工作的繁琐程度和劳动强度仓库里找到货物。
My point is more so that the relationship between technology and jobs is complex, and society and government have a major role to play in managing that relationship. Once again, corporations and makers of technology don’t get a free pass, but the ethics are not as straightforward as, say, renewable energy (“good”) or weapons design (“bad”). I think we should generally allow for the development of technology so that people can have the advantage of its benefits. The question of which people capture those benefits is a much bigger question about capitalism, regulation, and government.
我的观点是，技术与工作之间的关系非常复杂，而社会和政府在管理这种关系方面可以发挥重要作用。 再一次，企业和技术制造商没有获得免费通行证，但是道德观念并不像可再生能源(“好”)或武器设计(“坏”)那样简单。 我认为我们通常应该允许技术的发展，以便人们可以从其利益中受益。 哪些人获得这些利益的问题是关于资本主义，监管和政府的更大问题。
There is a world with robots that makes the rich richer and the poor poorer. There is also a world without robots that makes the rich richer and the poor poorer. Depending on the robot, the ‘bot itself might replace workers entirely, in which case the technology pretty directly destroys some jobs, or it might simply make people more efficient (as in the case of goods-to-person). But the bigger societal question is the one I see as more important. It’s not always technology itself that eliminates certain jobs, exactly, but broader trends — what is responsible for the disappearance of telemarketers? I mean, we still have, and use, phones.
有机器人的世界，使富人变得更富，穷人变得更穷。 还有一个没有机器人的世界，使富人更富，穷人更穷。 取决于机器人，“机器人”本身可能会完全替代工人，在这种情况下，该技术会直接破坏一些工作，或者可能只是使人们变得更有效率(如“人与人”的情况)。 但是，更大的社会问题是我认为更重要的问题。 并非总是能完全消除某些工作的技术本身，而是更广泛的趋势-电话推销员消失的原因是什么？ 我的意思是，我们仍然拥有并使用电话。
I offer Hot Take #1: Technology just accelerates existing trends, whatever the trends are. In some cases, the technology is underfunded, or regulated to the point of being impractical. In other cases, money is poured into a certain innovation, red tape is cut away, and it flourishes. How those financial and regulatory forces should approach robots as they improve, I leave an open-ended question — maybe some kind of ‘robot tax’, or something else to motivate giving humans better positions and work environments, which is what we really want.
我提供了热门建议＃1：无论趋势如何，技术都可以加速现有趋势。 在某些情况下，该技术的资金不足或受到监管以致不切实际。 在其他情况下，金钱被投入到某种创新中，繁文tape节被剥夺，然后蓬勃发展。 这些财务和监管力量如何随着机器人的改进而与之接触，我有一个悬而未决的问题-也许是某种“机器人税”，或其他一些激励人们为人类提供更好的位置和工作环境的动力，这正是我们真正想要的。
目前机器人技术处于什么状态？ (What is the state of robotics right now?)
I offer you this video, “Autonomously folding a pile of 5 previously unseen towels”.
我向您提供此视频，“自动折叠一堆5条以前看不见的毛巾”。
This video is at 50x speed…which means it takes the robot something like 20 minutes per towel. Granted, this is 9 years ago and it is pretty impressive that the towels are “never-before-seen”.
该视频的速度是50倍……这意味着每条毛巾需要20分钟左右的时间。 当然，这是9年前，这是相当令人印象深刻的是毛巾“从来没有见过的。”
This, too, is a lovely gem (the things that fancy universities and companies don’t show you…)
这也是一颗可爱的宝石(高档大学和公司没有向您展示的东西……)
This is where robotics is right now — there are a lot of heated debates about AI and ML, but in terms of robots interacting with their physical environment, there is a long, long way to go. (Each of those falls in the second video can be very expensive, by the way).
这就是现在的机器人技术-关于AI和ML的激烈辩论，但是就机器人与其物理环境的交互而言，还有很长的路要走。 (顺便说一句，第二个视频中的每个瀑布都可能非常昂贵)。
Hot Take #2: Every AI company is an ML company.
热点二：每个AI公司都是ML公司。
What’s the difference between AI (Artificial Intelligence) and ML (Machine Learning)? One joking answer is, “if it’s written in [the software language] Python, it’s ML, if it’s written in PowerPoint, it’s AI.”
AI(人工智能)和ML(机器学习)之间有什么区别？ 一个开玩笑的答案是： “如果使用[软件语言] Python编写，则为ML，如果使用PowerPoint编写，则为AI。”
The point of the joke is that the field of AI asks big, broad questions, like what is intelligence? What does it mean to be human and have human intelligence? Is intelligence actually something that’s part of human identity, or is it really empathy? How do people learn and how can we make learning machines? Is intelligence as humans understand it biased — is it a logical fallacy that we think the dolphins are dumber than us? (The dolphins didn’t mess up the earth with climate change, after all) Can we really understand our own intelligence, or would it require something more intelligent than us to understand ourselves? What about definitions — what should we define as intelligence? If we make a definition that is too narrow, your calculator is artificial intelligence; if too broad, it’s hard to classify the ‘progress’ we make toward AI.
开玩笑的意思是，人工智能领域提出了广泛而广泛的问题，例如什么是智力？ 成为人类并拥有人类智慧意味着什么？ 情报实际上是人类身份的一部分吗，还是真的有同理心？ 人们如何学习以及我们如何制造学习机？ 正如人类所理解的那样，智力是否有偏见？我们认为海豚比我们笨，这在逻辑上是谬论吗？ (毕竟，海豚并没有因气候变化而使地球混乱。)我们真的可以理解我们自己的智慧，还是需要比我们更聪明的东西来理解自己？ 定义呢？我们应该定义为智力吗？ 如果我们定义的范围太窄，则您的计算器就是人工智能； 如果范围太广，就很难对我们在人工智能方面的“进步”进行分类。
AI touches on computer science, design, neuroscience, even philosophy in a way. ML, on the other hand, is a specific category of software algorithms; a category of numerical methods and mathematical approaches. These types of algorithms have to look at 50,000 images of chairs to understand what a chair is. A human baby needs to only see about the number of chairs in its household. The baby, as it grows, can identify weird chairs too, sculptural chairs, art museum chairs that the ML algorithm might fail to “see”.
人工智能以某种方式触及计算机科学，设计，神经科学甚至哲学。 另一方面，ML是软件算法的特定类别。 一类数值方法和数学方法。 这些类型的算法必须查看50,000张椅子图片以了解椅子是什么。 一个人类婴儿只需要了解其家庭中的椅子数量即可。 随着婴儿的成长，它也可以识别出ML算法可能无法“看到”的怪异椅子，雕塑椅子，美术馆椅子。
AI asks the question, why is a baby so much better than an algorithm? Maybe we’re being unfair to the algorithm. Maybe the baby developed this understanding through evolution, or something, so really somewhere deep in its DNA is the memory of thousands of chairs seen by its ancestors. Sounds crazy, right? But this is a real theory. For ML companies to call themselves ‘AI’ is, to me, extreme arrogance. We have not actually decided whether ML qualifies as AI or not. In the neighboring field of neuroscience, we still don’t understand how our own brain works.
AI提出了一个问题，为什么婴儿比算法好得多？ 也许我们对算法不公平。 也许婴儿是通过进化或某种方式发展了这种理解的，所以真正在其DNA的深处是其祖先看到的数千把椅子的记忆。 听起来很疯狂吧？ 但这是一个真实的理论。 对于我来说，对于机器学习公司来说，称自己为“ AI”是极端的傲慢。 我们实际上尚未确定ML是否符合AI的条件。 在邻近的神经科学领域，我们仍然不了解自己的大脑如何工作。
And ML is just big data. Yeah yeah, it’s complicated and whatever (I told you there would be a lot of hot takes). But at the end of the day, ML is what you did in high school science or math, when you looked at a plot of data you collected and drew a line of best fit.
ML只是大数据。 是的，这很复杂，无论如何(我告诉过您，会有很多抢手货)。 但是，归根结底，ML是您在高中科学或数学中所做的，当您查看收集的数据并绘制出最合适的线时。
A quick refresher for some of you — this idea is that we can plot real-life data (which is messy and doesn’t fit on a precise line) and then make a line that we can use to help us make predictions, or make the information easier to understand. For example, data on housing prices with a line that has a very sharp upward slope can tell us that housing prices are increasing very quickly. Similarly, if it’s a curve that is fit to real-life data and changing over time (most of you are familiar with this in the age of COVID-19) as the curve gets flatter or steeper, we know that the situation is changing more or less rapidly and it informs our decisions.
让您快速回顾一下-这个想法是，我们可以绘制真实数据(混乱且不适合精确的直线)，然后绘制一条直线，以帮助我们进行预测或做出信息更容易理解。 例如，关于房屋价格的数据，其上升斜率非常陡峭，可以告诉我们房屋价格上涨非常快。 同样，如果这条曲线适合实际数据并且随着时间的推移而变化(在COVID-19时代，您中的大多数人都对此有所了解)，则该曲线会变得越来越平坦或更陡峭，我们知道情况的变化会越来越多或速度较慢，它可以告知我们的决定。
That’s ML.
那是ML。
Instead of a line or curve, we have even fancier graphs now. We even have weird multi-dimensional data representations so we can do funky math things with the data. But at its core? That’s all it is. That’s why it needs to see 50,000 chairs, because the thing-of-best-fit needs to be really, really good.
现在，我们有了更高级的图形，而不是直线或曲线。 我们甚至拥有怪异的多维数据表示形式，因此我们可以对数据进行时髦的数学运算。 但是其核心？ 仅此而已。 这就是为什么它需要看到50,000张椅子的原因，因为最合适的东西必须非常非常好。
A huge, important point here is that ML still relies on human intelligence!!! How do you think the algorithm gets trained on 50,000 labeled images of chairs? Human beings label those images so that during the “training” period, the ML can say “I think this is a chair” and then check the label of the image to determine the truth. The proliferation of ML companies has created work for human beings to label training data, work that is often low-paying and outsourced to the global south.
此处最重要的一点是， ML仍然依赖于人类的智力！！！ 您如何看待该算法在50,000张带标签的椅子图像上进行训练？ 人类对这些图像进行标记，以便在“训练”期间，ML可以说“我认为这是椅子”，然后检查图像的标签以确定真相。 ML公司 的激增 为人类创造了工作，以标记培训数据，这些工作通常是低薪的并且外包给全球南方。
Yes, after being ‘trained’ the algorithm can run off and perform independent, new tasks impressively, to the point of being called AI, but the core of any ML algorithm is the data it was trained on.
是的，在经过“训练”之后，该算法可以运行并执行令人印象深刻的独立的新任务，甚至被称为AI，但是任何ML算法的核心都是对其进行训练的数据。
This data can be biased, and that bias has severe implications, as pointed out in ongoing discussions by Data for Black Lives and the Algorithmic Justice League around facial recognition technology. To put it plainly, if your training data is 50,000 images of white men, your ML facial recognition algorithm will not be able to accurately recognize the faces of women and people of color.
正如“ Black Lives数据”和“算法正义联盟”围绕面部识别技术的持续讨论所指出的那样，此数据可能会产生偏差，并且这种偏差会产生严重的影响。 简而言之，如果您的训练数据是50,000张白人男性图像，则您的ML面部识别算法将无法准确识别女性和有色人种的面Kong。
To be clear, ML is still very powerful. It’s hard to process tons of data, and we had to come up with fancier math and more powerful computers, and there’s a reason this technology has reached its zenith only now. ML still keeps me up at night — but my nightmares are probably different than the Silicon Valley VC’s and startup engineers. The use of ML in facial recognition, determining credit scores, and other applications has been shown to be inaccurate at best and racist/discriminatory at worst. These issues— issues rooted in the people, the designers of the algorithms and tools, the biased data which algorithms are trained on — are what bothers me, not the idea that ML will someday get so good it becomes self-aware and takes over humans.
需要明确的是，ML仍然非常强大。 处理大量数据非常困难，我们不得不提出更先进的数学方法和功能更强大的计算机，并且这种技术只有到现在才达到顶峰是有原因的。 ML仍然让我彻夜难眠-但是我的噩梦可能与硅谷风投公司和初创工程师不同。 事实证明，在面部识别，确定信用评分和其他应用中使用ML最好是不准确的，而最坏情况是种族主义/歧视性的。 这些问题-植根于人的问题，算法和工具的设计人员 ，算法受过训练的偏见数据 -困扰着我，而不是ML有一天会变得如此好而变得自我意识并接管人类的想法。
如果智力也有些物理性怎么办？ (What if intelligence is also somewhat physical?)
All of this said, ML is still pretty powerful which is why some (ok, most) choose to call it a form of AI. Algorithms that can perform high-level analysis are indeed impressive. Look at this crazy, incredible shit:
综上所述，ML仍然非常强大，这就是为什么某些(好的，大多数)选择将其称为AI的原因。 可以执行高级分析的算法确实令人印象深刻。 看看这个疯狂的，令人难以置信的狗屎：
Let’s go back to what we deem as the best example of intelligence, a human being. We have many sensors to interact with our world, and we gain and compile information at incredibly fast speeds. On the software side, there’s the question of how our brains work to synthesize all this information. But doesn’t hardware matter too? How can we understand anything if we cannot see and interact with the world right in front of us?
让我们回到我们认为是人类最好的智能例子。 我们有许多传感器可以与我们的世界互动，并且我们以惊人的速度获取和编译信息。 在软件方面，存在一个问题，即我们的大脑如何工作以综合所有这些信息。 但是硬件也不重要吗？ 如果我们看不到眼前的世界并与之互动，我们如何理解任何东西？
I could go on and on about this subject, since this is the area of robotics that I am most excited about and have positioned myself in. A grad student I worked with at the MIT Media Lab once said, “who has better respect for nature — the roboticist or the biologist? I think maybe the roboticist.”
我可以继续讲这个主题，因为这是我最兴奋的领域，并且让我自己定位在机器人领域。我在MIT媒体实验室工作的一名研究生曾说过：“谁对自然有更好的尊重-机器人专家还是生物学家？ 我想也许是机器人专家。”
You don’t appreciate how efficient nature is until you start trying to copy it. Professor Hugh Herr, also at the Media Lab, is known for talking about how quiet muscles are — the motors in his prostheses whirr and click constantly, unlike the silent movement of mammalian muscles. Before we understand something as complex as the brain, we don’t even fully understand how muscles work.
在开始尝试复制自然资源之前，您不会意识到它的效率如何。 同样在媒体实验室工作的休·赫尔 ( Hugh Herr)教授以谈论安静的肌肉而著称，而假肢中的马达不断旋转并发出喀嗒声，这与哺乳动物肌肉的无声运动不同。 在我们了解像大脑这样复杂的事物之前， 我们甚至还没有完全了解肌肉的工作原理 。
Robots do not understand the ‘rules’ of the real world, and they sure as hell don’t have sensors as sensitive as dog’s noses or human eyesight. Robots do not understand physics. Robots do not understand that to pick up a cup, you should grab it around the center rather than an impossible point on the edge of the rim. You have to teach that to the robot, and someone’s PhD thesis at MIT was dedicated to just that problem (this field is called ‘manipulation’ and that problem is ‘grasping’) . The sensors of the robot, the physical structure of the robot — does that not also at least influence its intelligence? It is very hard to compete with nature on that front. There are also theories about human learning being embodied in way — your intelligence is not just in your brain, it’s also in the thousands of nerves all around your body, and some of our learning is impossible without our physical bodies. There are many of these expanded views of intelligence that I find so interesting.
机器人不了解现实世界的“规则”，他们确信地狱中没有像狗的鼻子或人眼一样敏感的传感器 。 机器人不懂物理。 机器人不知道要捡起杯子，应该抓住杯子的中心，而不是边缘的不可能的点。 您必须将其教给机器人，然后某人在麻省理工学院的博士学位论文专门针对该问题(该领域称为“操纵”，而该问题正在“抓紧”)。 机器人的传感器，机器人的物理结构–是否也至少不会影响其智能？ 在这方面很难与自然竞争。 还有一些关于人类学习是如何体现的理论-您的智力不仅在大脑中，而且还存在于您身体周围的成千上万条神经中，如果没有身体，我们的某些学习是不可能的。 我发现其中许多扩展的智能观点非常有趣。
This is not a declaration that we will never get to true ‘artificial intelligence’, but it is a declaration that right now, we still have a long, long way to go, and there are so many uncertainties along the way. For those worried that reaching true AI would lead to some kind of malevolent AI, well, malevolence is also one of those uncertainties.
这并不是说我们永远不会实现真正的“人工智能”，而是说现在，我们还有很长的路要走，而且还有很多不确定性。 对于那些担心达到真正的AI会导致某种恶意的AI的人来说，恶意也是这些不确定因素之一。
I feel that some of the hot shots in this field, especially in industry and Silicon Valley, don’t quite understand that. Here’s Hot Take #3: I feel we should focus more on the implications of these new and powerful technologies that are very real right now, like privacy and facial recognition as I mentioned, rather than worrying about an abstract malevolent AI future. Maybe we should even worry about how “Internet of Things” could lead to a Die Hard III-type scenario.
我觉得这个领域的一些热点，特别是在工业和硅谷，并不太了解。 这是热点问题3：我认为我们应该更多地关注这些非常实用的新技术的含义，例如我提到的隐私和面部识别，而不是担心抽象的恶意AI未来。 也许我们甚至应该担心“物联网”如何导致Die Hard III型场景。
But a robot takeover? I see this as somewhat similar to worrying about the day the sun explodes, when the threat of climate change is right in front of us. It’s humans using powerful technology for bad things that concerns me first, not the technology becoming independent.
但是机器人接管吗？ 我认为这有点像担心太阳爆炸的日子，那时气候变化的威胁就在我们眼前。 人类首先使用强大的技术来处理坏事，而不是让技术变得独立。
附录 (Appendix)
I had a lot of fun writing this and I hope you had fun reading it! Here are some fun articles, comics, and resources:
我写这篇文章很有趣，希望您阅读愉快！ 以下是一些有趣的文章，漫画和资源：
AI is Learning from Humans
人工智能正在向人类学习
Data for Black Lives
黑生命的数据
“Forget about AI, Extended Intelligence is the Future” by Joi Ito*
Joi Ito的“忘记人工智能，扩展智能就是未来” *
“What if: Robot Takeover” on xkcd.com
xkcd.com上的“如果：机器人接管”
AI is Learning from Humans from NY Times
人工智能正在向纽约时报的人类学习
*It is my belief we can appreciate non-problematic works by problematic individuals (to an extent) as long as we acknowledge the problems and participate in criticism. While I appreciate this particular article, I recognize Joi Ito’s complicity in facilitating donations by Jeffrey Epstein to MIT, and his association to Epstein when formerly at MIT.
*我相信，只要我们承认问题并参与批评，我们就可以(一定程度上)欣赏问题个人的非问题性作品。 当我欣赏这篇特别的文章时，我认识到Joi Ito在协助Jeffrey Epstein向MIT捐款以及他以前在MIT时对Epstein的协助中的同谋。
翻译自: https://medium.com/hardware-is-hard/robots-explained-4c489e3b344f
发那科机器人变量解释