Ghost vs. Tesla, Bringing true self-driving to every car, Raising $230M for auto tech | John Hayes
发布时间 2023-07-21 16:58:27 来源
摘要
In this episode, I'm speaking with John Hayes, co-founder and CEO of Ghost Autonomy.
00:00 - Intro
01:12 - John's background
05:19 - What is Ghost?
09:34 - The Levels of Autonomy
23:13 - Ghost's business model
29:13 - OEM adoption & benefits
39:33 - The future of AVs & commuting
44:11 - Wrapping up
Follow John:
Twitter - https://twitter.com/ghosthayes
LinkedIn - https://www.linkedin.com/in/johnmartinhayes
Ghost Autonomy - https://ghostautonomy.com
Check out the website (https://dealershipguy.com) for more and follow me on Twitter @GuyDealership!
Interested in advertising with CarDealershipGuy? Join the sponsor waitlist here: https://cdgpartner.com
This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions.
GPT-4正在为你翻译摘要中......
中英文字稿 ![](/static/new-window.png)
The product is essentially an acceleration product. You know when it's working and when it's not. Like you can see around you. You know you get it in Uber, you know if the driver is good or not. It's the same thing with this. We're basically like half car enthusiasts and half people who never want to drive.
该产品本质上是一个加速产品。你可以知道它是否在工作。就像你可以看到周围的情况一样。你可以在优步上得到它,你知道司机是否好。这和这个产品一样。我们基本上是半个汽车爱好者,半个不想开车的人。
What's up everyone? This is car dealership guy. You're listening to the car dealership guy podcast, which is my effort to give you access to the most unbiased and transparent insights into the car market. Let's get into today's episode.
大家好!我是汽车经销商。你正在收听的是我的汽车经销商播客,我努力为你提供对汽车市场最公正透明的见解。让我们开始今天的节目吧。
All right, this was a fascinating episode. I think autonomous vehicles in general and the whole idea seems like a very kind of fringe concept to many people. I think John Hayes is an interesting person because he's taken this idea and he's bringing it to reality. I think what they're doing is just a very different approach to autonomous vehicles. He really dives deep into that. He dives into the business model to the economics, how much money they've raised, how they're different in Tesla. Please don't forget to let me know your feedback. Just write tweet at me. I'm really curious to know what you're thinking, so I hope you enjoyed the episode. Let's get into it.
好的,这是一个非常有趣的一集。我认为自动驾驶车辆总体上来说,整个概念对许多人来说似乎是一个非常边缘的概念。我认为约翰·海耶斯是一个有趣的人,因为他将这个想法变成了现实。我认为他们所做的只是一种非常不同的自动驾驶车辆的方法。他真的深入研究了这个。他深入了解了商业模式、经济学,他们筹集了多少资金,他们与特斯拉的不同之处。请不要忘记告诉我您的反馈意见。只要在推特上给我发帖。我非常好奇知道您在想什么,所以我希望您喜欢这一集。让我们开始吧。
All views of car dealership guy and guests on this podcast are solely their opinions. None of the views expressed should be treated as financial advice. This podcast is for informational purposes only.
本播客中的汽车经销商和嘉宾的所有观点仅代表他们个人意见。不应将任何表达的观点视为财务建议。本播客仅用于提供信息。
John Hayes, welcome to the pod.
Good to be here. I've been really looking forward to this episode. I'm super curious to know about Ghost and just autonomy. We haven't had a CDG podcast episode on this topic, but there's a lot of things moving and shaking right now. I think before we get started on Ghost and what it does, can you just give us your brief background? How did you enter this hardware or software space? What is your background? How did you get to this point?
约翰·海耶斯,欢迎加入这期节目。
很高兴来到这里。我一直都非常期待这一集,对于幽灵和自主性我很好奇。虽然我们还没有在这个主题上做过CDG播客节目,但现在有很多事情正在发生和变化。在我们开始讨论幽灵及其功能之前,你能给我们简要介绍一下你的背景吗?你是如何进入硬件或软件领域的?你的背景是什么?你是怎么走到现在这一步的?
Light background is building software. I'm coming out from Canada right after college. I moved to Silicon Valley, joined a startup. I went through a virtual world startup, through a talent search startup and worked at Yahoo. Then the first big break was the company called Pure Storage, which was data storage on Flash Memory. We started that in 2009, we went public in 2015, which was about as fast as you could. The economy was really great back then. And then around 2017, I was getting that itch to start something new and I started Ghost. And so each time I'm changing domains, the interesting thing about Ghost is it doesn't really have anything to do with data storage. What it does have to do with is the observation we made back in Pure Storage was that consumer technology was leading into the data center. That's where all the innovation occurs, that's where all the volume is. What do you mean by that? What do you mean leading to the data center?
轻背景是构建软件的公司。我大学毕业后从加拿大来到这里。我搬到了硅谷,加入了一家初创公司。我经历过一个虚拟世界初创公司,经历过一个人才搜索初创公司,并在雅虎工作过。然后第一个重大突破是一家名叫Pure Storage的公司,它提供基于闪存的数据存储服务。我们从2009年开始,2015年上市,速度可以说是非常快了。那时经济真的非常好。然后到了大约2017年,我开始渴望开始新的事业,并创办了Ghost。所以每次我都在改变领域,而Ghost的有趣之处在于它与数据存储没有什么关系。它与我们在Pure Storage时发现的一个观察有关,即消费者技术正在引领数据中心发展。那是创新发生的地方,也是最重要的地方。你说的引领数据中心是什么意思?
Okay. So what we saw back in 2009 was you had just the MacBook Air had come out. The most expensive laptop had Flash. It had an SSD in it. You also saw the $200 laptop also had SSDs in it. And so you see the top and the bottom of the consumer market. And so what we predicted was no one would have any hard drives in the consumer market anymore. And what that moves is tons and tons, like billions of dollars of R&D to keep making this product better. And our assumption was that over time that solid-state storage in SSDs would get better faster than hard drives. And that basically came true. And if you look around your life, you don't have any hard drives anymore. And so our assumption was that R&D path would mean that the Flash member would take over absolutely everywhere. And so what we did was we took those consumer products that no one said everyone said, wouldn't work, wouldn't be reliable enough, and built some smart software around it and put it in data center where it runs 24 seven under the most durable applications.
好的。所以在2009年的时候,我们看到的是MacBook Air刚发布。最贵的笔记本电脑采用了闪存和固态硬盘(SSD)。同时,我们还看到廉价笔记本电脑也开始采用SSD。所以你可以看到消费市场的顶端和底端都在采用SSD。我们预测消费市场将不再使用硬盘。这个转变意味着数十亿美元的研发投入,以使这种产品变得更好。我们的假设是,随着时间推移,SSD固态存储技术的发展速度将超过硬盘。这一假设基本上成真了。如果你环顾你的生活,会发现已经没有硬盘了。因此,我们料想研发道路将意味着闪存将完全取代硬盘。因此,我们采取了那些被人们普遍认为不可行、不够可靠的消费品,并在其周围构建了一些智能软件,并将其放置在数据中心中,在最耐用的应用程序下全天候运行。
So essentially, you brought storage to the cloud?
Well, storage was in the cloud, but it was on a much cheaper medium, much more reliable, well proven medium.
Got it. And did you take pure storage public as CEO?
No, I was not CEO. So this is the first CEO job for me. I was a technical leader in that company. And so now I'm stepping up to CEO for the first time in this company.
那么你基本上是把存储引入了云端?
嗯,存储确实在云端,但它是在更便宜、更可靠、经受了很多考验的介质上。
明白了。那你是作为纯粹存储的首席执行官将公司上市的吗?
不,我不是首席执行官。这是我第一次担任CEO职位。在那家公司,我是技术领导者。所以现在我是第一次在这家公司担任CEO的职位。
So what was that like? I mean, you mentioned 2017, I saw, I noticed you were working at a VC fund in 2017 for a short stint. Was that the point where you had, you know, ghost, you had this idea, and then you left? Or how did that work out?
那是什么感觉呢?我的意思是,你提到了2017年。我注意到你在2017年短暂地在一家风投基金工作过。那是你构思这个点子并离开的时候吗?还是有其他原因?
So I was there as an entrepreneur in residence. And so what that means is kind of like entrepreneur welfare, where they pay you a salary, they give you a better welfare. And you're supposed to come up with a billion dollar company idea over like sometimes over a year, sometimes over two years. And so I was there for the purpose of starting a new company. And I met my co-founder, Volkmar there. There were other companies started out there at the same time.
所以我在那里担任企业家驻地。这意味着我像是享受企业家福利一样,他们付给你工资,给予更好的福利待遇。你的任务是在一年或者两年的时间内提出一个价值十亿美元的公司创意。因此我在那里是为了开始一家新公司。我在那里遇到了我的共同创始人Volkmar。同时还有其他公司也在同一时间创立起来。
And so for those few months, yeah, I was just, you know, going in every day, researching what's new in technology, because there has to be something new, like there has to be a reason to start a company, something that's changed. And then what market could be applied to? And the transportation automotive market is just one of the biggest markets in the world. And so if you can find technology to apply to that, that's going to be a great business if you can execute it.
所以在那几个月里,我每天都在研究科技领域的最新动态,因为一定会有新的东西,就像要创办一家公司,需要有改变的理由。然后我想到了能够应用于哪个市场?交通汽车市场是全球最大的市场之一。如果你能找到适用于这个市场的技术,并且能够执行得好,那将会是一项伟大的业务。
So explain to me simply what is Ghost? What is the thesis behind the company?
请简单地向我解释一下Ghost是什么?该公司的主要论点是什么?
The thesis is that when I live in Mount of California, so I see all the robo-taxi companies, Waymo's based there. And the thesis is is that when you look at how people actually use transportation, you know, you add in all of taxis, all rideshare, all public transportation, all bicycling, and adds up to 2% of trips. And so what I saw was that no one was going after the 98% of trips, which is people driving their own cars. And the barriers to that were, one, the hardware had to get much, much cheaper. And so when you see a robo-taxi, at the time, it was probably like a couple hundred thousand dollars of hardware. And they think they're going to get that to maybe tens of thousands, but that was never going to be the right scale to put on a consumer product. And so we started with a baseline where we wanted to assume that let's assume you put like a thousand dollars, like of cameras and computers, like really, really pair it down. There's something that you could put on every single car and then build autonomy software to make that work. And so that was the naive thesis back in 2017.
这个论点是,当我在加利福尼亚的山脉居住时,我看到了所有的机器人出租车公司,其中Waymo公司是其中一家。这个论点是,当你看到人们实际使用交通工具的方式时,你会发现,包括出租车、共享乘车、公共交通和自行车等,这些方式加起来只占出行的2%。因此,我观察到没有人去关注剩下的98%的出行方式,也就是人们自己驾驶汽车上路。而其中的障碍是第一,硬件需要变得便宜得多。当你看到一个机器人出租车的时候,那个时候它可能需要几十万美元的硬件成本。他们认为他们可能会将硬件成本降到几千美元,但那从来不会是适合消费者产品的规模。因此,我们从一个基线开始,假设你只放入一千美元的摄像头和计算机等设备,真的是尽可能简化,这是你可以应用在每辆汽车上并构建自动驾驶软件使其运作的解决方案。所以,这就是2017年的天真论点。
Yeah, and I want to hear how it evolved. But you're basically saying, let me retrofit someone's car so that it can become some form of self-driving, basically.
是的,我想听听它是如何演进的。但你基本上是在说,让我给某个人的汽车进行改装,使其基本上能够变成一种自动驾驶形式。
That's how we started out. So we were thinking, it's like, oh my god, the car companies were very difficult to sell through. It was a very long path. We got to consumer. Yeah, we would go direct to consumer. And starting about 2021, we decided to reorient the company and say, no, no, we're going to sell to auto manufacturers because they started wanting to have that conversation. So when we started in 2017, a typical response would be, oh, the auto companies will build what you're building. Like, because 2017 supercruise had come out, Tesla autopilot was out. It seemed like they could evolve their products. But now starting about 2021 or 2022, you're seeing a lot of turmoil in those projects inside the auto companies. And so now they're changing their tune where now they're actually looking to buy something rather than try and build it internally.
这就是我们最初的起步方式。所以我们想,哇哦,汽车公司的销售非常困难。这是一条很漫长的道路。我们接触到了消费者。是的,我们直接面向消费者销售。从2021年开始,我们决定重新定位公司,说不,不,我们要卖给汽车制造商,因为他们开始愿意进行这样的对话了。所以当我们在2017年开始的时候,典型的回应是,哦,汽车公司会制造你在制造的东西的。因为到2017年,supercruise已经发布了,特斯拉的自动驾驶系统也已经出来了。看起来他们可以进化他们的产品。但是从2021年或2022年开始,你会看到汽车公司内部的项目中出现了很多动荡。所以现在他们改变了立场,他们实际上开始寻找购买的东西,而不是试图自行建造。
Why is that? Is it just, you know, they can't get their shit together from a tech perspective? Is it just too hard? Well, what is it?
为什么会这样?难道他们只是无法在技术方面做好吗?是不是太难了?那又是什么原因呢?
I think that there's two things. There's a big thing. So one is that auto companies have not made software first-class engineering discipline. And so the hierarchy is built on mechanical engineering, electrical engineering. And when you look at the senior leadership, who came up through engineering, who didn't come up through marketing, it's almost always those two paths. And software engineering is just a completely different discipline. So they're trying to build these organizations, and they have to be run differently.
我认为有两件事。首先是有一个重大问题,即汽车公司尚未将软件作为一流的工程学科。因此,其层级结构是建立在机械工程和电气工程的基础上的。当你看到高级领导层时,他们都是通过工程学背景而非市场营销路线晋升上来的,几乎总是这两种路径。而软件工程则是完全不同的学科。因此,他们试图构建这些组织,并且必须有不同的管理方法来运营。
The other thing is that there's no evolutionary path between something that unreliably flashes a light at you and something that will actually drive a per 100% of the time. There's just orders of magnitude more complexity. And so, I think that they've discovered that they can't evolve from features like individual safety features and maybe some lane centering into something that's full autonomy, and that they're built in a completely different way.
另一件事是,一个不可靠地向你闪烁灯光的物体与一个能百分之百地驾驶车辆的自动驾驶系统之间并没有进化的路径。它们之间存在的差距是数量级的复杂性。因此,我认为他们已经发现,无法通过单独的安全功能和可能的车道居中等功能演化出全自动驾驶,它们需要以完全不同的方式构建。
So you think you can get the car that's in my garage right now to actually be full self-driving by just retrofitting some equipment on it?
那么你认为只需给我车库里的那辆车装备一些设备,就能使其实现完全自动驾驶,是这么想的吗?
Well, we think that the next car you buy will have the right equipment to be autonomous. And so we target it. But it won't necessarily be autonomous. That's up to the auto companies. But what we target is the exact same computers that they're already building into cars. So these are, you see it in the entertainment system, the exact same cameras that they're already building into cars. Like it seems like every car now has between like four and nine cameras and the radars and tying that together into an autonomous system. And you especially see this with electric vehicles where they're just changing the entire architecture to make it centered around a computer instead of being centered around the motor.
嗯,我们认为您购买的下一辆汽车将具备自动驾驶所需的合适装备。因此,我们以此为目标进行研究。但它不一定是自动驾驶的。这取决于汽车公司。但我们所追求的是与它们已经安装在汽车中的相同的计算机。所以,您可以在娱乐系统中看到,与它们已经嵌入在汽车中的完全相同的摄像头。就如同现在每辆汽车似乎都配备了四到九个摄像头和雷达,将它们整合到一个自动驾驶系统中。特别是在电动汽车上,您会看到他们正在改变整个架构,以将其重点放在计算机上而不是发动机上。
Can you explain to me the different levels of autonomy? I think before we even talk about, you know, the capabilities of ghosts, I'm trying, I want to level set and understand what exactly you're striving for here. The levels of autonomy are usually thought of as the number levels, like one, two, three, four, five is unattainable. It's some future that may never exist. One is like cruise control that we've had since the 70s. Okay. So most of the products you see on the market are level two, meaning that they do some assistance. They do some lane keeping, even things like blind spot information systems counted level two, like everything is just laughing because my wife hates it. Like the car will suddenly stop her and she'll just freak out because she's like, dude, it's completely surprises her out of nowhere. Yeah. And you probably find when you rent a car and like makes a noise at you and you're kind of confused as to what that noise is. That's a level two feature.
你能向我解释一下不同级别的自主性吗?在我们讨论灵活性能力之前,我想要先理清楚你在这里追求的是什么。通常,人们将自主性分为不同的级别,如一、二、三、四,五层级是无法达到的,可能是一个从未存在过的未来。一级就像是从70年代开始就有的巡航控制。所以市场上大多数产品都是二级的,意味着它们提供一些辅助功能,比如车道保持,甚至盲点信息系统也算是二级的,就像我的妻子讨厌的那样,一切都是笑话。就好像车突然会把她吓一跳,她觉得很突然。是的,当你租车时,如果车突然发出噪音,你可能会感到困惑,不知道那声音是什么。那就是二级功能。
And so, and so then in between you have level three, which is this strange zone where you've seen Mercedes implement a level three where they do a traffic jam assist. So if you're going slow, like under 35 miles an hour in just the right conditions, you can look away from the road and the car will just coast along. And BMW is talking about bringing up the same feature, but they're going to go up to 40 miles an hour. And so often it's just very limited conditions. And so the thing that everyone really wants is level four, meaning that you can set a destination or be driving along and you don't have to look at anything. The car will maybe try and track your attention. But if you don't give it to the car, then it will pull over to the side of the road. Like if it thinks it can't continue, it has enough self diagnostic that you don't have to be watching it to make sure it's not going to do something crazy. And so you've seen this appear in two places. So one is in robotaxis. So coming as like way among crews are the most prominent. You're seeing it in trucking. And more recently, you're seeing level four in China. So China is actually on the leading edge of taking level four technology and putting into consumer cars. And you can go there and like try. And that was a big shock to the auto industry. You starting earlier this year.
因此,中间还有第三级,这是一个奇怪的区域,在这个区域中,你会看到梅赛德斯实施了一个第三级的功能,就是在堵车时的辅助驾驶。所以如果你在慢速行驶,比如在35英里/小时以下的适当条件下,你可以将目光从道路上移开,汽车会自动行驶。宝马也在考虑引入同样的功能,但他们的速度将提高到40英里/小时。通常情况下,这些功能只在非常有限的条件下可用。所以每个人真正想要的是第四级,也就是你可以设定一个目的地或者一直开车,而不需要关注任何事情。汽车可能会试图引起你的注意,但如果你不理会它,它会停到路边。如果它认为无法继续行驶,它具有足够的自我诊断能力,你不必盯着它,以确保它不会做出疯狂的动作。你在两个地方见过这个功能。一个地方是出租车中,比如Waymo是最知名的。还有在卡车行业中也在使用。最近,中国也开始使用第四级技术,将其应用到消费级汽车中。你可以去那里试试。这对汽车行业来说是一个巨大的冲击,从今年开始就出现了这种情况。
What's Tesla? Tesla would themselves, they would say that they're a level two in that you have to watch it to make sure that it doesn't do something crazy. It works much of the time, but not all the time. Clearly, their engineering goal is to get to a level four, where you don't have to watch it. But they had to do probably years of improvement to get to that point. Yeah, but how is it possible it's level two? If you know, there's you see people just, you know, going to Tesla, obviously they say you have to, you know, hold the wheel or whatever for liability. But what is it really like practically speaking? I think practically speaking it's level two. And that's because if you just let it do it at once and you don't pay attention, you will have a clue. Almost everyone who is driven Tesla has found a moment where it scared them. And I think, you know, I live in California, I'm surrounded by Teslas and Tesla drivers. Everyone has sort of learned where it works and they turn it on and off, depending on where it's going to work. And I think that that's consistent with Tesla taking a very breadth-first approach. They wanted to do a little driving in a lot of places.
Tesla是什么?特斯拉会自称是二级自动驾驶,也就是说你必须留意它,以确保它不会做出什么疯狂的举动。它通常可以正常工作,但不总是如此。显然,他们的工程目标是达到四级自动驾驶,即不需要人类监督。但要达到这一点,他们可能需要数年的改进。是的,但它怎么可能只是二级自动驾驶呢?如果你了解,你会发现人们只需坐在特斯拉车里,他们显然说你必须像对待任何其他车辆一样自己握着方向盘,以规避责任。那么从实际角度来说,它是什么样子的?我认为从实际角度来讲,它就是二级自动驾驶。这是因为如果你只是一次让它自动驾驶,而不去留意,很快你就会有所察觉。几乎每个开过特斯拉的人都会遇到吓到他们的时刻。我住在加州,周围都是特斯拉和特斯拉车主。每个人都学会了在哪些地方它适用,他们会根据需要随时开启或关闭它。我认为这与特斯拉采取了广度优先的策略是一致的。他们希望在许多地方进行少量的自动驾驶实践。
What we're focused on is doing extremely good driving in fewer places. What are those places? Mainly freeways, like freeways express ways, highly structured environments where you don't have a lot of pedestrians and bicyclists and other road users like that.
我们的重点是在更少的地方进行非常出色的驾驶。那些地方是哪些呢?主要是高速公路,比如高速公路和高速路这样的高结构环境,在那里你不会遇到太多行人、骑自行车的人和其他类似的道路使用者。
You mentioned Waymo, you know, cruise comes to mind. What are the key differences here? I mean, you hear a ton about Waymo and Cruise that you're always in to press, you know, so what are the key differences here between Ghost and what Waymo and Cruise are doing?
你提到了Waymo,你知道,Cruise也跃入脑海。这里有哪些主要区别呢?我的意思是,你总是听到关于Waymo和Cruise的新闻报道,你知道,所以Ghost和Waymo、Cruise所做的事情有哪些主要区别呢?
So they have two major differences in their technology. One is that they use specialized sensors called LiDARs, which were originally used for engineering and site surveying. So you can measure exact distances with pretty good density around you. And so it's used for making a 3D map of an environment. The downside is they have to be outside the vehicle because it's laser, so it goes along the line of light. And they're quite extensive and difficult to maintain.
因此,它们的技术有两个主要的不同之处。首先,它们采用称为LiDAR的专用传感器,这种传感器最初是用于工程和现场测量。因此,您可以以相当高的密度测量您周围的确切距离,用于创建环境的三维地图。缺点是它们必须安装在车辆外部,因为它们使用激光技术,激光沿着光线传播。此外,它们非常精密且难以维护。
The second thing is that they use technology called HD maps, where they program in all of the detailed rules of how to drive on a particular street into a database and they use that database. And so mostly what they're looking for with their sensors is the other road users like cars and, and, you know, pedestrians and all the all the mobile things that they can incorporate into the HD map. And so that's why you see Waymo launching in particular cities because they spend up to a year before that doing a detailed survey of the city to build up this database of how the roads are connected and what all the rules are.
第二个问题是他们使用了被称为高清地图的技术,将在特定街道上行驶的详细规则编入数据库,并使用该数据库。因此,他们的传感器主要用于检测其他道路用户,如汽车、行人以及所有可以纳入高清地图的移动物体。这就是为什么你会看到 Waymo 在特定城市进行推出,因为他们在此之前会花上一年的时间,对该城市进行详细调查,建立起道路连接方式和所有规则的数据库。
We're taking an approach that's more similar to what Tesla is. So one, we don't use any specialized sensors. So our thesis is that, you know, ordinary cameras, again, there's an amazing amount of our data-approved cameras, plus AI means that you don't need specialized sensors. The second is we drive in what you call first-person mode, where instead of trying to anchor against a map by saying, where am I in the world? What are the rules at this point? We just interpret the scene visually as it comes in. And then the maps we use are ordinary navigation maps. So there isn't detailed information about exactly where you should drive on the road. Instead, we orient to what you can see. And so the advantage is that you make a system that works in a lot more places. So anything that looks like a highway it'll work in. You don't need to have pre-surveyed that. But the main thing is it's much cheaper and much lower maintenance.
我们采取的方法更类似于特斯拉。所以,首先,我们不使用任何专门的传感器。所以我们的观点是,你知道的,普通的摄像头,再加上令人惊艳的数据审批的摄像头数量,再加上人工智能,就不需要专门的传感器了。其次,我们采用所谓的第一人称模式驾驶,而不是通过问自己:“我在世界上的哪里?这一点上有什么规则?”来锚定在地图上。我们只是根据场景的可视化信息来解读。然后我们使用的地图是普通的导航地图,没有关于在道路上具体应该如何驾驶的详细信息。相反,我们根据你所能看到的来定位。所以优势是可以在更多地方运行。任何看起来像高速公路的地方它都能工作。你不需要事先测量。但最重要的是,它更便宜,维护成本更低。
Do you have an active prototype right now? Yeah. Yeah. We've been driving up to 5,000 miles a month in California now in Detroit, and we just crossed the border into Canada. It's just this week. So that's exciting for me as a Canadian. I put a tow back in Canada.
你们现在有一个正在使用的原型机吗?是的。是的。我们目前每个月在加利福尼亚和底特律一共行驶了5000英里,而且我们刚刚越过了加拿大边境,就在这周。作为一个加拿大人,这对我来说非常令人兴奋。我把车拖回了加拿大。
Wow. What was that like for you? Like were you the driver passenger? What was that like? I was here in Toronto. So our team was out the border. So they crossed the border. They have some pictures under the Canadian flag. So we're going to be letting people know that's happened. And maybe letting the police know because they didn't quite answer their phone. So we stayed under the radar enough, even though it's a totally wrapped car. So it looks unusual. But yeah, it worked perfectly. Canadian highways are pretty similar to American Highway.
哇,这对你来说是什么感觉?你是司机还是乘客?那个经历是怎么样的?我当时在多伦多。我们团队出去过边境。他们在加拿大国旗下拍了一些照片。所以我们将告诉人们发生了什么。也许通知一下警察,因为他们没有接电话。所以尽管这辆车完全包装起来,看起来很不寻常。但是,是的,它完美地起到了作用。加拿大的高速公路和美国的高速公路相当相似。
That's incredible. What was that feeling? I'm curious, like the first time you stepped into that car and you started driving, were you scared? You're like, I hope I don't fuck this up. It's almost like Jeff Bezos going on the the Blue Origin Rockets. Like, you know, I hope this thing doesn't explode when I think it. Yeah, it's probably not that dangerous. I mean, the first time we started driving late last year, like August, and the first time you go on the road, it did not work well. Like the steering is wrong, the speed is wrong. And it just takes like months of just sitting there and tuning it and fixing problems over and over. But that was the culmination of four and a half years of development and finally bringing it all together for the first time was just amazing for the entire team. It's like we just had we were just set up on the highway and just having people take runs. It's like, hey, this really works end to end and it's built on and there's nothing magic in the system.
太不可思议了。当时的感觉是怎样的?我很好奇,就像你第一次踏上那辆车并开始驾驶时,你害怕吗?你可能会想,“我希望我不要把这搞砸。”这就像Jeff Bezos登上蓝色起源火箭一样。你知道的,我希望这玩意不会在我想的时候爆炸。嗯,可能不那么危险。我指的是去年8月我们第一次开始驾驶,上路的第一次并不顺利。方向错了,速度也错了。需要花上几个月的时间坐在那里调试和解决问题。但那是四年半开发的巅峰,最终将一切汇聚到一起的时刻,对整个团队来说简直太神奇了。就好像我们在高速公路上摆好摊位,让人们进行测试。就像是说,“嘿,这东西完全正常运行,没有什么神奇的。”
So on that note, you mentioned four and a half years. What are the ways I like to come up with questions for my podcast is I just like to put myself in founders or CEOs or an executive shoes. And I do like this, you know, real timeline.
所以在这一点上,你提到了四年半。我喜欢为我的播客想出问题的方式就是将自己置身于创始人、首席执行官或高层管理者的角色中。我喜欢通过这种方式来构建一个真实的时间线。
And one of the things that I thought of when I put myself in your shoes, I said, 2017, we're now in 2023. Like they say, you know, being an entrepreneur is like, you know, eating glass or whatever. I mean, your feedback loop is like, in existence, you know, like you're not yet selling to market or, you know, maybe you're working on deals, but you're not like in market yet.
当我设身处地地想象自己是你的时候,我想到了一件事。我说,现在是2023年,而当时是2017年。就像大家常说的,创业就像吃玻璃或者其他什么。我的意思是,你的反馈循环还不存在,你还没有开始向市场销售,或者可能你正在开展交易,但你还没有完全进入市场。
And you're talking about six years of development building, trying to prove something. How like how, you know, and especially given you said you were from enterprise before that, I mean, enterprise, you build a product, let's go, let's get the first sale. It's like right away. Let's get that. Let's get that dopamine. Hey, let's bring some sales at the door.
而你是在谈论着长达六年的开发建设,试图证明某种东西。就像,你知道的那样,尤其是考虑到你说你之前是从企业跳槽来的,我的意思是,企业是这样,你建造一个产品,然后就可以去销售了,立刻就要去争取第一笔销售成果。就像立刻得到一些多巴胺一样。嘿,让我们带一些销售额到门口。
So how, what is it like running a company where for, you know, five, six years, it's like, you know, I wouldn't call it maybe an in existence feedback loop, but it's a very, it's tough, you know, and especially a CEO motivating the team. What, what does that like for you?
那么,你知道,在过去的五六年里经营一家公司是怎样的吗?我不会说这像一个没有反馈循环的存在,但确实非常困难,尤其是对于一位CEO来激励团队。那么,对你来说,这是什么感受呢?
It's, it's even more interesting with the investors who would, of course, love to see feedback loop because they're putting a lot of money into this year after year. One of our investors, Keith, your boy, you know, his theory is, I just tweeted that. I just tweeted that. Yeah. It's, it's sort of the, the anti lean startup, which is like, no, what you should do is it's like making a movie. You should make the movie and then sell tickets to the movie. Like it's your vision to make that work.
这是,对于那些愿意投入大量资金的投资者来说,情况变得更加有趣了,因为他们一年又一年地将资金投入其中。我们的一个投资者,Keith,你们的朋友,你知道的,他的理论是,“我刚才在推特上发了这条消息。是的,这有点像反刻苦创业,就像是,你应该先制作电影,然后卖票给观众,就像是你的愿景来使之成功。”
And I think that we're in a space where it's a bit like the last company where the product itself is, is obvious in a sense. In that if you sit in a car and the car drives, it's the right product. And you know that it's product that a lot of people want. Yeah. And that's where this company is very different from previous companies where, you know, I, I'm not going to buy a storage system for myself. You go home for Christmas and what do you do? It's like, I build data center equipment. This is like obvious. And when you sit there and we, you know, when we bring in partners from auto companies and share ones and they send the car, they're like, Oh, yeah, this is working. This is obviously working.
我认为我们现在所处的空间有点像上家公司,产品本身在某种程度上是显而易见的。就像你坐在车里,车自己开动,这就是正确的产品。而且你也知道很多人都想要这种产品。是的,这就是这家公司与以往公司非常不同的地方,你知道的,我不会给自己买一个存储系统。你回家过圣诞节的时候做什么?就像,我在建设数据中心设备。这是显而易见的。当我们与汽车公司的合作伙伴一起坐在那里,他们寄来了车,他们会说,哦,是的,这个是能工作的。显然是有效的。
And so, so a lot of it is one, I would say the harder part over that time was there was a lot of false starts. It probably took us, we had a long R and D pair, we had to solve some fundamental problems. Like, okay, you have a camera, you want to measure distance. What are your ways? And we went through about four completely unique ways to do that. You want to figure out the road geometry went through three completely unique ways to do that. We've rewritten how we do planning. Like, how do you decide where to drive? That's been totally rewritten three times. I think it's on the fourth. How do you control the car? That's been rewritten three times. And they think that to keep the team motivated, first, our team is mainly engineers. And so the cost and improvement is in solving problems.
所以,其中有很多的过程,我会说那段时间最困难的部分是一开始有很多失败的尝试。我们经历了一个漫长的研发过程,需要解决一些根本性的问题。比如,你有一个相机,想要测量距离,你有哪些方法?我们尝试了四种完全不同的方法。想要确定道路几何形状也经历了三种完全不同的方法。我们重写了车辆的规划方法,比如,如何决定驾驶的路线,已经改写了三次,现在是第四次了。如何控制车辆也改写了三次。他们认为为了保持团队的积极性,首先,我们的团队主要是工程师。他们改进问题解决的成本就是团队进步的动力。
Yeah, solving problems and having the sort of the right sense of when should you start over and build something completely new. And we've done that multiple times over the years to get it right. Because otherwise you get stuck and you make a lot of compromises. But this is definitely the longest development cycle I've ever done in my life. Because we're at six years, we probably have at least another year of pretty hard development. We're starting to get some feedback from the auto companies now. Some of it is determining their psychology. But you have to create internal posts. But a lot of it is making sure that everyone in the company goes in the car at least once a month. And they can see that it's getting better and get that sort of, the product is essentially an acceleration product. You know when it's working and when it's not.
是的,解决问题并且有一种正确的感觉,知道什么时候应该重新开始,全新地构建一些东西。这些年来,我们已经多次这样做,以确保做到正确。否则你就会陷入困境,做出很多妥协。但这绝对是我一生中开发周期最长的项目。因为我们已经进行了六年,可能还需要至少一年的艰难开发。我们开始从汽车公司那里获得一些反馈了。其中一些是在确定他们的心理状态。但是你必须发布内部帖子。但很大一部分工作是确保公司里的每个人每个月至少一次坐进车里。他们可以看到它正变得越来越好,从而产生一种产品加速的感觉。你知道什么时候它工作,什么时候它不工作。
You can see around you, you get it in Uber, you know if the driver is good or not. It's the same thing with this. And that's also been essential for helping people figure out what they should do next. Because they know something is wrong. How can I contribute? If I work out controls, I can make the controls better, I can make the visualization better. And so it's very personal. Almost everyone who works at the company wants this product. We're basically like half car enthusiasts and half people who never want to drive. And so everyone wants it. And they know what they want from themselves. And so as long as they see that continuing to prove, that's where the motivations come from.
你可以在你周围看到,你乘坐 Uber 就可以获取到这一点,你知道司机是好还是不好。这和这个产品是一样的。这个对于帮助人们弄清楚接下来应该做什么也是至关重要的,因为他们知道某些事情出了问题。我可以如何贡献?如果我调整控件,我可以使控件更好,我可以使可视化更好。所以这是非常个人的。几乎每个在公司工作的人都想要这个产品。我们基本上可以说是半个车迷和半个不想开车的人。所以每个人都希望这个产品。他们知道他们想要什么。只要他们看到持续的进步,动力就会来源于此。
What's it like to actually retrofit a car? Is it like, hey, come to my garage 30 minutes, let me slap some hardware on this thing? Or what is that like?
实际对汽车进行改装是什么感觉?是不是就像说:“嘿,来我车库待个30分钟,让我给这玩意装上一些硬件”?或者说,那个过程是怎么样的呢?
Well, so when we build cars, it probably, it takes like a couple of weeks because the guys like literally take apart the entire car we're doing interfacing with the controls of the car. We have to add computers into the car. We wire up cameras all around. Some stuff is glued on. Some stuff like all the doors come off. And we install radars. It's a process that takes weeks right now. And that's one of the things that convinced us that this really does have to be manufactured in. This is complicated enough that doing after the fact, we'll probably be hundreds of thousands of dollars to do retrofit. And retrofitting isn't a great market. People just don't spend that much on cars after they buy them for things that aren't maintenance.
当我们建造汽车时,可能需要几个星期,因为工人会彻底拆解整辆车,并与车辆控制系统进行接口。我们必须在车辆中添加电脑,并固定摄像头。有些部件是粘贴在上面的,有些部件(如车门)会拆下来,然后我们再安装雷达。这是一个需要几个星期的过程。这也是我们确信这些汽车必须在制造过程中进行生产的原因之一。这个过程已经够复杂了,如果事后再进行改装,可能需要花费几十万美元。而且改装市场并不是一个很好的市场,人们在购买汽车后通常不会花太多钱来做非维修性质的事情。
Yeah. And I think also the other thing to consider with consumers is, if I retrofit my car, did I avoid the warranty? Yeah, no one knows. It's sort of uncharted territory and the auto companies will threaten that you have. Yeah, which is where I think it makes sense to partner with auto manufacturers, almost be like an authorized whatever, so that you can just cross that hurdle.
是的。我认为还有另一件需要考虑的事情是,如果我对我的车进行改装,会不会取消保修?是的,没有人知道。这是一个未知领域,汽车公司会威胁说会取消保修。这就是我认为与汽车制造商合作并成为授权合作伙伴是有意义的原因,这样你就可以顺利解决这个问题。
Tell me more about the economics of this business. I have a couple questions. Starting very high level, how much funding have you raised?
So we've raised just over 200 million funding, about 230 million. And so all of that has been equity funding, mostly entirely from financial investors.
告诉我更多关于这个企业的经济情况。我有几个问题。先从非常高的层面开始,你们筹集了多少资金?
我们已经筹集了超过2亿美元的资金,大约是2.3亿美元左右。目前所有的资金都是股权投资,主要来自金融投资者。
Who was your first investor?
So first investor was Mike Spicer out of Silver Hill. He was also the first investor in pure storage back in the day. So had some confidence that we would make something work.
你们的第一位投资者是谁?
我们的第一位投资者是来自Silver Hill的Mike Spicer。他也是当时Pure Storage的第一位投资者。因此,他对我们能够取得成功有一定的信心。
Had some confidence that you know what you're doing.
Maybe, maybe. Hopefully. Yeah, hopefully. Okay. And so what's the money being put towards? Is it development, employees, anything else?
It's almost all R&D. And what's nice about making an extensive hardware is the cars themselves are at that expensive, the building that expensive. So almost all of it is R&D. And it's probably two-thirds software, one-third hardware R&D.
有点信心你知道自己在做什么。
也许吧,希望是这样。是的,希望如此。好的。那么这笔钱用来做什么?是用于开发、雇员还是其他什么?
几乎全部都用于研发。制造一款复杂的硬件产品的好处是汽车本身并不昂贵,建造的成本也不高。所以几乎全部都是用于研发。大概有三分之二是软件研发,三分之一是硬件研发。
Got it. And so this is the previous company, Pure Storage, it was like a lot of raised capital went building an enterprise sales force, because that was it. Well, we have 12 customers in the world. So it's almost all R&D. And we're selling into the engineering departments of those companies.
明白了。那么这是之前的公司Pure Storage,他们花了很多资本来建立一个企业销售团队,因为这是他们的主要工作。我们世界上只有12个客户,所以几乎都是研发工作。我们是向这些公司的工程部门销售产品。
And what's your cost of, you know, a piece of hardware to equip one car? How much does that cost?
So the final cost to build should be under $1,500. And yeah, for the company. And then for the consumer or the manufacturer, how do you think about that? So the way we think about it is a bit different. One, we're not actually interested in making money off of hardware. You know, you put a computer in a car and it's going to be the same computer. And I think that cars would benefit from standardizing that the camera we buy from a camera maker. We don't make cameras. Someone else makes it. And so our goal is to get the car companies to just put a computer in every single car. And then our business model is based on licensing the software at the point of activation. And whether people pay in terms of an upfront cost or with a bit of monthly cost is neutral to us. Because we don't have expense until they actually start using software.
你们一台用于装备一辆车的硬件的成本是多少?那需要多少费用?
所以建造的最终成本应该在1500美元以下。对于公司来说,是这样的。然后对于消费者或制造商来说,你们是如何考虑的呢?我们对此有点不同的看法。首先,我们实际上并不对硬件赚钱有兴趣。你知道,你把一台电脑放在车里,它就是一台相同的电脑。而且我认为汽车会受益于标准化,就像我们从一个摄像机制造商购买摄像头一样。我们不生产摄像头,是其他公司生产的。所以我们的目标是让汽车公司在每辆车上都安装一个电脑。然后我们的商业模式是基于在激活时许可软件。无论人们是一次性费用支付还是分期付费,对我们来说都是中立的。因为直到他们真正开始使用软件之前,我们都没有支出。
So you're basically betting or you're goal, what you're optimizing for is to make to for the manufacturers to put the necessary equipment as a just natural course of progression in these cars. And then you piggyback off that, that hardware that's already pre installed. And you sell them to software, which is like light years ahead, because you've been working on it for so long and you've come such a long way.
Yeah. And this is a very different business model for the auto manufacturers. We're very used to saying everything I sell has a piece of hardware and the cost associated with it. And then they have to inventory manage all of that. And we're saying no, speculatively, put a computer in there that you'll probably use anyways, because you still have your screens that you have to run, you still have level two features, like it's not going to go completely unused. But you're speculatively putting this in there, knowing that a pretty good percentage of customers will upgrade and use all the hardware.
所以你基本上是在赌注,或者你的目标是让制造商自然而然地在这些汽车中安装必要的设备。然后你利用已经预先安装的硬件,然后将它们销售给已经进行了长时间研发并且取得了很大进步的软件。
是的,这对于汽车制造商来说是一种非常不同的商业模式。我们很习惯于说我销售的每个东西都有一块硬件和相关的成本。然后他们必须对所有这些进行库存管理。而我们却说,不,可以先在里面投放一台计算机,因为你仍然有屏幕需要运行,你仍然有二级功能,它不会完全闲置。但你是在投机性地放入这个东西,知道相当大比例的客户会升级并使用所有硬件。
What are you building specifically on a software side that you think is defensible? Like if I'm, and I agree with you that the manufacturers, they don't have that like software engineering muscle or that's not at least our core competency. But what is it about go specifically that will get you to a point where you know, you are the software they need, they have to come to you as opposed to, you know, a competitor or building it themselves, whatever it may be.
在软件方面,你具体在建设什么,你认为这是可以防御的?比如,我同意你的观点,制造商们没有软件工程的实力,或者说这至少不是我们的核心优势。但是具体而言,围绕围棋方面的软件,有哪些可以让你成为他们必不可少的软件,而不是选择竞争对手或自己开发的因素?
So software is always an accumulating advantage, meaning that hardware, you redesign the whole thing every few years. And you see this with cars where they come out with an all new car where they pre-rese-rese-side everything. But software, you manage over very, very long periods of time. And so you have a cumulating advantage and you have in that you just add to it and add to it. And you replace the hardware underneath and that's okay. And that's how all the software you use has been developed since the beginning of time. So if you think about it, it's like you replace your iPhone every couple years, but all the software remains the same. And it just follows and you substitute that out.
因此,软件始终是一种累积优势,意味着硬件每隔几年就需要重新设计整个系统。就像汽车一样,它们推出全新的车型,重新设计了一切。但是软件的情况不同,你需要在非常非常长的时间内管理它。因此,你具有累积优势,只需不断添加和替换。你可以替换底层硬件,这也是可以接受的。这就是为什么自从软件产生以来,你使用的所有软件都是这样开发出来的。所以如果你思考一下,替换掉iPhone,但所有的软件保持不变。这只是一个平常的替代过程。
Now, the other thing about the economics of a software company is the more users you have, the more RDR. So you have this positive feedback loop. And I believe that the auto companies themselves, none of them have the scale necessary to do real worldwide development. And it will always be an external vendor that can have enough distribution through multiple car companies in order to feed the R&D to actually develop it. So some of it is, we're very good at running software projects, but there's a reason why in a lot of domains of software, you tend to have winner take all solutions because once you get that flywheel going, whoever has the most users has the biggest R&D budget has the most efficient sales. And that's especially true in AI products where you have the most data flow as well. But then in general, like let's talk about competitors.
现在,关于软件公司的经济学,一个用户越多,收入越高。因此,存在着这种正向反馈的循环。我相信汽车公司本身没有足够的规模来进行真正的全球开发。而始终会是外部供应商通过多个汽车公司的分销来推动研发的实际发展。所以其中一部分是,我们非常擅长运营软件项目,但在很多软件领域中,你往往会看到胜者通吃的解决方案,因为一旦这个正循环开始,拥有最多用户的公司就拥有最大的研发预算和最高效的销售。在人工智能产品中尤其如此,因为数据流最多。但总的来说,让我们谈谈竞争对手问题。
So the main competitor, the market leader right now is Mobileye. Mobileye has been oriented around selling a chip. And that's been their whole business. And if you look at their business model is how many chips can I sell? And they haven't yet transitioned to being able to sell software in order to and just letting the chip be a commodity. So we would rather let let Qualcomm, let in media, you know, they have other markets for their chips that's driving their R&D faster. And they're going to make their chip better faster than Mobileye will make it faster. Because they just have much, much larger distribution than hundreds of millions of units. And so what we're betting on is that platform will get better. And then our software, because that's our sole focus will get better as well.
所以目前主要的竞争对手,市场领导者是Mobileye。Mobileye的定位一直是销售芯片,这就是他们整个业务的核心。如果你看一下他们的商业模式,就是卖出多少个芯片。但他们还没有过渡到能够销售软件,让芯片成为一个普通商品。所以我们宁愿让高通、因特尔等其他公司来竞争,因为他们在芯片市场有其他的销售渠道,这推动了他们的研发速度。他们会比Mobileye更快地提升芯片质量。因为他们拥有数亿个单位的更大市场分销量。我们所押注的是这个平台会变得更好。而我们的软件,因为这是我们唯一关注的领域,也会变得更好。
Who do you think are from the car manufacturers? Who do you think is the likely winner here? As this as the scales as, you know, we think, I don't know, five years down the road, whatever, who is the winner here? Which you mean by winner? I think who gets the most who gets the most adoption and who benefits the most from this from the car manufacturers? It's always it's a very fragmented business to begin. Like, there is there any car manufacturer with more than 5% of any market? And I don't I I don't know exactly why that is. But I think what I'm seeing is the American car manufacturers are probably the most ahead. And then the Europeans, the Koreans are, you know, they're behind, but they're aggressively catching up. And you're talking about autonomy. Are you starting to want to must trip economy?
你认为谁来自汽车制造商?你认为这里可能的赢家是谁?就像我们所看到的,五年后,或者其他什么时候,谁是赢家?你所指的赢家是指什么?我认为是谁在汽车制造商中获得最多采纳并从中受益最多的人。一开始它总是一个非常碎片化的行业。像,有没有任何汽车制造商在任何市场上占据超过5%的份额?我不知道为什么会是这样。但是我认为我所看到的是,美国汽车制造商可能是最领先的。然后是欧洲和韩国,他们落后,但他们在积极迎头赶上。你说到自动驾驶。你是不是开始想要进行经济旅行了?
Yeah, autonomy. Just where they are in autonomy. And the Japanese car manufacturers are cut of nowhere. I the stalking horse in everyone's mind is where the Chinese manufacturer is going to do. Since they're going to be entering the market, maybe by the end of this decade, with electric vehicles that are much cheaper than anything that can be produced by a European or American manufacturer. And they're far ahead on software in comparison to those companies. So that's going to be the big rebalancing is like, how much market share are the Chinese companies with their large domestic market and large distribution going to take away from the incumbents?
是的,自主性。他们就是在自主性方面发展。而且日本汽车制造商突然冒出来了。我觉得大家在想的就是,中国制造商将会做什么。因为他们将会在这个十年末之前进入市场,推出比欧洲或美国制造商生产的电动汽车更便宜的产品。而且在软件方面,他们比这些公司更领先。所以,重要的平衡将是,中国公司凭借其庞大的国内市场和广泛的分销渠道,会从现有的企业那里获得多少市场份额。
On a on a bit of a different note, as you as you're talking about all this, you know, the technical details and I'm thinking about building this company again, what was it like for you through COVID, especially being, you know, car focused, like just hardware software? How did you how did you do that? You know, like remote work in person work? And you know, enterprise sales background, tell me a little bit about that.
在谈论这一切的时候,换个话题,特别是在你谈论这些技术细节的时候,我正在考虑重新建立这家公司,对于你来说,在COVID期间是什么样的经历呢?特别是考虑到你的专注于汽车领域,像硬件和软件。你是如何做到的呢?远程工作还是现场工作?以及你在企业销售方面的背景,能给我讲一点相关情况吗?
So we went a lot remote that are hard to team and our operations team never went remote because you're working on a physical car. It's like you can't put a car at everyone's house. We were probably remote until like middle of 2021. And then we started bringing it back. This is probably earlier than a lot of other companies. And remote was pretty difficult. Like it was not nearly as productive because it's not just that you're working on cars and that your product. The thing that saved us is at the time we were doing a ton of driving because we were still developing a lot of foundational technology. But the but having people co located is a huge accelerator for that. And so I think we slowed down a lot in 2020. And I think a lot of companies did. They slowed down a lot. And it wasn't until we really brought back people back in say 2021 that we started accelerating again. And we went from a complex system that just was not stabilized. Like we ground on that for like six months to what's the simplest, most brute force thing you can do to make it work. And so that's just happened over and over and over again. Where you don't know it's like, is this possible or not you're doing something hard? Are we just like not smart enough? Like is anyone able to do this? And you just have to back off and just try something that's a lot more obvious. And then you can get back into a learning cycle where you can figure out how to make it more complicated.
我们选择了一些很偏远的地方,这些地方很难组建团队,我们的运营团队从未远程办公,因为他们在进行实体汽车的工作。就好像你不能把汽车放在每个人家里一样。我们可能一直到2021年中期才开始恢复常态。这可能比其他很多公司早。远程工作相当困难。因为你不仅仅在工作于汽车和产品,降低了效率。挽救我们的因素是当时我们进行了大量的车辆驾驶,因为我们仍在开发许多基础技术。但是,人员集中在一起是一个巨大的加速器。所以我认为在2020年我们进展放慢了很多。我认为很多公司都是如此。他们速度明显放慢了。直到2021年我们真正地开始让人员回来,我们才重新加速。我们从一个未稳定的复杂系统,苦苦调试了六个月,才变为最简单、最粗暴的方法来使其正常运作。这种情况一再发生。你不知道这个问题是否可行,你在做一些困难的事情。我们是不是没那么聪明?还是有人能做到这一点?你只能退后一步,尝试一些更加明显的方法。然后你可以进入一个学习循环,找出如何使其变得更加复杂。
Yeah, I can see that makes sense. You mentioned earlier on the conversation, going back to just autonomy, you mentioned that you're you're targeting, you know, the mass market and like these are freeways expressways. Well, why is that like, is it do you think that you also mentioned that level five is so that this utopian vision? Is it really do you see a future where we are just, you know, entering these vehicles that move for us? No one said, you know, at the driver's seat or no one has their hand in the steering wheel. Is that not is that not going to happen?
是的,我明白这是有道理的。你在之前的对话中提到过,回到只谈自主技术,你提到你的目标是瞄准大众市场,像高速公路这样的道路。那为什么会这样呢?你还提到了五级自动驾驶是一种乌托邦的愿景。你真的认为未来我们会进入一种只是坐在这些为我们移动的车辆中的场景吗?没有人坐在驾驶座上,也没有人握着方向盘。这是不可能发生的吗?
So if you asked me a year ago, I would say definitely not. And the main reason why definitely not is because sooner or later, you have to communicate to a car what you want to do. And there are scenarios where it will be easier to just directly control the car than trying to explain to it or like click on something or like tell it where you want to go. It's like you go to a concert that's in a field and there's a person like waving a baton. Well, you're never going to teach a computer how to do that. So you're going to want to provide some sort of human instruction. What's changed is that in last year, you have large language models that come out that have actually made it a lot more practical to map sort of verbal, human intent into motion. And so I think you can get there, but it's not going to look exactly like how people see it right now. So right now you get your robot taxiing like you enter a destination and it's this very sort of scripted experience. And I don't think that that works for individually owned cars. Like you don't want to sit down in your car and then have to do a bunch of data entry before you get going. You want to sit down, you want to go. And then as you move along, you'll progress through different levels of autonomy.
所以如果你一年前问我,我肯定会说不可能。而绝对不可能的主要原因是,迟早你必须向汽车传达你想要做的事情。在某些情况下,直接控制汽车会更容易,而不是试图向它解释,或者点击一些东西,或者告诉它你想去哪里。就像你去一个在野外的音乐会,有一个人在挥舞指挥棒。呃,你永远不会教会一个计算机怎么做这个。所以你需要提供一些人类的指导。改变的是,去年出现了一些大型语言模型,实际上更加实用,可以将口头的人类意图转化为运动。所以我认为你可以做到这一点,但它不会像现在的人们所看到的那样。所以现在你的机器人出租车可以实现你输入目的地这样的操作,这是一种非常程式化的体验。我不认为这对于个人拥有的汽车是有效的。你不想坐下来,然后在出发前要进行大量的数据输入。你想坐下来,你就想开走。随着你的行动,你将逐步进入不同级别的自主驾驶。
So the way that we set it up is like if ever you like of the steering wheel, the car is driving. It doesn't have a very complex goal. Its goal is to like keep following the road and keep following the highway. But it's okay. The idea is that you're not displaying intent to drive. So the car fills that in. And you don't have to press a button and you don't have to set a speed. We just figure all that out from this surrounding traffic. And as a driver, as a driver, like I'm used to using adaptive cruise control in the highway, you know, it gets closer to the car, gets further from the car, breaks for me, simple. I love it.
所以我们设置的方法是,如果你不再握方向盘,车就会自动驾驶。它并没有非常复杂的目标,只是跟随道路和高速公路行驶。但没关系,重点是你不必表达出开车的意图,车会代替你完成。你不需要按按钮,也不需要设置速度。我们只需要根据周围的交通状况来解决这些问题。作为司机,我习惯在高速公路上使用自适应巡航控制,它会与前车保持距离,自动制动,简单又方便。
How as a driver, a consumer, how would this be different for me on the highway?
作为一名驾车者和消费者,在高速公路上,这对我的影响会有何不同?
So for one thing, you wouldn't even turn it on. It's always on. So that's one of our philosophies is that your car is always smarter. You don't have to press a button to say, please be smart now. The other thing is that you don't set a speed. We know what the speed is. We actually survey all of the surrounding traffic. So you know, adaptive cruise control tends to be you go set speed until you approach another car and then you're locked into all of the car. What we did is we surveyed the traffic and we figure out the following distance and the speed. And so the intention is that you can go through a wide variety of driving scenarios. Like I often it's like, oh, you have a fast car and then you're in stop and go. And there's traffic that's close.
首先,你甚至不需要打开它。它一直都是开着的。所以我们的一个理念就是你的汽车总是更智能。你不需要按下按钮说:“请变得聪明起来。”另外一点是你不需要设置速度。我们知道速度是多少。我们实际上会调查周围的车流情况。所以你知道,自适应巡航控制通常是你设置一个速度,直到你接近其他车辆,然后你就被锁定在了这个车辆上。而我们所做的是调查交通情况,然后确定跟随的距离和速度。因此,我们的意图是让你能够应对各种驾驶情景。就像经常会发生的情况是,你开着一辆快车,然后你陷入了车来车往的堵车中。
For us, our goal is that you don't make any adjustment through all of that. The car is figuring out how to adapt between those environments. So you're not changing the following distance. You're not changing the speed at all. And so we're getting people used to the idea that your car by default just does the right thing. If by default follows the road, if by default, we'll keep a reasonable following distance. All you have to do is nothing.
对我们来说,我们的目标是让你在整个过程中都不需要做任何调整。汽车正在努力适应这些环境之间的变化。因此,你无需改变跟车距离。你根本不用改变车速。因此,我们让人们习惯于这样一个想法,即你的车默认情况下会自动执行正确的动作。默认情况下会跟随道路行驶,如果需求,默认情况下会保持合理的跟车距离。你所需要做的就是什么都不用做。
And so instead of the car just slowing down unless you press a button, it's going to keep driving along the road and sort of the simplest thing. And then what you do is you layer navigation on top of that. You say, I want to drive to this exit or I want to or maybe actually enter a destination at which point the car becomes more active in terms of changing lanes as well. And that's the goal by the end of this year is to do that navigation layer on the freeway.
所以,车不再只是在你按下按钮之前减速,而是会继续沿着道路行驶并寻找最简单的路径。然后,你可以在此基础上进行导航。你可以说,我想驶向这个出口,或者实际上输入一个目的地,这样车辆在换道方面会更积极。这就是今年底的目标,实现高速公路上的导航层。
How far do you think from having the first ghost equipped vehicle on the streets and not through your company? Obviously through a manufacturer or consumer? It's going to be the second half of this decade. So after 2025, like maybe 2026 to 2028. So that is the development cycle of cars. They want to make these decisions like two years before they actually trip. So that's about the time range that we think is the target. And what you see right now is the auto manufacturers are locked into what they're doing at least through 2025. And then it's kind of open. And the conversations we have with them say it's like, okay, we're on this level two path level two plus plus, but we want something more, but we don't know what that is yet. And different companies have organized themselves in different ways to try and answer that question.
你认为离街头出现第一辆装备了幽灵技术的汽车还有多远,而不是通过你的公司,显然是通过制造商或消费者?可能是在本十年的下半期,所以在2025年之后,可能是2026年到2028年之间。这是汽车的开发周期。他们希望在实际交付的两年前做出这些决定。所以我们认为这是一个目标时间范围。你现在所看到的是汽车制造商至少在2025年之前都有固定计划,而之后还不确定。我们与他们的对话表明:我们正在二级技术道路上迈进,但我们希望有所突破,但我们还不知道具体是什么。不同的公司以不同的方式组织自己来回答这个问题。
Are you finding that certain manufacturers, whether and even splitting them up by domestic, Asian, European, are you finding that certain manufacturers are more receptive to conversations about this than others? Or is it just very, you know, just differs kind of brand by brand, not so much by where they're from in the world? It definitely matters where they're from in the world. So we're finding like the European manufacturers that do development in Silicon Valley are considerably more open than the ones that only do development in Germany. The American manufacturers, it's kind of the same thing where they're more open to the conversation. And what we found is that the better product they ship, the more reality they've been exposed to about how hard the next step is going to be, and the more open they are to a conversation.
你是否发现,某些制造商,无论是针对国内、亚洲还是欧洲市场,他们对于这个问题的讨论是否更具有积极性?或者说,这种情况是否更多地取决于品牌而不是制造商所属的国家?当然,所属国家确实有所影响。我们发现,在硅谷进行开发的欧洲制造商比仅在德国进行开发的制造商更加开放。美国的制造商也是类似的情况,他们更愿意进行讨论。我们发现,制造商们对于这个问题的讨论态度更加开放,这与他们所提供的产品质量有关,以及他们已经经历过下一步骤的艰难程度所带来的现实感。
Got it. So you're saying, as they're experiencing that, they're realizing that they need help solving this. So yeah, so you get this inverse behavior where the companies that haven't shipped any good features are usually harder to deal with because they haven't tried versus the company that shipped anything. It's like, you're talking like Ford and GM and Mercedes, who've done a lot of internal development, have a lot more experience with the problem. And so it's easier to have a productive conversation with them because they know where you actually run into roadblocks. And we can go directly to that part. Yeah, they felt the pain points. They're looking for the doctor.
明白了。所以你的意思是,当企业经历到这种情况时,他们意识到需要帮助来解决这个问题。所以是的,你会看到一种相反的行为,那些没有推出任何优秀功能的公司通常更难处理,因为他们没有试过,而那些已经发布过产品的公司则不同。就像你所说的,福特、通用和奔驰这样的公司,他们在内部开发方面积累了很多经验,对这个问题更有经验。因此,与他们进行富有成效的交流比较容易,因为他们知道你在哪些地方会遇到困难。我们可以直接谈论那些问题。是的,他们感受到了痛点,他们在寻找医生。
I can't believe we missed this question, but why the name ghosts? We got there. It's a simple one. It's a simple one. So the inspiration is like the call of the Japanese commie, which is like, okay, what animates your car? We're not building a car. We're just making your car have intent and want to go somewhere. And that's the ghost in the car, that animates it. Nice. I love it.
我简直不敢相信我们错过了这个问题,但是幽灵的名字是为什么呢?我们明明已经到那儿了。这个问题很简单。它很简单。所以灵感来自于日本传呼机的呼叫,这就像是问,你的汽车是由什么来驱动的?我们并不是在制造一辆汽车。我们只是让你的汽车有意义,并且渴望去某个地方。这就是车里的灵魂,它给它注入了生命。太好了,我喜欢。
Well, John, very eye opening in the world of autonomy. I think before we wrap up, I just give me a bit of macro. What is the self driving? Even for the, even if we're not self-driving, but what does the commute look like for the average person in like 2030? How is it different from today? The first thing that it'll be a lot less stressful. Like everyone, anyone who's done a long commute knows you get to the end of the commute and you're just tired from the commute. And so the first thing is like, if you can just, you know, sit back and like let the car ride itself, one thing I've noticed with my using the product myself is that usually I drive like probably like the top 5% in terms of speed, maybe the top 1%, depending on your measuring. And our car just goes slower because it's like kind of around the speed limit, but a little faster. And I find that I'm kind of okay with that. Like just because, you know, offloading into the to the car to do the driving just makes me feel a lot less stress.
嗯,约翰,自动驾驶领域真是大开眼界啊。在我们结束之前,我想简要了解一下大局。什么是无人驾驶呢?即使我们不使用自动驾驶,2030年普通人的通勤方式与今天有何不同?首先,通勤会减轻很多压力。每一个长途通勤的人都知道,到达目的地后都会因为通勤而感到疲惫。如果你能坐下来,让车自己行驶就好了。我也亲自使用了这个产品,发现我平时开车速度可能是最快的5%,也许是最快的1%,具体要看如何衡量。我们的车只是按照速度限制行驶,但还是稍微比限速快一点。我发现这对我来说也没问题,因为让车自动驾驶让我感觉少了很多压力。
We've done a lot of long test trips like all the way from recently we did San Francisco to Las Vegas and the stress has dropped off. And I think that one of the effects of that is going to be that people will be willing to commute a lot further than they would otherwise. And so we'll see if like eventually autonomy kicks off a new frontier of development where you can develop even further from workplaces because the commute is low stress enough that people will be willing to just go further in order to have more space and, you know, be in a community they prefer compared to where their job is. And on the tributus mentioned SF to Vegas, what was that like? Like what percentage of the time did you have to, you know, put your hands on the wheel? How did that work? It's almost never along that trip. The longest stretch we did without any intervention, I think it was about 120 miles. And the reason for that intervention is when you cross the border into Nevada, we have to change our license plate to go from a California registration to our Nevada registration.
我们已经做了很多长途测试旅行,比如最近我们从旧金山开往拉斯维加斯,压力减轻了。我认为其中一个影响将是人们愿意通勤的距离比以往更远。因此,我们会看到,自动驾驶可能会开启新的发展前景,使人们可以离工作地点更远地开发,因为通勤压力足够小,人们愿意去更远的地方以获得更多空间,并且生活在一个他们喜欢的社区,而不是他们工作的地方。在提到从旧金山到拉斯维加斯的行程时,那是什么感觉?你大概有多少时间需要将手放在方向盘上?是怎么运作的?那次旅行几乎没有需要干预的情况。我们最长的一段没有任何干预的行程大概是120英里。需要干预的原因是当你跨越内华达州的边界时,我们需要更换车牌,将车辆的注册从加利福尼亚转到内华达州来。
But you can now do very, very long stretches without any intervention. This is going, there's tons of trucks around there. The road itself is pretty flat, but at times it gets pretty patchy. But the system is very consistent through that. And so we're spending a lot of time, not just on the long trips, but around Detroit where the highways are extremely variable to fix all those cases. And they're under construction because it's summer. Make sure we handle all those cases. Our goal is that you just never intervene and it never asked you to intervene under any circumstances. And you know, the car changing lanes on its own and everything. Yeah. Yeah, change lanes. Right now you indicate you tell it to change lane, but it's checking the side and checking the rear and the lane change occurs. And then the next step by the end of the year is to make that completely automatic.
但现在你可以在很长的路段上不进行任何干预。这里有很多卡车。道路本身相当平坦,但有时会出现坑洼。但是系统在这些情况下非常稳定。因此,我们花了很多时间,不仅在长途旅行中,而且在底特律附近,那里的高速公路非常多变,我们正在修复所有这些情况。它们正在建设中,因为现在是夏天。确保我们处理掉所有这些情况。我们的目标是您永远不必干预,且无论在任何情况下都不会要求您干预。而且,你知道的,车辆自主变道等等。是的,变道。现在你只需要指示它改变车道,但它会检查侧面和后面,然后进行车道变更。到年底,我们计划实现完全自动化的变道。
Can this work in an environment where it's not perfect sunshine, cloudy, like, you know, there's a little rain, you know, something like that. Does that automatically kill it or what happens then? No, so cloudy is great. That's actually way easier than sunshine. So rain is fine. The main issue was the windshield wipers. The rain itself didn't seem to cause much disturbance on that on the lane markers, we've tested rain until like, you don't want to drive because the window is associated with water. The interesting things we're running into are glare. So think of a tunnel exit where there's like one bright light and then but the rest of the tunnel is dark. In Detroit, it's also dealing with salt spray is another issue because salt spray just creates lots of gunk on your windshield and on the lens by extension. And so those are the next challenges is to solve those cases. But you know, ordinary guy time light rain, you know, variable road conditions, glare on the road, we spent a lot of time this year solving for for glare where even people, I think, can't see where the lane markers are. Our goal is to be superhuman in all those respects.
在一个并非晴朗完美的环境中,比如多云、有一些雨,那会对它产生影响吗?自动驾驶系统是否会立刻失效?不会的,多云天气其实更容易应对。下雨也没问题,主要问题是雨刷。雨本身对车道标线的影响不大,我们测试过甚至下雨到你都不想开车的地步。有趣的问题是反光。可以想象一下隧道出口,有一个亮灯,其他地方都很黑暗。在底特律,还要应对盐雾问题,因为盐雾会在挡风玻璃上形成许多污垢,然后进一步影响感应器。所以下一步的挑战就是解决这些问题。不过,对于普通情况,像轻微的雨、不同的路况和道路反光,我们今年已经花了很多时间来优化。我们的目标是在所有这些方面都达到超人水平。
I love it. Yeah, you're going to need little windshield wipers on the on the sensors because I can tell you I'm driving sometimes and it's like, Oh, I don't have my lost my sensors, rain is, you know, crazy rain or maybe there's some salt or whatever, something on it. Yeah, so we have it easy. We just put it behind the windshield. So it's the same windshield wipers. So same sensor. Nothing, nothing is outside the car. Nothing's on the roof. It's all inside the car. So if you can see system can see, it's great.
我喜欢它。是的,你需要在传感器上安装小型雨刷,因为有时候我开车时会发现,哦,我的传感器丢失了,是因为下着大雨,或者可能有些盐或其他东西附着在上面。是的,所以对我们来说很容易,我们只需将其安装在挡风玻璃上。所以使用的是相同的挡风玻璃雨刷,同样的传感器。没有任何东西在汽车外部,也没有任何东西在车顶上。所有的东西都在车内,所以如果系统能够看得到,那太棒了。
John Hayes, John, where can people learn more about you ghost? Where should people go? If they want to learn more. So our website is ghostatonic.com myself. I'm on Twitter at ghosthays, H-A-Y-E-S.
约翰·海斯,约翰,请问人们该在哪里了解更多关于你的鬼魂?他们应该去哪里呢?如果他们想要了解更多。我们的网站是ghostatonic.com,而我个人则在Twitter上使用用户名ghosthays,H-A-Y-E-S。
John Hayes, thank you so much. This is very fascinating. I'm going to be I'm going to be paying close attention where this goes. So I want to see, you know, I'm really curious to see how this evolves and I'm rooting for him. And I want to I want to see these on every car in the street. So this is going to be fun.
约翰·海斯,非常感谢你。这非常有趣。我将会密切关注此事的发展。因此,我很好奇想看到它的进展,我支持他。我希望在街上每辆车都能使用这个。所以这会很有趣的。
All right. Thank you. Appreciate it. Thank you.
好的,谢谢。非常感谢。谢谢。
All right. Hope you enjoyed that episode. Please give the podcast a rating. Consider subscribing to the show and check the show notes for links to what we talked about. Thanks for tuning in. I'll see you guys next time.
好的。希望你们喜欢那一集。请给这个播客一个评分。考虑订阅该节目,并查看节目笔记中我们讨论的链接。感谢你们收听。下次见。