Why Do Humanoid Robots Still Struggle With the Small Stuff?
Companies are developing and now promoting a future full of humanoid robots.
公司正在开发并正在推广充满人形机器人的未来。
T he last time I covered the science of humanoid robots, the state of the art looked downright Orwellian — by which I mean, “ four legs good, two legs bad .” It was 2015.
上次我介绍人形机器人科学时，最先进的技术看起来完全是奥威尔式的——我的意思是，“四条腿好，两条腿坏”。那是2015年。
Boston Dynamics’ first “Spot” quadruped had taken YouTube by storm , confidently trotting up stairs and recovering from vicious kicks.
波士顿动力公司的第一款“Spot”四足机器人在 YouTube 上掀起了一场风暴，它自信地小跑上楼梯，并从恶毒的踢击中恢复过来。
Also popular at the time: humanoids falling down.
当时还流行：人形摔倒。
Constantly.
不断地。
I felt sorrier for those tottering metal lobsters than I ever did for Spot.
我对那些摇摇欲坠的金属龙虾比对斑点感到更难过。
Bipedal locomotion is hard .
双足行走很困难。
Cut to now.
切到现在。
Humanoids have apparently become so advanced that Tesla is mothballing some electric car models to make way for its Optimus humanoid robot, and start-ups are preselling android butlers with a straight face.
人形机器人显然已经变得如此先进，以至于特斯拉正在封存一些电动汽车模型，为其擎天柱人形机器人让路，而初创公司则在预售一脸严肃的机器人管家。
Hype aside, I was genuinely curious: Did a paradigm shift happen in the field when I wasn’t looking?
撇开炒作不谈，我真的很好奇：当我不注意的时候，这个领域是否发生了范式转变？
Sure, “AI” happened (that is, in the post-ChatGPT sense).
当然，“人工智能”发生了（即后 ChatGPT 意义上的）。
I certainly hadn’t overlooked that.
我当然没有忽视这一点。
But I had no idea what it possibly had to do with robots not falling down anymore.
但我不知道这与机器人不再摔倒有什么关系。
AI breakthroughs have made humanoid robots more capable than ever.
人工智能的突破使人形机器人比以往任何时候都更有能力。
But they still struggle with everyday tasks like stairs and doors.
但他们仍然难以完成楼梯和门等日常任务。
For a reality check, I called Scott Kuindersma, who recently left Boston Dynamics after many years there, and Jonathan Hurst of Agility Robotics.
为了核实现实情况，我给斯科特·库德斯玛（Scott Kuindersma）和敏捷机器人公司（Agility Robotics）的乔纳森·赫斯特（Jonathan Hurst）打了电话，后者在波士顿动力公司工作多年后最近离开了。
Both scientists had been present and involved during the robot-faceplant days.
两位科学家在机器人面部植物时代都在场并参与其中。
Surely today’s robotic bipedal marvels can ascend a few stairs and open a door without breaking a nonexistent sweat, something they famously struggled with a decade ago.
当然，今天的双足机器人奇迹可以爬上几级楼梯并打开一扇门，而不会流一滴不存在的汗，这是他们在十年前众所周知的挣扎。
I asked each researcher: Can your flagship robot — Boston Dynamics’ Atlas or Agility’s Digit , two of the most credible and pedigreed humanoids on Earth — handle any set of stairs or doorway?
我问每位研究人员：你们的旗舰机器人——波士顿动力公司的 Atlas 或 Agility 的 Digit，地球上最可靠、最纯正的两种人形机器人——能走任何楼梯或门口吗？
“I don’t think it’s totally solved,” Kuindersma said.
“我认为这个问题还没有完全解决，”库德斯玛说。
Don’t get me wrong: I don’t believe that some sock-faced robot zombie is close to taking over my household chores.
别误会我的意思：我不相信某个长着袜子的机器人僵尸会接管我的家务。
But stairs and doors?
但是楼梯和门呢？
It’s 2026.
现在是 2026 年。
Why are humanoids still this …
为什么人形还是这样……
hard?
难吗？
To be fair, a paradigm shift did happen.
公平地说，范式转变确实发生了。
Three, actually.
实际上是三个。
First, deep learning — neural networks running on fast GPU chips — turbocharged computer vision and reinforcement learning, which radically improved the speed and sophistication with which robots could perceive and interact with their environments.
首先，深度学习（在快速 GPU 芯片上运行的神经网络）增强了计算机视觉和强化学习，从根本上提高了机器人感知环境并与其交互的速度和复杂程度。
Then in 2016, a revolution in actuation (roboticist-speak for “making parts move”) began: Heavy hydraulic mechanisms were replaced by smaller, “proprioceptive” electric motors that gave legged robots animal-like nimbleness.
然后在 2016 年，一场驱动革命（机器人学家所说的“让零件移动”）开始了：重型液压机构被更小的“本体感受”电动机取代，赋予腿式机器人动物般的灵活性。
Most recently came the large language models.
最近出现了大型语言模型。
Adapting chatbot technology for robots, it turns out, lets them autonomously plan and perform multistep tasks, such as coring an apple or emptying a dishwasher (in demos, at least).
事实证明，为机器人采用聊天机器人技术可以让它们自主计划和执行多步骤任务，例如给苹果去核或清空洗碗机（至少在演示中）。
I n philosophy, “qualia” refers to the subjective qualities of our experience: what it’s like for Alice to see blue or for Bob to feel delighted.
在哲学中，“感受性”指的是我们体验的主观品质：爱丽丝看到蓝色或鲍勃感到高兴是什么感觉。
Qualia are “the ways things seem to us,” as the late philosopher Daniel Dennett put it.
正如已故哲学家丹尼尔·丹尼特所说，感受性是“事物在我们看来的方式”。
In these essays, our columnists follow their curiosity, and explore important but not necessarily answerable scientific questions.
在这些文章中，我们的专栏作家追随他们的好奇心，探索重要但不一定可以回答的科学问题。
These advances created the night-and-day difference between “ Running Man ,” the hulking, halting version of Atlas that won second place in 2015’s DARPA Robotics Challenge, and the svelte, smooth Atlas recently shown breakdancing and autonomously moving irregular items from one bin to another (while dealing with interference from a hockey stick–wielding human).
这些进步创造了“Running Man”与在 2015 年 DARPA 机器人挑战赛中获得第二名的笨重、停顿版本的 Atlas 和最近展示的苗条、流畅的 Atlas 之间的天壤之别。Atlas 最近展示了霹雳舞并自动将不规则物品从一个垃圾箱移动到另一个垃圾箱（同时应对挥舞曲棍球棒的人类的干扰）。
That fluid gait, for example, comes from deep reinforcement learning.
例如，这种流畅的步态来自深度强化学习。
Roboticists once coordinated each movement with various hand-engineered algorithms, using equations to model the (simplified) physics of the robot.
机器人专家曾经用各种手工设计的算法来协调每个动作，使用方程来模拟机器人的（简化）物理过程。
Now they train neural networks to act as “whole-body controllers” by running countless digital simulations of the humanoid.
现在，他们通过运行无数的人形数字模拟来训练神经网络充当“全身控制器”。
This process teaches the network a “policy” for how to translate feedback from its environment into actions.
这个过程教会网络如何将环境反馈转化为行动的“策略”。
“We use reinforcement learning to build a policy that’s handling the body coordination, collision avoidance, balance, all that stuff,” Kuindersma said.
“我们使用强化学习来制定处理身体协调、避免碰撞、平衡等所有问题的策略，”库德斯马说。
There’s no longer any need to model a robot’s leg as a linear inverted pendulum , for example.
例如，不再需要将机器人的腿建模为线性倒立摆。
“That’s just gone by the wayside,” he said.
“那只是半途而废了，”他说。
The Atlas robot from Boston Dynamics shows off in a video from early 2025.
波士顿动力公司的 Atlas 机器人在 2025 年初的视频中进行了展示。
Boston Dynamics/Anadolu Agency via Getty Images
波士顿动力/阿纳多卢通讯社，盖蒂图片社
This strategy was aided by the proprioceptive actuators pioneered by Sangbae Kim of the Massachusetts Institute of Technology in his Cheetah series of robots .
这一策略得到了麻省理工学院 Sangbae Kim 在他的 Cheetah 系列机器人中首创的本体感受执行器的帮助。
“Reinforcement learning has existed for a long time, you know.
“你知道，强化学习已经存在很长时间了。
People tried it before,” Kim said.
人们以前尝试过，”金说。
“But if you use conventional [motors], the robot just breaks” every time it fails to perfectly execute a policy in the real world — or encounters an obstacle or disturbance.
“但如果你使用传统的[电机]，每当机器人无法在现实世界中完美执行策略或遇到障碍或干扰时，它就会崩溃”。
Kim’s actuators got around the problem with controllable “compliance,” or flexible springiness.
Kim 的执行器通过可控的“顺应性”或灵活的弹性解决了这个问题。
Over the past decade, they’ve gotten cheaper and more widely accessible.
在过去的十年里，它们变得更加便宜并且更容易获得。
“Reinforcement learning solved a lot of the [bipedal] locomotion problem, but the hardware was the enabler,” Kim said.
“强化学习解决了很多[双足]运动问题，但硬件才是推动者，”金说。
To have robots which work like humans, I think we have to master physics.
为了拥有像人类一样工作的机器人，我认为我们必须掌握物理学。
If reinforcement learning and compliant actuation were gifts to humanoid robotics, multimodal AI put a bow on it.
如果强化学习和顺从驱动是人形机器人的礼物，那么多模态人工智能则向它致敬。
In 2023, Google DeepMind introduced “vision-language-action” (VLA) models, which can take in video and natural language and produce movement commands as outputs.
2023 年，谷歌 DeepMind 推出了“视觉-语言-动作”（VLA）模型，该模型可以接收视频和自然语言并生成运动命令作为输出。
“If you say ‘I’m thirsty,’ it knows you probably want to drink, and it can [generate] the steps that [the robot] needs to take: Go find a thing, and then pick it up in this way,” said Carolina Parada, head of robotics at Google DeepMind.
“如果你说‘我渴了’，它知道你可能想喝水，并且它可以[生成][机器人]需要采取的步骤：去找一个东西，然后以这种方式拿起它，”谷歌 DeepMind 机器人技术主管 Carolina Parada 说。
“This is something that, before three years ago, you would have to go hard-code.” In a stroke, VLAs united previously disparate approaches to robotic perception, planning, and control into one general-purpose pipeline.
“在三年前，这是必须进行硬编码的事情。” VLA 一举将之前不同的机器人感知、规划和控制方法整合到一个通用管道中。
Robust embodiment, check.
稳健的体现，检查。
Generalizable intelligence, check.
通用情报，检查。
(A start, anyway.) So why don’t they add up to humanoids being scientifically “solved” — at least in principle?
（无论如何，这是一个开始。）那么，为什么它们不能让类人生物得到科学“解决”——至少在原则上如此？
Pulkit Agrawal, who studies robot learning at the appropriately named Improbable AI Lab at MIT, had an answer when I reached him there last month.
Pulkit Agrawal 在麻省理工学院的 Improbable AI 实验室研究机器人学习，当我上个月联系到他时，他得到了答案。
“To have robots which work like humans,” he said, “I think we have to master physics.”
“要拥有像人类一样工作的机器人，”他说，“我认为我们必须掌握物理学。”
He wasn’t referring to cosmic matters like general relativity or quantum gravity, nor to the virtual “ world models ” that currently excite leading AI researchers such as Yann LeCun .
他指的不是广义相对论或量子引力等宇宙问题，也不是目前让 Yann LeCun 等领先人工智能研究人员兴奋的虚拟“世界模型”。
Instead, Agrawal is talking about mastering something a high school science student ought to be familiar with: force and inertia.
相反，阿格拉瓦尔谈论的是掌握高中理科学生应该熟悉的东西：力和惯性。
Press images of the Neo from 1X (left) and Tesla’s Optimus (right) imagine a future of humanoid helpers.
1X 的 Neo（左）和 Tesla 的 Optimus（右）的新闻图片想象了人形助手的未来。
The whole point of the humanoid form factor, after all, is to deliver what Kim calls “multipurpose mobile manipulation,” or the ability to move almost anywhere (including on stairs and through doors) and handle almost anything (from unloading pallets to screwing in light bulbs), without hurting anyone in the process.
毕竟，人形外形的全部意义在于提供 Kim 所说的“多用途移动操纵”，或者说能够在几乎任何地方移动（包括在楼梯上和穿过门）并处理几乎任何东西（从卸载托盘到拧紧灯泡），而在此过程中不会伤害任何人。
In short, what we do every day.
简而言之，就是我们每天所做的事情。
“These things are about [controlling] forces, if you want to do them at speeds of a human,” Agrawal said.
“如果你想以人类的速度做到这一点，这些事情都是关于[控制]力量的，”阿格拉瓦尔说。
“Force control has been a thing in classical [robotics].
“力控制一直是经典[机器人]中的一个问题。
But in modern machine learning land, it’s not been that widespread.”
但在现代机器学习领域，这种情况还没有那么普遍。”
Force control is simple in principle.
力控制原理上很简单。
Picture a robot arm drawing on a whiteboard — without smashing the tip of the marker.
想象一下机器人手臂在白板上画画，而不会弄碎记号笔的笔尖。
Roboticists have known how to make this happen for more than 40 years : They program the arm to behave as if it has an imaginary spring and shock absorber attached to it.
40 多年来，机器人专家们就知道如何实现这一目标：他们对手臂进行编程，使其表现得就像有一个想象中的弹簧和减震器一样。
“One can make the spring really soft in the direction pointing into the whiteboard, and stiffer along the surface of the whiteboard,” Kuindersma said.
“可以让弹簧在指向白板的方向上变得非常柔软，而沿着白板的表面则变得更硬，”库德斯玛说。
“That way the robot maintains the right pressure with the marker while precisely writing the lines and curves of the letters.” This feedback can be driven by force sensors built into the robot’s joints, but the catch is that the classical approaches require a lot of knowledge about the robot, environment, and task in order to work, he further explained.
“这样，机器人就能在精确地书写字母的线条和曲线的同时，保持正确的标记压力。”他进一步解释说，这种反馈可以由机器人关节中内置的力传感器驱动，但问题是经典方法需要大量有关机器人、环境和任务的知识才能工作。
That approach to controlling force works great for industrial robots with specific tasks to perform, and it even helped with humanoid locomotion.
这种控制力的方法非常适合执行特定任务的工业机器人，甚至有助于人形运动。
But it was impossible to generalize.
但这是不可能一概而论的。
Kim’s proprioceptive electric actuators, also called quasi-direct drive actuators, simplified things.
Kim 的本体感受电动执行器，也称为准直接驱动执行器，使事情变得简单。
Not only were they designed to absorb unexpected impacts without damage, they were also very “transparent,” which meant that the motor converted electrical current into a proportional amount of force (and vice versa) with relatively little error.
它们不仅设计用于吸收意外冲击而不造成损坏，而且非常“透明”，这意味着电机将电流转换为成比例的力（反之亦然），且误差相对较小。
In essence, the motor itself became a force sensor, which meant “you can remove cost and complexity from your robot by eliminating dedicated force sensors,” Kuindersma said.
从本质上讲，电机本身变成了力传感器，这意味着“您可以通过消除专用力传感器来降低机器人的成本和复杂性，”Kuindersma 说。
As reinforcement learning eclipsed manual programming as a way of controlling humanoid movement, “classic” force control was not forgotten.
随着强化学习作为控制人形运动的一种方式使手动编程黯然失色，“经典”的力控制并没有被遗忘。
It just got abstracted and delegated, in a way, to both hardware and AI.
在某种程度上，它只是被抽象并委托给硬件和人工智能。
“From an AI point of view, it’s not like you have to be thinking about force control,” Hurst said.
“从人工智能的角度来看，你不必考虑力控制，”赫斯特说。
“It’s more like you kind of know that you need a quasi-direct drive motor to get close [to the force regulation necessary], then put [the neural network] in simulation and iterate a million times — and then you can put it on the robot and get cool behaviors.”
“这更像是你知道你需要一个准直接驱动电机来接近[必要的力调节]，然后将[神经网络]放入模拟中并迭代一百万次 - 然后你可以将它放在机器人上并获得很酷的行为。”
Those neural networks are learning generalized policies that control the positions of a robot’s body parts.
这些神经网络正在学习控制机器人身体部位位置的通用策略。
Force regulation often happens only indirectly in simulation training, or sometimes as a side effect when learned from video or human input.
力调节通常仅在模拟训练中间接发生，或者有时作为从视频或人工输入学习时的副作用。
But those methods don’t explicitly teach the physics of force — at least, not yet.
但这些方法并没有明确地教授力的物理学——至少现在还没有。
“A lot of the signals that are required for doing intelligent force control are not present in [video and human demonstration] data,” Kuindersma said.
“[视频和人体演示]数据中不存在进行智能力控制所需的许多信号，”库德斯马说。
DeepMind’s Parada acknowledged that the VLA models basically just learn to move between specifically defined poses — and this approach goes a long way.
DeepMind 的 Parada 承认，VLA 模型基本上只是学习在特定定义的姿势之间移动——这种方法大有帮助。
“We’ve been surprised ourselves at how far you can push it, without any other sensing,” she said.
“我们自己都感到惊讶，在没有任何其他感知的情况下，你可以将它推到多远，”她说。
In 2015, the most advanced humanoid robots in the world competed at the DARPA Robotics Challenge Finals.
2015 年，世界上最先进的仿人机器人参加了 DARPA 机器人挑战赛决赛。
The tech has since improved.
此后技术得到了改进。
But only so far.
但仅限于此。
As long as robot bodies remain relatively stiff and heavy compared to ours, “they have high inertia, and they’re not [as] compliant,” Agrawal said, which means that without force control, they will struggle with precision tasks in complicated environments.
阿格拉瓦尔说，只要机器人的身体与我们的身体相比保持相对僵硬和沉重，“它们就会具有很高的惯性，并且不那么顺从”，这意味着如果没有力控制，它们将在复杂的环境中难以完成精确的任务。
“If you’re going to touch delicate objects and you have small errors, bad things are going to happen.” Picture a regular egg and another made of solid steel: One of them needs to be picked up much more carefully.
“如果你要触摸精致的物体并且犯了一些小错误，那么糟糕的事情就会发生。”想象一个普通的鸡蛋和另一个由实心钢制成的鸡蛋：其中一个需要更加小心地拿起。
One way to get around this problem, used by many impressive systems alongside positional accuracy, is just to go slow.
解决这个问题的一种方法就是放慢速度，许多令人印象深刻的系统都在使用这种方法来提高位置精度。
Imagine trying to move a chair with your car, Agrawal said: “If I go slowly, I can be precise on how I move [my position], and then I can control where the chair goes, so the [force] problem goes away.” That’s part of why Atlas moves like molasses while grasping auto parts but glides like a gymnast when it’s not touching anything except the floor.
想象一下尝试用你的车移动一把椅子，阿格拉瓦尔说：“如果我走得很慢，我就可以精确地移动[我的位置]，然后我可以控制椅子的移动方向，这样[力]问题就消失了。”这就是为什么阿特拉斯在抓住汽车零部件时像糖蜜一样移动，但在除了地板之外不接触任何东西时却像体操运动员一样滑行。
“It would be an overstatement to say that force control is absolutely required in every useful manipulation task — that’s just not true,” Kuindersma said.
库德斯马说：“如果说每一项有用的操纵任务都绝对需要力控制，那就有点夸张了——事实并非如此。”
But he, Hurst, and Parada all readily grant that clever force workarounds won’t deliver the all-purpose mobile dexterity our robot butlers need.
但他、赫斯特和帕拉达都承认，巧妙的力量解决方法无法提供我们的机器人管家所需的通用移动灵活性。
Even if today’s VLA-brained bots, refined by reinforcement learning, had “an internet-sized” amount of positional data to train on, “it’s very likely you [would] have to do some additional work,” Parada said.
即使今天通过强化学习改进的 VLA 大脑机器人有“互联网大小”的位置数据可供训练，“很可能你需要做一些额外的工作，”帕拉达说。
“Humans feel the forces that are working against you when you’re trying to open a bottle.” Humanoids, for the most part, still don’t, which means they have not mastered physics — at least not in the way we have, from a lifetime of interacting with our environments through the extraordinarily complex musculoskeletal and nervous systems gifted to us by evolution.
“当你试图打开瓶子时，人类会感受到对你不利的力量。”类人生物在很大程度上仍然没有掌握物理学，这意味着他们还没有掌握物理学——至少没有像我们那样掌握物理学，从一生中通过进化赋予我们的极其复杂的肌肉骨骼和神经系统与我们的环境相互作用。
That’s a big reason why even doors and stairs aren’t fully “solved” for present-day humanoids.
这就是为什么即使是门和楼梯对于当今的类人生物来说也没有完全“解决”的一个重要原因。
These stairs, that door?
这些楼梯，那扇门？
Probably.
大概。
But all stairs and doors, plus everything else?
但所有的楼梯和门，加上其他一切？
“There’s no world in which there are actually useful, autonomous [humanoid] robots that are only doing position-based control,” Kuindersma said.
“世界上不存在真正有用的、仅进行基于位置控制的自主[人形]机器人，”库德斯玛说。
“Force as a first-class citizen is absolutely required.”
“作为一等公民，武力是绝对需要的。”
So how do we get over the wall, scientifically speaking?
那么，从科学角度来说，我们如何翻墙呢？
Most of the experts I asked suspect that it will take a new blend of hardware and software advances.
我询问的大多数专家都怀疑这需要硬件和软件进步的新结合。
Tactile sensors for better data collection and robot hands that combine high power, compliance, and transparency with low inertia would accomplish a lot, and nobody believes that true material breakthroughs (like replacing motors with artificial muscles ) will be necessary.
用于更好地收集数据的触觉传感器和将高功率、合规性、透明度和低惯性结合在一起的机器人手将取得很大的成就，而且没有人相信真正的材料突破（例如用人造肌肉取代电机）是必要的。
“The hardware is exceptional, and if you’re blaming [it], you’re making excuses,” said Russ Tedrake, another longtime MIT roboticist I spoke to.
“硬件非常出色，如果你责怪它，那你就是在找借口，”我采访过的另一位长期担任麻省理工学院机器人专家的 Russ Tedrake 说道。
“If you put a human brain through the hardware we have today — by teleoperating it, for instance — it’s incredibly capable.” Finding more intelligent ways to control it is key.
“如果你将人脑置于我们今天拥有的硬件中——例如通过远程操作——它的能力令人难以置信。”找到更智能的方法来控制它是关键。
The Digit robot from Agility Robotics demonstrates fine motor control in an unstructured environment.
Agility Robotics 的 Digit 机器人在非结构化环境中展示了精细的运动控制。
When asked how to achieve that, everyone had a different answer.
当被问及如何实现这一目标时，每个人都有不同的答案。
Agrawal is studying how to combine force control with reinforcement learning by having humanoids learn compliant behaviors in simulation , instead of moving between rigidly defined positions.
Agrawal 正在研究如何将力控制与强化学习结合起来，让类人机器人在模拟中学习顺从行为，而不是在严格定义的位置之间移动。
Tedrake, whose work on “large behavior models” (a cousin of VLAs) produced the apple-coring robot demo, recently argued in Science Robotics for a ChatGPT-style regime of “large-scale data collection and large pretrained models.” Frank Park, who wrote the book on modern robotics — literally, the textbook titled Modern Robotics — believes that current AI approaches should be torn down to the studs and replaced with ones that make physics fundamentals (such as force and acceleration) learnable at a foundational level.
Tedrake 在“大型行为模型”（VLA 的近亲）方面的工作制作了苹果去核机器人演示，他最近在《科学机器人》杂志上主张采用 ChatGPT 式的“大规模数据收集和大型预训练模型”制度。弗兰克·帕克（Frank Park）撰写了有关现代机器人学的书（字面意思是一本名为《现代机器人学》的教科书），他认为当前的人工智能方法应该被彻底淘汰，并代之以那些使物理基础知识（例如力和加速度）在基础水平上可学习的方法。
“The VLA architecture is just all wrong,” he told me.
“VLA 架构完全错误，”他告诉我。
“I believe that approach is doomed to fail.”
“我相信这种做法注定会失败。”
In all these conversations, what struck me most wasn’t the debates about which kinds of sensors, data, or AI architecture could “solve” humanoid robotics.
在所有这些对话中，最让我印象深刻的并不是关于哪种传感器、数据或人工智能架构可以“解决”人形机器人技术的争论。
Rather, it was the sense that the scientific ethos of the field had changed.
相反，这是该领域的科学精神发生了变化的感觉。
Hurst, who had just spun Agility Robotics out of his Oregon State University lab when we first spoke, put a fine point on it.
当我们第一次交谈时，赫斯特刚刚从俄勒冈州立大学的实验室中创立了敏捷机器人公司，他对此提出了一个很好的观点。
“I remember Gill Pratt, who was the director of the MIT Leg Lab and then the program manager for the DARPA Robotics Challenge, saying that his big worry was that we’d end up using reinforcement learning and AI to make robots walk and run before we ever actually understood how it works,” he said.
“我记得吉尔·普拉特（Gill Pratt），麻省理工学院腿部实验室主任，后来担任 DARPA 机器人挑战赛的项目经理，他说他最担心的是，我们最终会使用强化学习和人工智能来让机器人行走和奔跑，然后我们才真正理解它是如何工作的，”他说。
“And in a lot of ways, we’re kind of doing that.” ( Editor’s note: Gill Pratt recalled this conversation differently.
“在很多方面，我们都在这么做。” （编者注：吉尔·普拉特对这次谈话有不同的回忆。
He acknowledged that machine learning could allow performance beyond our formal understanding, but not that this was a cause for worry.
他承认机器学习可以让性能超出我们的正式理解，但并不值得担心。
)
）
Tedrake agreed but said that it’s hardly the first time we’ve taken scientific and engineering leaps without a firm grip on the fundamentals.
泰德雷克对此表示同意，但他表示，这并不是我们第一次在没有牢牢掌握基础知识的情况下实现科学和工程的飞跃。
“If you look at electricity and magnetism, there was the Volta stage where you’re sticking electrodes in frogs,” he said.
“如果你观察电和磁，就会发现伏尔塔阶段就是将电极贴在青蛙身上，”他说。
“And then we had Faraday, who did exactly the right experiments, and then eventually we had Maxwell tell us the governing equations.
“然后我们有法拉第，他做了完全正确的实验，最后我们让麦克斯韦告诉我们控制方程。
I think we’re in the Volta stage.”
我认为我们正处于 Volta 阶段。”
So when will humanoids be solved?
那么人形机器人什么时候能得到解决呢？
“Robots are still bad, and it will take time.
“机器人仍然很糟糕，这需要时间。
But the bones are good.
不过骨头还好。
Both are true,” Tedrake said.
两者都是真的，”泰德雷克说。
“And it’s still hard.”
“而且还是很难。”
Get highlights of the most important news delivered to your email inbox
获取发送到您的电子邮件收件箱的最重要新闻的摘要
Quanta Magazine moderates comments to facilitate an informed, substantive, civil conversation.
《Quanta》杂志对评论进行审核，以促进知情、实质性、文明的对话。
Abusive, profane, self-promotional, misleading, incoherent or off-topic comments will be rejected.
辱骂、亵渎、自我推销、误导、语无伦次或偏离主题的评论将被拒绝。
Moderators are staffed during regular business hours (New York time) and can only accept comments written in English.
版主在正常工作时间（纽约时间）有工作人员，并且只能接受英文评论。