发布会·中文精读

Tesla 自动驾驶日

Tesla Autonomy Day

B 站版本为网友搬运,时间点可能差几秒
中文精读由 AI 基于原始视频与文稿改写,非官方翻译,可能存在偏差或错误;表述以原视频为准。说明

2019 年 4 月 22 日,特斯拉在帕洛阿尔托总部办了一场"自动驾驶日"。马斯克上台时迟到了,先道歉,然后说这是公司第一次办这种活动,希望以后能办得勤一点,让外界跟上进展。

开场那位产品负责人讲了一句很实在的话:过去两年,所有关于特斯拉的讨论都绕着 Model 3 的产能打转,没人注意到这家公司在后台干了什么。他们做了一颗芯片,把整车的视觉神经网络从头重写了一遍。

那天摊开的东西,是三年的工作量。

先把数字放在这儿。特斯拉自研的 FSD 计算机,双芯片合计 144 TOPS。英伟达当时最好的车规级自动驾驶芯片 Xavier,单颗 21 TOPS。

七倍。而且这些芯片不在实验室里,它们正在下线的每一台 Model S、Model X、Model 3 上装着。

马斯克说这话的时候语气很平:

一开始这听起来不太可能。特斯拉一家从来没设计过芯片的公司,怎么会做出世界上最好的芯片?但客观上这件事已经发生了。不是领先一点点,是领先很多。(马斯克)

一颗没人要求过的芯片

2016 年,市面上找不到一颗从零开始为神经网络设计的推理芯片。所有人都在做同一件事:给 CPU、GPU、DSP 加几条指令,让它们跑神经网络时快一点。没有人原生地做这件事。

2016 年 2 月,Pete Bannon 入职特斯拉。他之前做了 35 年芯片,其中 12 年在 PA Semi(后来被苹果收购),参与过几十种架构,是 iPhone 5 的主设计师之一。他到岗后问马斯克:你愿不愿意花掉做全定制系统设计需要的所有钱?

马斯克反问:我们会赢吗?

Pete 说:当然。

马斯克说:那我加入。

这就是整个项目的起点。没有需求文档,没有竞品分析,就两句话。

三年从零到量产

Pete 在台上把时间线摆了出来,密度很高:

  • 2016 年 2 月,招到第一批人,开始想一颗全自动驾驶专用芯片长什么样
  • 18 个月做设计,2017 年 8 月交付制造
  • 2017 年 12 月拿回芯片,上电,第一次就跑通了
  • 2018 年 4 月,改了几处,出了 B0 版本
  • 2018 年 7 月,芯片通过认证
  • 2018 年 12 月,开始量产质量级别的零件
  • 2019 年 3 月,装进 Model S 和 Model X
  • 2019 年 4 月,Model 3 开始生产

从招第一个人到三款车全部量产,三年多一点。Pete 说这是他职业生涯里最快的系统开发项目,原因是垂直整合:电源设计、信号完整性分析、封装设计、系统软件、固件、板级设计、系统验证,这些团队特斯拉本来就有,可以直接拉过来用,不用外购,也不用排期。

目标也很清楚,就三条:功耗压在 100 瓦以下,这样新计算机能塞进老车里做后装改装;零件成本要低,低到可以装两套做安全冗余;神经网络性能至少 50 TOPS。

最后一条超额完成。两颗芯片加起来 144 TOPS。

批大小为 1,这是安全设计不是性能设计

Pete 解释了一个容易被忽略的选择。

谷歌的 TPU v1 批大小是 256。意思是它要攒够 256 张图才开始算,攒不够就等着。这在数据中心里没问题,反正请求永远排队。

车上不行。

特斯拉把批大小设成 1:一张图进来,立刻算。这样延迟最小。Pete 的说法是,最小化延迟就是最大化安全。一辆时速 100 公里的车,每多等 100 毫秒就多跑 2.8 米。

他们还做了一个当时看起来冒险的赌注:GPU 只放了很弱的一颗。因为 2016 年的时候,神经网络输出之后还有大量后处理要跑在 GPU 上,但他们判断随着网络变强,后处理会越来越少。

这个赌赢了。Pete 说后处理确实一路下降,那颗弱 GPU 现在够用。

为什么是 8 位整数

芯片工作在热约束环境里,功耗预算卡得很死。所以每一个算术单元的选择都要算账。

他们选了整数加法,能耗是同等浮点加法的九分之一。选了 8 位乘 8 位整数乘法,功耗远低于其他乘法方案,精度也够用。

内存更狠。Pete 给了一个数字:数据跑到片外 DRAM 上,能耗大约是访问片上 SRAM 的 100 倍。

所以整个架构围绕一件事展开:把中间结果留在片上。神经网络每一层的输出只被下一层用一次,用完就扔。既然是一次性的,为什么要写回 DRAM?

每个神经网络加速器配了 32MB SRAM。两颗芯片,四个加速器,总共 128MB 的片上暂存。

Pete 还引了伯克利 Mark Horowitz 的一篇论文:在普通整数 CPU 上执行一条指令,加法本身只占 0.15% 的功耗,剩下 99.85% 全是控制开销和记账。所以自研加速器的思路很直接,把这些控制开销砍掉,只留算术。

两套完整的计算机,坏一套车照开

板上不是一颗芯片,是两套完全独立的计算机。各自有 SoC、DRAM、闪存,各自启动,各自跑自己的操作系统。

摄像头分成两组,一半接蓝色电源,一半接绿色电源。两套计算机同时接收全部八路视频,独立处理,独立算出一份行驶计划,然后互相比对。两份计划一致才执行。

任何一部分挂掉,车继续开。摄像头挂一半,继续开。电源回路挂一路,继续开。一颗芯片彻底死掉,继续开。

马斯克补了一句判断标准:

这套计算机发生故障的概率,显著低于一个人失去意识的概率。这是关键指标,至少低一个数量级。(马斯克)

车上还有一层冗余:2016 年 10 月起,所有特斯拉都配了双电机转向,任何一个电机失效还能转向;所有电源线和数据线都有备份,剪断任意一根车照开;主电池包完全断电,辅助电源还能支撑转向和刹车。

马斯克说,从硬件角度,这些车从 2016 年 10 月起就是按 Robotaxi 的标准设计的。2016 年 10 月之前生产的车,他们不打算升级,因为改造成本比造一台新车还高。

72 瓦、80% 的成本、21 倍的帧率

实测数据来自在路上跑的车,跑完整的 Autopilot 软件栈:功耗 72 瓦,其中大约 15 瓦用于跑神经网络。目标是 100 瓦以下,达成了。

硅成本是上一套方案的 80%。也就是说换用自研方案之后,每台车在这块还省钱了。

性能对比更能说明问题。拿那个 350 亿次运算的窄视场摄像头网络来跑:

  • 用芯片上的 12 个 CPU 跑,1.5 帧每秒,完全不够开车
  • 用那颗 GPU 跑,17 帧每秒,还是不够
  • 用神经网络加速器跑,2100 帧每秒

Pete 的原话是"白天和黑夜的区别"。他说自己做过这么多项目,性能提升超过三倍的都没见过。

换成整机对比:同样的网络,老硬件上跑到 110 帧每秒,新计算机用上四个加速器,2300 帧每秒。21 倍。

再往上叠一层:英伟达 Xavier 单芯片 21 TOPS,FSD 计算机双芯片 144 TOPS,七倍。Pete 提醒听众,如果 Xavier 要达到同样算力,功耗会按同比例上涨,散热会变成大问题,而散热直接影响续航。

神经网络很笨,笨到你无法想象

Andrej Karpathy 接手了下一段。他在斯坦福读的博士,主攻图像识别和深度学习,在斯坦福开了一门卷积神经网络课,第一年 150 人,三年后涨到 700 人,是斯坦福最大的课之一。马斯克介绍他的时候说,他是世界上最好的计算机视觉研究者之一,可能没有"之一"。

Andrej 开场先泼了盆冷水。

他放了四张图:大提琴、鬣蜥、剪刀。人一眼就能认出来。计算机看到的是什么?一张几百万个数字的网格,每个数字是一个像素的亮度值。

从这堆数字到"鬣蜥"这个概念,中间隔着一整个视觉皮层。而鬣蜥可以有很多姿态、很多光照条件、很多背景、很多裁剪方式,网络必须对所有这些变化都稳健。

然后他讲了训练过程。网络初始化时权重完全随机,预测也完全随机,可能看着鬣蜥说这是船。训练就是不断告诉它"这是鬣蜥",通过反向传播和随机梯度下降,把几百万个连接各调一点点,让"鬣蜥"的概率从 0.1% 涨到 14%,再涨上去。这个过程重复几百万次。

接下来是他整场演讲的核心论点。

他放了一张日本狆的照片。人说一句"哦,日本狆长这样",再看几张其他狗的照片,就能挑出哪几只是日本狆。一个样本够了。

计算机不行。你得给它一千张日本狆,不同姿态、不同亮度、不同背景、不同裁剪,它才能抓住那个模式。

这个反差决定了自动驾驶的技术路线。因为路上没有"日本狆"这种标签,只有车道线、障碍物、可行驶区域。而这些东西没有教科书,只能靠数据喂出来。

数据引擎:把"我不知道我不知道"变成可训练的样本

Andrej 描述了一个循环,他管它叫 data engine。

先用一批随机采样的种子数据,人工标注,训练,部署到车上。然后建一套触发机制,专门捕捉车上的失误:网络自己觉得不确定、驾驶员接管、系统检测到异常,这些情况自动把数据传回来。

传回来的数据进入单元测试,确认问题存在。接着向车队要更多类似场景的数据,标注,并进训练集,重训,重新部署,再循环。

他举了个具体例子。如果只在高速直道上采随机数据,网络会学到一个错误的世界模型:车道线永远是直的。碰到弯道,它会预测出一条直的车道线。Andrej 说这不能怪网络,它不知道左边那棵树重不重要,不知道右边那辆车重不重要,它从零开始,唯一能告诉它"什么重要"的工具就是标注数据。

他还讲了一个更细的例子。一辆车后面挂着自行车。早期网络会输出两个检测框,一个车,一个自行车。严格说都对,但下游的规划器不想处理"自行车会跟着车跑"这件事。真实世界里那就是一辆车。于是团队用机器学习的方法向车队要了六张类似的图,全部标注成一辆车,重训,问题解决。

他强调了一句:所有用于训练的数据都是匿名的,从第一天就按隐私要求设计的。

五十万张加塞照片,没有一个标注员

人工标注贵,也慢。Andrej 讲了两个绕开人工标注的项目。

第一个是加塞检测。

做法不是写代码判断"左转向灯亮了吗""这个框在水平移动吗"。而是让车队在检测到有车从右侧车道进入中间车道、或从左侧进入中间车道时,自动回传数据。然后把时间倒放,就能自动得到标注:这辆车在 1.3 秒后会插到前面。

一次能拿到大约 50 万张图,全部自动标注好,没有一个人工参与。训练完之后先不启用,跑影子模式:网络一直预测"我认为这辆车要加塞",然后比对真实结果,把误报和漏报都传回来,再进训练集。反复几轮,等误报漏报比例满意了,才真正把控制权交给这个网络。

Andrej 说这个功能大约三个月前就推给用户了。如果车主感觉车对加塞的反应变好了,那就是车队学习在规模上跑起来的效果。

你开车的时候就在标注数据

第二个项目是路径预测。

Andrej 的说法很妙:你开车的时候,本身就在标注数据。你转方向盘,就是在告诉系统该怎么通过这个环境。

车队回传全部八路摄像头的视频,配上 GPS、IMU、方向盘角度、轮速,就能还原出这个人在这段路上走过的精确轨迹。这条轨迹就是监督信号。用这些轨迹训练网络,网络就学会从画面直接预测路径。

这在学术上叫模仿学习:拿真实世界里人类开车的轨迹,让网络去模仿。

演示视频里,车开到一个看不见弯道后面的路口,网络预测出的绿色路径直接拐了过去。马斯克在台下插了一句,说网络在预测它根本看不见的东西,而且置信度极高。

Andrej 补充,这个预测是三维的,只是渲染成二维投到画面上。系统知道地面的坡度,这对驾驶极其重要。

这套东西已经上线了。五个月前,车还过不了高速上的苜蓿叶匝道,现在能过了。今天演示的穿过路口,很大一部分也来自这套自动标注训练出的预测。

视觉怎么测距:人没有从眼睛里发射激光

Andrej 转到激光雷达的话题,先讲了个常识。

在座各位今天都是开车来的,用的是神经网络加视觉,没有人从眼睛里往外打激光,但都安全到了。

人脑从视觉里提取深度,至少用了三条线索。第一是双目视差,两只眼睛在同一时刻给出两个独立观测,大脑做三角测量。第二是结构来自运动,很多动物眼睛长在两侧,视野重叠很少,它们靠晃动头部获得多个观测点,同样能三角测量。第三是单目线索,就算闭一只眼完全不动,人对深度仍有感觉,因为场景里有大量强单目线索,比如透视和消失点。

他放了一张经典视错觉图:两根蓝条其实一样长,但大脑根据画面的消失线自动判定其中一根更长。人工神经网络也吃这套。

然后他给了三种从纯视觉估计深度的工程方法。

第一种是多视角立体视觉。一段六秒的行车视频,用八个摄像头的画面做三维重建,能还原出整条街的几何结构。这套方法在车上也用了简化版。

第二种是用传感器给视觉打标签。前向雷达测出的距离,用来标注视觉网络输出的框。人不用告诉网络"这辆车 25 米远",雷达直接告诉它。训练够了,网络就能自己从画面里推断距离。演示里雷达点(圆圈)和视觉框(立方体)在俯视图上基本重合,说明网络学会了不同车型的真实尺寸,并能据此反推距离。

第三种叫自监督。只喂原始视频,不给任何标签,让网络预测每一帧的深度,约束条件是这些预测在时间上必须自洽。想自洽,唯一的办法就是预测对。

Andrej 的结论很硬:视觉识别是自动驾驶的必要条件,不是加分项。激光雷达点云的信息密度远低于图像。他举了个例子:路边那个东西是塑料袋还是轮胎?激光雷达只会给出几个点,视觉能分辨。骑车的人回头看了一眼,他是要并进来还是继续直行?施工区的牌子上写的是什么?所有这些问题的答案只存在于视觉里。

他的原话是,整个道路基础设施就是为人类视觉消费而设计的。所有标志、所有红绿灯,都是给人眼看的。

所以他把激光雷达称作捷径:绕过了视觉识别这个根本问题,给出一种虚假的进展感,最终是一根拐杖。它能做出很快的演示。

长尾:九的马拉松

Andrej 用一张图收尾。

自动驾驶要做的是处理 99.99% 的情况,剩下那几个九最难。他放了一串图:先是正常行驶的车,然后造型奇怪的车,然后全黑的车,然后车驮着车,然后翻倒的车,然后腾空的车。

这些场景来自车队,出现的频率比任何竞争对手都高。而解决这些问题的速度,正比于你在真实世界里遇到它们的频率。

这就是他给出的特斯拉优势的全部论证。

影子模式:让新算法先闭嘴开车

Stuart Bowers 讲软件工程。他在 Facebook 做过广告基础设施的机器学习平台,在 Snap 做过大规模竞价系统,现在负责把硬件、视觉、规划控制、测试、内核补丁、持续集成、仿真捏成一个能上车的产品。

他先给了一个数字:导航辅助驾驶累计跑了 7.7 亿英里。

然后是流程。任何新算法先上影子模式,在路上跑但不控制车,看它在真实场景里会怎么做。他放了一段演示:一辆车快速超车,系统在仿真环境里画出几条候选路径,和人类实际开出的路径叠在一起比对。

调好之后进入受控部署,也就是早期访问计划,几千个愿意给反馈的用户,在闭环里真开。有人接管就回传片段,团队把片段放回开环重放,问一个问题:我们现在的软件,离人类的行为是更近了还是更远了。

他们还把四个 FSD 计算机装进自建机架,组成集群跑这套数据基础设施。整个软件栈都是自己写的,从内核补丁到图像信号处理器调参,所以能拿到更准的数据,迭代更快。

正式铺开之前,他们让每个用户确认车辆行为,按一下确认键。靠这个攒了 900 万次成功确认的变道,零事故。这个月他们取消了确认步骤,现在高速上每天有超过 10 万次自动变道。

他还提到下一代的自动紧急制动:不只是为当前挡在路上的人刹车,而是预测哪些人会进入你的路径。这套东西现在跑影子模式,当季推给车队。

一辆车一周跑 60 小时,账就变了

马斯克最后上台,讲的是钱。

先摆现状。美国一辆汽油车的平均使用成本,按 AAA 的数据是每英里 62 美分。网约车是每英里 2 到 3 美元。特斯拉测算的 Robotaxi 运营成本是每英里不到 18 美分,而且还会降。

再看资产利用率。普通人一周开车 10 到 12 小时,一天一两个小时。一辆能自动驾驶的车,一周能跑 55 到 60 小时。一周总共 168 小时。

同样的钢铁和电池,产出翻了五倍。马斯克说,如果你把这个看成一次系统升级,把汽车的效用提升五倍,那是经济效率的巨大跃迁。

单车经济:一台 Model 3 Robotaxi 的成本低于 3.8 万美元,长期可能降到 2.5 万美元以下。年毛利大约 3 万美元,按 11 年算超过 30 万美元。他特意说明这个测算是保守的,假设只有 50% 的行驶里程载客。 单台 Robotaxi 的净现值,他给的数量级是"几十万美元"。

商业模式上,车主可以把自己的车加进或移出特斯拉网络,特斯拉抽成 25% 到 30%。马斯克把它比作 Uber 加 Airbnb 的混合体。车不够的地方,特斯拉自己投放车辆。

他还改了一条租赁条款:Model 3 租期结束后没有购买选项,车必须还回来。因为还回来的车可以直接进车队。

为什么别人复制不了?马斯克的答案有两个。一是全栈整合:整车设计制造、自研计算机硬件、自研芯片、自研 AI 软件,全在自己手里。二是车队规模:当时路上有 42.5 万台配 Hardware 2 以上的车,八摄像头、前向雷达、十二个超声波;当季末会到 50 万台;一年后超过 100 万台配 FSD 计算机。

他打了个比方,说这就像谷歌搜索引擎的优势:用户越多,查询和结果就越是在替谷歌编程。

时间表:今年功能完整,明年睡觉

马斯克把自动驾驶分成三级。

第一级是功能完整,车能在所有场景下自己开,但人还得盯着。他的说法是今年做到。

第二级是从特斯拉自己的判断出发,认为人不需要再摸方向盘、不需要再看窗外。他的估计是明年第二季度左右。

第三级是说服监管机构承认这件事。他估计明年年底,至少在部分司法管辖区。

Robotaxi 的时间表:明年有第一批车上路,车里没人,在部分辖区运营。

他对监管的判断是,监管者最终是被数据说服的。数据足够多、足够清楚地证明自动驾驶更安全,他们会听,可能会花时间消化,但最后会得出正确结论。

他给了个安全数据:目前 Autopilot 的安全性是普通驾驶员的约两倍,而且会继续拉开。

高精地图和地理围栏都是拐杖

有人问高精地图和限定区域运营。马斯克的回答没有余地。

高精地图是个错误。你要么需要高精地图,那么环境一变车就废了;要么不需要高精地图,那你花在画地图上的时间就是浪费。激光雷达和高精地图是两根拐杖,回头看,它们显然是错的、愚蠢的。(马斯克)

地理围栏同理。他的原话是,如果你需要地理围栏来限定区域,你就没有真正的自动驾驶。

他还讲了 GPS 的定位。GPS 误差会变,道路实际情况也会变,可能今天在修路绕行。GPS 可以当辅助,帮你更自信地开过自己熟悉的街区,但绝对不能做主信号。一旦 GPS 成为主信号,就是在自找麻烦。

关于激光雷达,他补充了一段区分。SpaceX 的龙飞船对接空间站用的是激光雷达,那是他自己推动从零开发的,因为那个场景里激光雷达合理:没有现成的视觉基础设施,距离远、目标明确、成本不敏感。但在车上,它是昂贵且无用的。

他还给了个物理层面的理由:如果你要主动发光测距,别用可见光波段,因为被动光学已经把可见光的信息全拿到了。要用就用能穿透遮挡的波段,比如 20.8 毫米的雷达波,而不是 400 到 700 纳米的可见光。所以车上配的是前向雷达加十二个超声波,再加八个摄像头。

真正的瓶颈是电池和产能

有人问电池供应会不会成为瓶颈。马斯克承认会是约束,并给出了一个反直觉的销售策略:故意把销量往标准续航版本上引。

原因是能量密度。标准续航版大约 50 千瓦时,长续航版大约 75 千瓦时。同样的电池产能,做小电池包能多造三分之一的整车。而在自动驾驶时代,最重要的指标是路上跑着多少台能自己赚钱的车,不是每台车能跑多远。

百万英里电池怎么做到?他算了笔简单账:250 英里续航的电池包,跑 100 万英里需要 4000 次循环。这个循环寿命,特斯拉的储能产品 Powerpack 早就做到了。

产能那边,2013 年生产 2.3 万台,2018 年约 25 万台,五年十倍。他预计未来五六年还会有类似的增长。全球现在有约 20 亿辆汽车和卡车,年产能约 1 亿台,全部换成自动驾驶电动车需要很长时间。他强调制造被严重低估了:很多人以为设计对了就能瞬间造出全世界需要的量,这是不成立的。

他还提到,去年为了让 Model 3 上量,把超级工厂 2 号里所有 2170 电池产线都改成了车用电池。储能业务长期在捡剩下的电芯,因为储能不需要车规级的认证和安全冗余。

那些问答里藏着的判断

整场活动有大量问答,几个回答比主演讲更有信息量。

有人问下一代芯片。Pete 说至少比现在这套好三倍,大约两年后,而且下一代已经做了一半。

有人问芯片谁代工。答案是三星。

有人问知识产权怎么保护。Pete 说申请了大约十二项专利,但底层是线性代数,这东西很难专利化。真正的护城河是权重,权重是加密的。马斯克补了一句:如果有人能破解,我很想见见这个人,我会立刻雇他。更硬的护城河是车队:那些权重是靠几十亿英里真实驾驶数据不断更新出来的,别人没有这个车队。

有人问能不能把停着的车当算力用。Andrej 提到特斯拉有个叫 Dojo 的训练计算机项目,专门用来吞海量视频做无监督训练,那是另一场发布会的内容。至于车本身,它是推理优化过的机器。

有人问数据量有多大。Andrej 的回答是,规模不是重点,多样性才是。如果全是高速直道,多了没用。所以触发机制做得非常精细,只取此刻真正需要的那部分数据。

有人问仿真能不能替代真实里程。Andrej 说仿真他们用得很重,开发和评估都在用,也成功用于训练。但训练数据没有真实世界的替代品,仿真搞不定外观物理,也搞不定周围所有智能体的行为。他举了几个例子:雪地、风中的树、施工区里互相影响的行人小孩动物、狗、异域动物,还有卡车驮着卡车驮着卡车这种你根本想不出来的场景。

马斯克接过话:

搞仿真,本质上是在给自己批改作业。你知道你要仿真什么,所以你肯定能解出来。但你不知道你不知道什么。世界非常怪异,有几百万个长尾案例。如果有人能做出一个精确匹配现实的自动驾驶仿真,那本身就是人类能力的巨大成就。他们做不到。(马斯克)

有人问加塞和并线这种人类博弈怎么处理。Andrej 说长期看必须靠车队学习:什么场景下人会并线,什么场景下人觉得不安全不并线,用这些数据训练分类器,分类器写的"代码"比人写的更准。马斯克补充说会有个比"疯狂麦克斯模式"更激进的设置,也许叫"洛杉矶交通模式",但那是用户自己选的,默认还是保守。

有人问如果自己买了十台 Model 3 跑 Robotaxi 行不行。马斯克说可以,但只能通过特斯拉网络,因为几年前就在车里写了这条限制。他还说这事会很麻烦,你试试就知道了。

有人问事故责任。如果车在自动驾驶状态下撞了人,谁负责?马斯克的回答只有两个字:特斯拉。

把这套逻辑拿去检查你手上的项目

这场发布会信息密度很高,但真正能迁移的东西不多,我挑几条能直接用的。

  • 判断一个 AI 项目能不能成,先看它的数据回流闭环。不是看模型结构,不是看团队履历,是看它有没有机制在真实场景里发现自己的错误,把错误变成训练样本,再部署回去。这个循环转一圈要多久,决定了项目天花板。
  • 算一下专用方案和通用方案的账。英伟达做通用芯片,服务所有客户;特斯拉只服务自己,可以把批大小设成 1、把 GPU 砍到最弱、把 99.85% 的控制开销删掉。如果你的需求量足够大,专用方案的优势通常是数量级的,不是百分比的。
  • 看推理芯片,先看能效比,再看峰值算力。72 瓦跑完整个自动驾驶栈,靠的是把中间数据全留在片上。片外 DRAM 的能耗是片上的 100 倍,这条账在任何边缘设备上都成立。
  • 别拿仿真当真实数据的替代品。仿真适合做回归测试和评估,不适合当训练数据的主要来源,因为你在用自己已知的世界观训练自己。真实数据的价值在于它会给你想不到的东西。
  • 冗余要在架构层设计,不能事后加装。特斯拉 2016 年 10 月就把双电机转向、双电源线、双数据线全铺进车里,两年半后才真正用上。等到需要的时候再改,成本比造新车还高。
  • 如果你在做需要长期积累的产品,想清楚你的"车队"是什么。特斯拉的车队是几百万台带八个摄像头的车。你的产品里,那个会随使用量增长、别人复制不了的资产是什么?如果没有,那护城河就只是时间差。

最后补一句马斯克在问答里说的话。有人问他怎么看那些押注激光雷达的竞争对手,他说:他们会全部放弃激光雷达,这是我的预测,记住我的话。

这话是 2019 年 4 月说的。剩下的,交给时间。

YouTube 平台字幕(自动或人工),可能有识别错误,人名与数字请以视频为准。已按语句合并成段落,并去掉 um、uh 等口头填充词。 文稿来源:YouTube(自动/人工字幕)。

0:01[Music] [Applause] [Music] [Applause] [Music] [Applause] [Music]

20:48[Applause] [Music] [Applause] [Music] you [Music] [Applause]

31:55[Music] [Applause] [Music] [Applause] [Music] you [Music] [Applause]

48:13[Music] [Applause] [Music] you [Music] [Applause] [Music] [Applause]

1:04:30[Music] hi everyone I'm sorry for being late welcome to our very first analyst day for autonomy I really hope that this is something we can do a little bit more regularly now to keep you posted about the development we're doing with regards to autonomous driving about three months ago we were getting prepped up for our q4 earnings call with Elon and quite a few other executives and one of the things that I told the group is that from all the conversations that I keep having with investors on a regular basis that the biggest gap that I see with what I see inside the company and what the outside perception is is our ability of autonomous driving and it kind of

1:10:20makes sense because for the past couple of years we've been really talking about model three ramp and you know a lot of the debate has revolved around model three but in reality a lot of things have been happening in the background we've been working on the new force of you having chip we've had a complete overhaul of our neural net for vision recognition etc so now that we finally started to produce our full self-driving computer we thought it's a good idea to just open the veil invite everyone in and talk about everything that we've been doing for the past two years so about three years ago we wanted to use we wanted to find the best possible chip

1:10:58for full autonomy and we found out that there's no chip that's been designed from ground up for neural nets so we invited my colleague Pete Bannon the VP of silicon engineering to design such chip for us he's got about 35 years of experience of building chips and designing chips about 12 of those years where for a company called PA semi which was later acquired by Apple so he worked on dozens of different architectures and designs and he was the lead designer I think for Apple iPhone 5 but just before joining Tesla and he's gonna be joined on the stage by Illamasqua thank you [Applause]

1:11:45actually I was gonna introduce Pete but ones done so it's just the the best trip and System Architect that that I know in the world and and it's honor to have you and your and your team at Tesla and we'll take away just tell him I think how to work at unity never done thanks Eli it's a pleasure to be here this morning and a real treat really to tell you about all the work that my colleagues and I've been doing here at Tesla for the last three years I think

1:12:20will tell you a little bit about how the whole thing got started and then I'll introduce you to the full self-driving computer and tell you a little bit about how it works we'll dive into the chip itself and go through some of those details I'll describe how the custom neural network accelerator that we design works and then I'll show you some results and hopefully it will all still be awake by then I was hired in February

1:12:41of 2016 I asked Elon if he was willing to spend all the money it takes to do full custom system design and he said well are we gonna win and I said well yeah of course so he said I'm in and so that got us started we hired a bunch of people and started thinking about what a full what a custom-designed chip for full autonomy would look like we spent eighteen months doing the design and in August of 2017 we released the design for manufacturing we got it back in December it powered up and it actually worked very very well in the first try we made a few changes and released a B zero Rev in April of 2018 in July of 2018 the chip was qualified and we

1:13:21started full production of production quality parts in December of 2018 we had the autonomous driving stack running on the new hardware and we were able to start retrofitting employee cars and testing the hardware and software out in the real world just last March we started shipping of the new computer in the Model S and X and just earlier in April we started production in the model 3 so this whole program from the hiring of the first few employees to having it in full production in all three of our cars is just a little over three years and it's probably the fastest system development program I've ever been associated with and it really speaks a lot to the advantages of having

1:14:06a tremendous amount of vertical integration to allow you to do concurrent engineering and speed up deployment in terms of goals we were totally focuses exclusively on Tesla requirements and that makes life a lot easier if you have one and only one customer you don't have to worry about anything else one of those goals was to keep the power under 100 watt so we could retrofit the new machine into the existing cars we also wanted a lower

1:14:31part cost so we could enable for redundancy for safety at the time we had a thumb in the wind I submitted that it would take at least 50 trillion operation was a second of neural network performance to drive a car and so we wanted to get at least that much and really as much as we possibly could batch sizes how many items you operate on at the same time so for example Google's TPU one has a batch size of 256 and you have to wait around until you have 256 things to process before you can get started we didn't want to do that so we designed our machine with a batch size of one so as soon as an image shows up we process it immediately to minimize latency which maximizes safety

1:15:08we needed a GPU to run some post-processing at the time we were doing quite a lot of that but we speculated that over time the amount of post-processing on the GPU would decline as the neural networks got better and better and that has actually come to pass so we took a risk by putting a fairly modest GPU in the design as you'll see and that turned out to be a good bet security super important if you don't have a secure car you can't have a safe car so there's a lot of focus on security and then of course safety in terms of actually doing the chip design as ela and alluded earlier there was really no ground-up neural network accelerator in existence

1:15:48in 2016 everybody out there was adding instructions to their CPU or GPU or DSP to make it better for inference but nobody was really just doing it natively so we set out to do that ourselves and then for other components on the chip we purchased industry standard IP for CPUs and GPUs that allowed us to minimize the design time and also the risk to the program another thing that was a little unexpected when I first arrived was our ability to leverage existing teams at Tesla Tesla had wonderful power supply design teams signal integrity analysis package design system software firmware board designs and a really good system validation program that we were able to

1:16:32take advantage of to accelerate this program here's what it looks like over there on the right you see all the connectors for the video that comes in from our cameras that are in the car you can see the two self-driving computers in the middle of the board and then on the left is the power supply and some control connections and so I really love it when a solution is boiled down to its barest elements you have video computing and power and and it's straightforward

1:17:00and simple here's the original hardware 2.5 enclosure that the computer went into and we've been shipping for the last two years here's the new design for the FSD computer it's basically the same and that of course is driven by the constraints of having a retrofit program for the cars I'd like to point out that this is actually a pretty small computer it fits behind the glove box between the glove box and the firewall in the car it does not take up half your trunk as I

1:17:29said earlier there's two fully independent computers on the board you can see them they're highlighted in blue and green to either side of the large SOC you can see the DRAM chips for that we use for storage and then below left you see the flash chips that represent the file system so these are two independent computers that boot up and run their own operating system yeah if I

1:17:51can add something that's the general principle here is that if any part of this could fail and the call will keep driving so you have cameras fail you could have power circuits fail you could have one of the Tesla pulse rifle self-driving computer chips fail car keeps driving the probability of this computer failing is substantially lower than somebody losing consciousness that's that's the key metric at least an order of magnitude yep so one of the things that we additional thing we do to keep the machine going is to have redundant power supplies in the car so one one machines running on one just power supply and the other ones on the other the cameras are the same so

1:18:32half of the cameras run on the blue power supply the other half around the green power supply and both chips receive all of the video and process it independently so in terms of driving the car the basic sequence is collect lots of information from the world around you not only do we have cameras we also have radar GPS maps the I M use ultrasonic sensors around the car we have wheel ticks steering angle we know what the acceleration and deceleration of the car is supposed to be all of that gets integrated together to form a plan once we have a plan the two machines exchange their independent version of the plan to make sure it's the same and assuming

1:19:14that we agree we then act and drive the car now once you've driven the car with some new control you have what costs want to validate it so we validate that what we transmitted it was what we intend to transmit to the other actuators in the car and then you can use the sensor suite to make sure that it happens so if you ask the car to accelerate or brake or steer right or left you can look at the accelerometers and make sure that you are in fact doing that so there's a tremendous amount of redundancy and overlap in both our data acquisition and our data monitoring capabilities here moving on to talk about the full self-driving chip a little bit it's

1:19:49packaged in a thirty seven point five millimeter BGA with sixteen hundred balls most of those are used for powering ground but plenty for signal as well if you take the lid off it looks like this you can see the package substrate and you can see the dye sitting in the center there if you take the dye off and flip it over it looks like this there's 13,000 C four bumps scattered across the top of the dye and then under net underneath that are twelve metal layers and if you which is obscuring all the details of the design so if you strip that off it looks like this this is a 14 nanometer FinFET CMOS process it's 260 millimeters in size which is a modest-sized iso for

1:20:30comparison typical cell phone chip is about a hundred millimeters square which so we're quite a bit bigger than that but a high end GPU would be more like six hundred eight hundred millimeter square so so we're sort of in the middle I would call it the sweet spot it's it's a comfortable size to build there's 250 million logic gates on there and a total of six billion transistors which even even though I work on this all the time that's mind boggling to me the chip is manufactured and tested to AEC q100 standards which is a standard automotive criteria next I'd like to just walk around the chip and explain all the different pieces to it and I'm sort of

1:21:10gonna go in the order that a pixel coming in from the camera would visit all the different pieces so up there in the top left you can see the camera cellular interface we can ingest 2.5 billion pixels per second which is more than enough to cover all the sensors that we know about we have an on-chip network that distributes data from the memory system so the pixels would travel across the network to the memory controllers on the right and left edges of the chip we use industry standard LPD ddr4 memory running at four hundred four thousand two hundred and sixty six gigabits per second which gives us a peak bandwidth the sixty eight gigabytes a second which is a pretty healthy

1:21:44bandwidth but again this is not like ridiculous so we're sort of trying to stay in the comfortable sweet spot for cost reasons the image signal processor has a 24-bit internal pipeline that allows us to do take full advantage the HDR sensors that we have around the car it does advance tone mapping which helps to bring out details and shadows and then it has advanced noise reduction which just improves your overall quality of the images that we're using in the neural network the neural network accelerator itself there's two of them on the chip they each have 32 megabytes of SRAM to hold temporary results and minimize the amount of data that we have

1:22:20to transmit on and off the chip which helps reduce power each array has a 96 by 96 multiply add array with in place accumulation which allows us to do 10,000 multiply ads per cycle there's dedicated riilu Hardware dedicated pooling hardware and the each of these deliver 306 excuse me each one delivers 36 trillion operations per second and they operate at two gigahertz the two of them together on a die who delivers 72 trillion operations a second so we exceeded our goal of 50 tariffs by a fair bit there's also a

1:22:59video encoder we encode video and use it in a very variety of places in the car including the backup camera display there's optionally a user feature for dash camp and also for a clip logging data to the cloud which Stewart and Andre will talk about more later there's a GPU on the chip it's modest performance it has a support for both 32 and 16 bit floating point and then we have 12 a 72 64-bit CPUs for a general-purpose processing they operate at 2.2 gigahertz and this represents about two and a half times the performance available in the current solution there's a safety system that

1:23:37contains two CPUs that operate in lockstep this system is the final arbiter of whether it's safe to actually drive the actuators in the car so this is where the two plans come together and we decide whether it's safe or not to move forward and lastly there's a safety system and then basically the job of the safety system is to ensure that this chip only runs software that's been cryptographically signed by Tesla if it's not been signed by Tesla then the chip does not operate now I've told you

1:24:08a lot of different performance numbers and I thought it'd be helpful maybe to put it into perspective a little bit so throughout this talk I'm going to talk about a neural network from our narrow camera it uses 35 Giga three five billion operations 35 Giga apps and if we used all 12 CPUs to process that network we could do one-and-a-half frames per second which is super slow I'm not nearly adequate to drive the car if we use the 600 gigaflop GPU the same

1:24:35network we'd get 17 frames per second which is still not good enough to drive the car with a cameras the neural network accelerators on the chip can deliver 21 frames per second and you can see from the scaling as we moved along that the amount of computing in the CPU and GPU are basically insignificant to what's available in the neural network accelerator it's it really is night and day so moving on to talk about the neural network accelerator we're just gonna stop for some water on the Left there's

1:25:10a cartoon of a neural network just to give you an idea what's going on the data comes in at the top and visits each of the boxes and the data flows along the arrows to the different boxes the boxes are typically convolutions or d convolutions with real ooze the green boxes are pooling layers and the

1:25:30important thing about this is that the data produced by one box is then consumed by the next box and then you don't need it anymore you can throw it away so all of that temporary data that gets created and destroyed as you flow through the network there's no need to store that off chip and DRAM so we keep all that data in SRAM and I'll explain why that's super important in a few minutes if you look over on the right side of this you can see that in this network of the 35 billion operations almost all of them are convolution which is based on dot products the rest are deconvolution also based on dot product and then riilu and pooling which are relatively simple

1:26:08operations so if you were designing some hardware you'd clearly target doing dot products which are based on multiply ad and really kill that but imagine that you sped it up by a factor of 10,000 so 100% all of a sudden turns into 0.1% 0.01 percent and suddenly the riilu and pooling operations are going to be quite significant so our hardware doesn't our hardware design includes dedicated resources for processing riilu and pooling as well now this chip is

1:26:41operating in a thermally constrained environment so we had to be very careful about how we burn that power we want to maximize the amount of arithmetic we can do so we picked integer add it's 9 times less energy than a corresponding floating point add and we picked 8 bit by a bit integer multiply which is significantly less power than other multiply operations and is probably enough accuracy to get good results in terms of memory we chose to use SRAM as much as possible and you can see there that going off chip to DRAM is approximately a hundred times more expensive in terms of energy consumption than using local SRAM so clearly we want to use the local SRAM as much as

1:27:25possible in terms of control this is data that was published in a paper by Mark Horowitz at is SCC where he sort of critiqued how much power it takes to execute a single instruction on a regular integer CPU and you can see that the add operation is only 0.15 cent percent of the total power all the rest of the power is control overhead and bookkeeping so in our design we set to basically get rid of all that as much as possible because what we're really interested in is arithmetic so here's the design that we finished you can see that it's dominated by the 32 megabytes of SRAM there's big banks on the left and right and in the center bottom and

1:28:05then all the computing is done in the upper middle every single clock we read 256 bytes of activation data out of the SRAM array 128 bytes of weight data out of the SRAM array and we combined it in a in a 96 by 96 mole at array which performs 9000 multiply ads per clock at 2 gigahertz that's a total of 3.6 336

1:28:32Point air at 8 Tara ops now when we're done with a dot product we unload the engine so that we shift the data out across the dedicated riilu unit optionally across a pooling unit and then finally into a write buffer where all the results get aggregated up and then we write out 128 bytes per cycle back into the SRAM and this whole thing cycles along all the time continuously so we're doing dot products while we're unloading previously results doing pooling and writing back into the memory if you add it all up to your Hertz you need one terabyte per second of Estrin man was to support all that work and and so the harder supplies that so one terabyte per second a bandwidth per

1:29:15engine there's two on the chip two terabytes per second the chip has the accelerator has a relatively small instruction set we have a DMA read operation to bring data in from memory we have a DMA write operation to push results back out to memory we have three dot product based instructions convolution deconvolution inner product and then two relatively simple a scale is a one input one output operation and ly is this two inputs and one output and then of course stop when you're done we

1:29:48had to develop a neural network compiler for this so we take the neural network that's been trained by our vision team as it would be deployed in the older cars and when you take that and compile it for use on the new accelerator the compiler does layer fusion which allows us to maximize the computing each time we read data out of the SRAM and put it back it also does some smoothing so that the the demands on the memory system aren't too lumpy and then we also do channel channel padding to reduce bang conflicts and we do Bank aware SRAM allocation and this is a case where we could have put more hardware in the design to handle Bank conflicts but by

1:30:28pushing it into software we save Hardware in power at the cost of some software complexity we also automatically insert DMAs into the graph so that data arrives just in time for computing without having to stall the machine and then at the end we generate all the code we generate all the weight data we compress it and we add a CRC check sum for reliability to run a program all the neural network descriptions our programs are loaded into SRAM at the start and then they sit there ready to go all the time so to run a network you have to program the address of the input buffer which presumably is a new image that just arrived from a camera you set the output

1:31:11buffer address you set the pointer to the network weights and then you set go and then the Machine goes off and will sequence through the entire neural network all by itself usually running for a million or two million cycles and then when it's done you get an interrupt and can post-process the results so moving on to

1:31:30results we had a goal to stay under 100 watts this is measured data from cars driving around running the full autopilot stack and we're dissipating 72 watts which is a little bit more power than the previous design but with the dramatic improvement in performance it's still a pretty good answer of that 72 Watts about 15 watts is being consumed running the neural networks in terms of cost the silicon cost of this solution is about 80% of what we were paying before so we are saving money by switching to this solution and in terms of performance we took the narrow camera neural network which I've been talking about that has 35 billion operations in

1:32:12it we ran it on the old hardware as in a loop as quick as possible and we delivered a hundred and ten frames per second we took the same data the same Network compile it for hardware for the new FST computer and using all four accelerators we can get 2,300 frames per second processed so a factor of 21 I think this is perhaps the most significant slide it's night and day I've never worked on a project where the performance increase was more than three so this was pretty fun if you compare it

1:32:49to say in videos drives the AVR solution a single chip delivers 21 ter ops our full self-driving computer with two chips is 144 tear offs so to conclude I think we've created a design that delivers outstanding performance 144 tear ups for a neural network processing has outstanding power performance we managed to jam all of that performance into the thermal budget that we had it enables a fully redundant computing solution it has a modest cost and really the important thing is that this FSD computer will enable a new level of safety and autonomy in Tesla's vehicles without impacting their cost or range something that I think we're all looking forward to when we do Q&A after

1:33:40each segment so if people have questions about the hardware they can ask right now the reason I would I ask Pete to do just a detailed far more detailed and perhaps most people would appreciate dive into the Tesla full self-driving computer is because it at first it seems improbable how could it be that Tesla who has never designed a chip before would design the best chip in the world but that is objectively what has occurred not not best by a small margin best by a huge margin it's in the cars right now all Tesla's being produced right now have this computer we switched over from the Nvidia solution for SMX about a month ago I switched over model

1:34:35three about ten days ago all cars being produced have the have all the hardware necessary compute and otherwise for full self-driving I'll say that a game old Tesla cars being produced right now have everything necessary for full self-driving all you need to do is improve the software and later today you will drive the cars with the development version of the improved software and you will see for yourselves questions repeat

1:35:20a trip two or three global equities research very very impressive in every shape and form I was wondering like I've I took some notes you are using activation function Arielle you the rectified linear unit but if we think about the deep neural networks it has multiple layers and some algorithms may use different activation functions for different hidden layers like soft Max or tan h do you have flexibility for

1:35:53incorporating different activation functions rather than Lu in your platform then I have a follow-up yes we do we have informations of tan edge and sigmoid for example one last question like in the nanometers you mentioned 14 nanometers as I was wondering wouldn't make sense to come to at a lower maybe 10 nanometers two years down or maybe seven at the time we started the design not all the IP that we wanted to purchase was available in ten nanometers we had to finish the design in 14 it's

1:36:29maybe worth pointing out that we finished this design like maybe one and a half two years ago and began design if the next generation we're not talking about the next generation today but we're about halfway through it that will all the things that are obvious for a next-generation chip we're doing

1:36:57oh hi you talked about the software as the piece now you did a great job I was blown away understood ten percent of what you said but I trust that it's in good hands Thanks so it feels like you got the hardware pieces done and that was really hard to do and now you have to do the software piece now maybe that's outside of your expertise how should we think about that software piece what can ask for better introduction to so Andre and Stuart I think yeah are there any fun dating questions for the chip part before the next part of the presentation is neural nets and software so maybe I'm the chip side the last slide was 144 trillions of operations

1:37:44per second versus was it Nvidia 21 that's right and maybe can you just contextualize that for a finance person why that's so significant that gap thank you well I mean it's a factor of seven and performance Delta so that means you can do seven times as many frames you can run neural networks that are seven times larger and more sophisticated so it's a it's a very big currency that you can spend on on lots of interesting things to make the car better I think that savior of power usage is higher than ours Xavier powers I don't know

1:38:19that it's like the best my knowledge the power requirements would increase at least to the same degree of factor of seven and and costs would also increase by a factor seven so yeah I mean how power is a real problem because it also reduces range so it has the helper power is very high and then you have to get rid of that power by the thermal problem becomes really significant because you had to go to get rid of all that power

1:38:56thank you very much I think we have you know a lot of ask the questions if you guys don't mind the day of running but long just we're gonna do that the drive demos afterwards so if you've got if you if you if anybody needs to pop out and do drive demos a little sooner you're welcome to do that I do want to make sure we answer your questions yep Pradeep Romani from UBS Intel and

1:39:27AMD to some extent have started moving towards a chip lab based architecture I did not notice a chaplet based design here do you think that looking forward that would be something that might be of interest to you guys from an architecture standpoint a chip based architecture yes we're not currently considering anything like that I think that's mostly useful when you need to use different styles of technology so if you want to integrate silicon germanium or DRAM technology on the same silicon substrate that gets pretty interesting but until the die size gets obnoxious I

1:40:04wouldn't go there the strategy here in it this is the started you know basically three little over three years ago where's design build a computer that is fully optimized and aiming for full self-driving then writes software that is designed to work specifically on that computer and get the most out of that computer so you have tailored to hardware that is that is a master of one trade self-driving the in-video is a

1:40:35great company but they have many customers and so when as they as they apply their resources they need to do a generalized solution we care about one thing self-driving so that it was designed to do that incredibly well the software is also designed to run on that hardware incredibly well and the combination of the software and the hardware I think is unbeatable

1:41:05I the chip is designed to cross this video input in case you use let's say lidar would it be able to process that as well or is that is it primarily for video I explained to you today is that lidar is is a fool's errand and and anyone luck relying on orb lidar is doomed doomed expensive expensive

1:41:34sensors that on are unnecessary it's like having a whole bunch of a expensive appendices that compact one appendix is bad well now don't put a whole bunch of them that's ridiculous you'll see so just two questions on just on the power consumption is there way to maybe give us like a rule of thumb one you know every watt is reduces range by certain percent or a certain amount just so we can get a sense of how much a model three the the target consumption is 250 watts per mile

1:42:21it depends on the nature of the driving as to how many miles that effect in city it would have a much bigger effect than on highway so you know if you're driving for an hour in a city and you had a solution hypothetically that you know was it was it was a kilowatt you'd lose four miles on a model three so if you're

1:42:48only going say 12 miles an hour then that's like there were to be a 25 second impact in range in city it's basically powers of the power that the the power of the system has a massive impact on city range which is where we think of most most of the Robo taxi market will be so the power is extremely important

1:43:17I'm sorry thank you what's the primary design objective of the next-generation ship we don't want to talk too much about the next-generation ship but it's it'll be at least let's say three times better than the current system about two years away is is the

1:43:52chip being mainly you you don't mean you facture the chip you contract that out and how much cost reduction does that save in the overall vehicle cost the the 20% cost reduction I cited was the the piece cost per vehicle reduction not that wasn't a development cost I was just the actual yeah I'm saying but like if I'm manufacturing these in mass is this saving money in doing it yourself yes a little bit I mean most chips are

1:44:21made for most people don't make chips with there aren't five it's very unusual I think you don't see any supply issues without getting the chip mass-produce the cost saving pays for the development I mean the basic strategy going to Elon was we're gonna build this chip it's gonna reduce the costs Anil on said hmm times a million cars a year deal that's

1:44:48correct yes sorry if they're really chips specific questions we can answer them others there will be a Q&A opportunity after after Andre talks and and after Stuart talks so there will be two other Q&A opportunities this is very chip specific then I'll be here all afternoon yeah and

1:45:10exactly if people be here at the end as well so good REO Thanks that died photo you had there's the neural processor takes up quite a bit of the die I'm curious is that your own design or there's some external IP there yes that was the custom design for by Tesla and then I guess the follow-on would be there's probably a fair amount of opportunity to reduce that footprint as you tweak the design it's actually quite

1:45:39dense so in terms of reducing it I don't think so it'll will greatly enhance the functional capabilities in the next generation okay and then last question can you share where you're you're fapping this part what what where are we oh is Samsung yes thank you grant Tanaka

1:46:09Tanaka Kapil just curious how defensible your chip technologies and design is from it from a IP point of view and hoping that you won't won't be offering a lot of the IP the outside for free Thanks we have filed on the order of a dozen patents on this technology fundamentally it's linear algebra which I don't think you can patent ah not sure I think if

1:46:40somebody started today and they're really good they might have something like what we have right now in three years at but in two years we'll have something something three times better talking about the intellectual property protection you have the best intellectual property and some people just steal it for the fun of it I was wondering if we look at a few interactions with Aurora that companies to industry believes they stole your intellectual property I think the key ingredient that you need to protect is the weights that associate to various parameters do you think your chip can do something to prevent anybody maybe encrypt all the weights so that

1:47:30even you don't know what the weights are at the chip level so that your intellectual property remains inside it and nobody knows about it and nobody can just feel it man I'd like to meet the person that could do that because they were I would hire them in a heartbeat yeah so every a hard problem yeah I mean

1:47:52we do encrypt the the it's it's a hard trip to crack so if they can crack it's very good if they give any crack it and then also also figure out the software and the neural net system and everything else they can design it from scratch like that's that's all it's our intention to prevent people from stealing all that stuff and we if they do we hope it at least takes a long time it will definitely take them a long time yeah I mean I felt like if we were if it was alcohol to do that how would we do it be very difficult but the thing that's I think a very powerful sustainable advantage for us is the fleet nobody has the fleet those weights are consummating updated and improved

1:48:31based on billions of miles driven Tesla has a hundred times more cars with the full self-driving Hardware than everyone else combined you know we we have by the end of this quarter we'll have 500,000 cars with the full eight camera set up twelve ultrasonics someone will still be on Hardware two but we still have the data gathering ability and then by a year from now we'll have over a million cars with full self-driving computer hardware everything yeah

1:49:09should we have fun it's just a massive data advantage it's similar to like you know how like the Google search engine has a massive advantage because people use it and people the people are programming effectively program google with the queries and the results yeah just pressing on that and please reframe the questions I'm tackling and if it's appropriate but you know when we talked to Wayne Moe or Nvidia they do speak with equivalent conviction about their leadership because of their competence in simulating miles driven can you talk about the advantage of having real-world miles versus simulated miles because I think they expressed that you know by the time you get a

1:49:55million miles they can simulate a billion and no Formula One racecar driver for example could ever successfully complete a real-world track without driving in a simulator can you talk about the advantages it sounds like the that you perceived to have associated with having data ingestion coming from real-world miles versus simulated miles absolutely the simulator we have a quite a good simulation too but it just does not capture the long tail of weird things that happen in the real world if the simulation fully captured the real world well I mean that would be proof that we're living in a simulation I think it doesn't I wish but

1:50:41it simulations do not capture the real world they don't the real world is really weird and messy you need the you need the cars in the road and we actually get it get into that in Andre and Stuart's presentation yeah so okay when we move on to 200 the last question was actually a very good Segway because one thing to remember about our F is the computer is that it can run much more complex neural nets for much more precise image recognition and to talk to you about how we actually get that image data and how we analyze them we have our senior director of AI Andre karpati who's gonna explain all of that to you Andre has a PhD from Stanford University

1:51:32where he studied computer science focusing on education recognition and deep learning Andre why don't you just talk do your own intro if there's a lot of PhDs from Stanford that's not important yes yeah we don't care come on Thank You Andre started the computer vision class at Stanford that's much more significant that's what matters just it so if can please talk about your background in a way that is not bashful

1:52:01just tell what about the SEC redun yeah and then sure yeah so yeah I think I've been training neural networks basically for what is now a decade and these neural networks were not actually really used in the industry until maybe five or six years ago so it's been some time that I've been trained these neural networks and that included you know institutions at Stanford at at opening I at Google and really just training a lot of neural networks not just for images but also for natural language and designing architectures that couple those two modalities for for my PhD so

1:52:33the computer computer science class oh yeah and at Stanford actually taught the convolutional neural Oryx class and so I was the primary instructor for that class I actually started the course and designed the entire curriculum so in the beginning it was about 150 students and then it grew to 700 students over the next two or three years so it's a very popular class it's one of the largest classes at Stanford right now so that was also really successful I mean on Dre's like really one of the best computer vision people in the world arguably the best okay thank you yeah so

1:53:02hello everyone so Pete told you all about the chip that we've designed that runs neural networks in the car my team is responsible for training of these neural networks and that includes all of data collection from the fleet neural network training and then some of the deployment onto that so what do then you know that works

1:53:21do exactly in the car so what we are seeing here is a stream of videos from across the vehicle across the car these are eight cameras that send us videos and then these neural networks are looking at those videos and are processing them and making predictions about what they're seeing and so the some of the things that we're interested in there's some of the things you're seeing on this visualization here our lane line markings other objects the distances to those objects what we call drawable space shown in blue which is where the car is allowed to go and a lot of other predictions like traffic lights traffic signs and so on now for my talk

1:53:57I will talk roughly into in three stages so first I'm going to give you a short primer on neural networks and how they work and how they're trained and I need to do this because I need to explain in the second part why it is such a big deal that we have the fleet and why it's so important and why it's a key enabling factor to really training this you know networks and making them work effectively on the roads and in the third stage I'll talk about a vision and lidar and how we can estimate depth just from vision alone so the core problem that these networks are solving in the car is that a visual recognition so four unites these are very this is a very

1:54:27simple problem you can look at all of these four images and you can see that they contain a cello about an iguana or scissors so this is very simple and effortless for us this is not the case for computers and the reason for that is that these images are to a computer really just a massive grid of pixels and at each pixel you have the brightness value at that point and so instead of just seeing an image a computer really gets a million numbers in a grid that tell you the brightness values at all the positions the maker knows if you will it really is the matrix yeah and so

1:55:01we have to go from that grid of pixels and brightness values and to high-level concepts like iguana and so on and as you might imagine this iguana has a certain pattern of brightness values but iguanas actually can take on many appearances so they can be in many different appearances different poses and different brightness conditions against the different backgrounds you can have a different crops of that iguana and so we have to be robust across all those conditions and we have to understand that all those different brightness palette patterns actually correspond to a goannas now the reason you and I are very good at this we have a massive neural network inside

1:55:33our heads there's processing those images so light hits the retina travels to the back of your brain to the visual cortex and the original cortex consists of many neurons that are wired together and that are doing all the pattern recognition on top of those images and really over the last I would say about five years the state-of-the-art approaches to processing images using computers have also started to use neural networks but in this case artificial neural networks but these artificial neural networks and this is just a cartoon diagram of it are a very rough mathematical approximation to your visual cortex we'll really do have neurons and they are connected together

1:56:11and here I'm only showing three or four neurons and three or four in four layers but a typical neural network will have tens to hundreds of millions of neurons and each neuron will have a thousand connections so these are really large pieces of almost simulated tissue and then what we can do is we can take those neural networks and we can show them images so for example I can feed my iguana into this neural network and the network will make predictions about what it's seen now in the beginning these neural networks are initialized completely randomly so the connection strengths between all those different neurons are completely random and therefore the predictions of that

1:56:43network are also going to be completely random so it might think that you're actually looking at a boat right now and it's very unlikely that this is actually an iguana and during the training during a training process really what we're doing is we know that that's actually in iguana we have a label so what we're doing is we're basically saying we'd like the probability of iguana to be larger for this image and the probability of all the other things to go down and then there's a mathematical process called back propagation a stochastic gradient descent that allows us to back propagate that signal through those connections and update every one

1:57:13of those connections and update every one of those connections just a little amount and once the update is complete the probability of iguana for this image will go up a little bit so it might become 14% and the probably of the other things will go down and of course we don't just do this for this single image we actually have entire large data sets that are labeled so we have lots of images typically you might have millions of images thousands of labels or something like that and you are doing forward backward passes over and over again so you're showing the computer here's an image it has opinion and then you're saying this is the correct answer and it Tunes itself a

1:57:45little bit you repeat this millions of times and you sometimes you show images the same image to the computer you know hundreds of times as well so the network training typically will take on the order of few hours or a few days depending on how big of a network you're training and that's the process of training a neural network now there's something very unintuitive about the way neural networks work that I have to really get into and that is that they really do require a lot of these examples and they really do start from scratch they know nothing and it's really hard to wrap your head around with around this so as an example here's a cute dog and you probably may not know

1:58:18the breed of this dog but the correct answer is that this is a Japanese spaniel now all of us are looking at this and we're seeing Japanese spaniel more like okay I got it I understand kind of what this Japanese spaniel looks like and if I show you a few more images of other dogs you can probably pick out other Japanese spaniels here so in particular those three look like a Japanese spaniel and the other ones do not so you can do this very quickly and you need one example but computers do not work like this they actually need a ton of data of Japanese spaniels so this is a grid of Japanese spaniels showing them in a thousand of examples showing them in different poses

1:58:50different brightness conditions different backgrounds different crops you really need to teach the computer from all the different angles what this Japanese spaniel looks like and it really requires all that data to get that to work otherwise the computer can't pick up on that pattern automatically so with us all this imply about the setting of self-driving of course we don't care about dog breeds too much maybe we will at some point but for now we really care about Ling line markings objects where they are where we can drive and so on so the way we do this is we don't have labels like iguana for images but we do have images from the fleet like this and we're interested

1:59:21in for example in line markings so we a human typically goes into an image and using a mouse annotates the ling line markings so here's an example of an annotation that a human could create a label for this image and it's saying that that's what you should be seeing in this image these are the Ling line markings and then what we can do is we can go to the fleet and we can ask for more images from the fleet and if you ask the fleet if you just do an Evo of this and you just ask for images at random the fleet might respond with images like this typically going forward on some highway this is what you might just get like a random collection like

1:59:53this and we would annotate all that data if you're not careful and you only annotate a random distribution of this data your network will kind of pick up on this this random distribution on data and work only in that regime so if you show it slightly different example for example here is an image that actually the road road is curving and it is a bit of a more residential neighborhood then if you show the neural network this image that network might make a prediction that is incorrect it might say that okay well I've seen lots of times on highways lanes just go forward so here's a possible prediction and of course this is very incorrect but the

2:00:28neural network really can't be blamed it does not know that the Train on the the tree on the left whether or not it matters or not it does not know if the car on the right matters for not towards the lane line it does not know that the that the buildings in the background matter or not it really starts completely from scratch and you and I know that the truth is that none of those things matter what actually matters is that there are a few white lane line markings over there and in a vanishing point and the fact that they curl a little bit should pull the prediction except there's no mechanism by which we can just tell the neural network hey those linge line markings

2:00:59actually matter the only tool in the toolbox that we have is labelled data so what we do is mean to take images like this when the network fails and we need to label them correctly so in this case we will turn the lane to the right and then we need to feed lots of images of this to the neural net and neural that over time will humilate will basically pick up on this pattern that those things there don't matter but those leg line markings do and we learn to predict to the correct lane so what's really critical is not just the scale of the data set we don't just want millions of images we actually need to do a really good job of covering the possible space

2:01:31of things that the car might encounter on the roads so we need to teach the computer how to handle scenarios where it's light and wet you have all these different specular reflections and as you might imagine the brightness patterns and these images will look very different we have to teach a computer how to deal with shadows how to deal with Forks in the road how to deal with large objects that might be taking up most of that image how to deal with tunnels or how to do with construction sites and in all these cases there's no again explicit mechanism to tell the network what to do we only have massive amounts of data we want to source all those images and we want to annotate the

2:02:04correct lines and the network will pick up on the patterns of those now large and varied datasets basically make these networks work very well this is not just a finding for us here at Tesla this is a ubiquitous finding across the entire industry so experiments and research from Google from facebook from Baidu from alphabets deepmind all show similar plots where neural networks really love data and love scale and variety as you add more data these neural networks start to work better and get higher accuracies for free so more data just makes them work better now a number of companies have a number of people have kind of pointed out that potentially we

2:02:44could use simulation to actually achieve the scale of the data sets and we're in charge of a lot of the conditions here and maybe you can achieve some variety in a simulator now at Tesla and that was also kind of brought up in the question questions just before this now at Tesla this is actually a screenshot of our own simulator we use simulation extensively who use it to develop and evaluate the software we've also even used it for training quite successfully so but really when it comes to training data from your networks there really is no substitute for real data the simulator simulations have a lot of trouble with modeling appearance physics and the

2:03:19behaviors of all the agents around you so there are some examples to really try that point across the real world really throws a lot of crazy stuff at you so in this case for example we have very complicated environments with snow with trees with wind we have various visual artifacts that are hard to simulate potentially we have complicated construction sites bushes and plastic bags that can go and that can kind of go around with the wind complicated construction sites that might feature lots of people kids animals all mixed in and simulating how those things interact and flow through this construction zone might actually be completing completely intractable it's

2:03:55not about the movement of any one pedestrian in there it's about how they respond to each other and how those cars will respond to each other and how they respond to you driving in that setting and all of those are actually really tricky to simulate it's almost like you have to solve the self-driving problem to just simulate other cars in your simulation so it's really complicated so we have dogs exotic animals and in some cases it's not even that you can't simulate it is that you can't even come up with it so for example I know that you can have truck on truck on truck like that but in the real world you find this and you find lots of other

2:04:28things that are very hard to really even come up with so really the variety that I'm seeing in the data coming from the fleet is just crazy with respect to what we have in a simulator we have a really good simulator simulation you're fundamentally a grain you're grading your own homework so you you know you if you know that you're gonna simulate it okay you can definitely solve for it but as Andre is saying you don't know what you don't know the world is very weird and has millions of corner cases and if

2:05:02you somebody can produce a self-driving simulation that accurately matches reality that in itself would be in a monumental achievement of human capability they can't there's no way yeah yeah so I think the three points are I really tried to drive home until now are to get neural networks to work well you require these three essentials you require a large data set a very data set and a real data set and if you have those capabilities you can actually train you all that works and make them work very well and so why is Tesla is such a unique and interesting position to really get all these three essentials right and the answer to that of course

2:05:41is the fleet we can really source data from it and make our neural network systems work extremely well so let me take you through a concrete example of for example making the object detector work better to give you a sense of how we develop these in all that works how we iterate on them and how we actually get them to work overtime so object detection is something we care a lot about we'd like to put bounding boxes around say the cars and the objects here because we need to track them and we need to understand how they might move around so again we might ask human annotators to give us some annotations for these and humans might go in and my

2:06:12tell you that ok those patterns over there are cars and bicycles and so on and you can train your neural network on this but if you're not careful the neural network all will make mispredictions in some cases so as an example if we stumble by a car like this that has a bike on the back of it then the neural network actually went when I joined would actually create two detections it would create a car detection and a bicycle detection and that's actually kind of correct because I guess both of those objects actually just but for the purposes of the controller in a planner downstream you really don't want to deal with the fact that this bicycle can go with the car

2:06:43the truth is that that bike is attached to that car so in terms of like just objects on the road there's a single object a single car and so what you'd like to do now is you'd like to just potentially annotate lots of those images as this is just a single car so the process that we that we go through internally in the team is that we take this image or a few images that show this pattern and we have a mechanism a machine learning mechanism by which we can ask the fleet to source us examples that look like that and the fleet might respond with images that contains those patterns so as an example these six images might come from the fleet they

2:07:17all contain bikes on backs of cars and we would go in and we would annotate all those as just a single car and then the the performance of that detector actually improves and the network internally understands that hey when the bike is just attached to the car that's actually just a single car and it can learn that given enough examples and that's how we've sort of fixed that problem I will mention that I talked quite a bit about sourcing data from the fleet I just want to make a quick point that we've designed this from the beginning with privacy in mind and all the data that we use for training is anonymized now the fleet doesn't just respond with bicycles on backs of cars

2:07:50we look for all the thing we'll look for lots of things all the time so for example we look for boats and the fleet can respond with boats we look from construction sites and the fleet can send us lots of construction sites from across the world we look for even slightly more rare cases so for example finding debris on the road is pretty important to us so these are examples of images that have streamed to us from the fleet that show tires cones plastic bags and things like that if we can source these at scale we can annotate them correctly and then your network will learn how to deal with them in the world here's another example animals of course

2:08:22also a very rare occurrence an event but we wanted the neural network to really understand what's going on here that these are animals and we want to deal with that correctly so to summarize the process by which we iterate on neural network predictions looks something like this we start with a seed data set that was potentially sourced at random we annotate that data set and then we train your lab works on that data set and put that in the car and then we have mechanisms by which we notice inaccuracies in the car when this detector may be behaving so for example if we detected the neural network might be uncertain or if we detect that or if there's a driver

2:08:57intervention or any of those settings we can create this trigger infrastructure that sends us data of those inaccuracies and so for example if we don't perform very well on Lane line detection on tunnels then we can notice that there's a problem in tunnels that image would enter our unit tests so we can verify that we've actually fixing the problem over time but now what you do is to fix this inaccuracy you need to source many more examples that look like that so we asked the fleet to please send us many more tunnels and then we label all those tunnels correctly we incorporate that into the training set and we retrain the network redeploy and iterate the cycle

2:09:29over and over again and so we refer to this iterative process by which we improve these predictions as the data engine so iteratively deploying something potentially in shadow mode sourcing inaccuracies and incorporating the training set over and over again and we do this basically for all the predictions of these neural networks now so far I've talked about a lot of explicit labeling so like I mentioned we asked people to annotate data this is an expensive process in time and also it was especially yeah it's just an expensive process and so these annotations of course can be very expensive to achieve so what I want to talk about also is really to utilize the

2:10:10power of the fleet you don't want to go through this human annotation bottle like you want to just stream in data and automate it automatically and we have multiple mechanisms by which we can do this so as one example of a project that we recently worked on is the detection of currents so you're driving down the highway someone is on the left or on the right and they cut in in front of you into your lane so here's a video showing the autopilot detecting that this car is intruding into our lane now of course we'd like to detect a current as fast as possible so the way we approach this problem is we don't write explicit a code for is the left blinker on is the

2:10:47right blinker on track the keyboard over time and see if it's moving horizontally we actually use a fleet learning approach so the way this works is we ask the fleet to please send us data whenever they see a car transition from a right lane to the center lane or left to center and then what we do is we rewind time backwards and we automatically can annotate that hey that car will turn will in 1.3 seconds cutting in front of the on preview and then we can use that for training than your lat and so the neural net will automatically pick up on a lot of these patterns so for example the cars are typically Yod then moving this way maybe the blinker is on all that stuff happens

2:11:22internally inside the neural net just from these examples so we ask the fleet to automatically send us all this data we can get half a million or so images and all of these would be annotated for currents and then we train the network and then we took this cut in network and we deployed it to the fleet but we don't turn it on yet we run it in shadow mode and in shadow mode the network is always making predictions hey I think this vehicle is going to cut in from the way it looks this vehicle is going to cut in and then we look for mispredictions so as an example this is an clip that we had from shadow mode of the cut in network and it's kind of hard to see but

2:11:57the network thought that the vehicle right ahead of us and on the right is going to cut in and you can sort of see that it's it's slightly flirting where the lane line is trying to it's sort of encroaching a little bit and the network got excited and they thought that that was going to be cut in that vehicle will actually end up in our center lane that turns out to be incorrect and the vehicle did not actually do that so what we do now is we just turn the data engine we source that ran in the shadow mode is making predictions it makes some false positives and there are some false negative detections so we got overexcited and sometimes and sometimes

2:12:24we missed the cut in when it actually happened all those create a trigger that streams to us and that gets incorporated now for free there's no humans harmed in the process of labeling this data incorporated for free into our training set we retrain the network and redeployed the shadow mode and so we can spin this a few times and we always look at the false positives and negatives coming from the fleet and once we're happy with the false positive or negative ratio we actually flipped a bit and actually let the car control to that Network and so you may have noticed we actually shipped one of our first versions of a copy intact architecture approximately I think three months ago

2:12:58so if you've noticed that the car is much better at detecting currents that's fleet learning operating at scale yes it actually works quite nicely so that's plate learning no humans were harmed in the process it's just a lot of neural network training based on data and a lot of shadow mode and looking at those results another very centrally like everyone's training the network all the time is what it amounts to whether that whether order polish on or off network is being trained every mile that's driven for the car that's harder to or above is training the network yeah another interesting way that we use this in a scheme of fleet learning at the other

2:13:38project that I will talk about is a path prediction so while you are driving a car what you're actually doing is you are annotating the data because you are steering the wheel you're telling us how to traverse different environments so what we're looking at here is a some person in the fleet who took a left through an intersection and what we do here is we we have the full video of all the cameras and we know that the path that this person took because of the GPS the inertial measurement unit the wheel angle the wheel ticks so we put all that together and we understand the path that this person took through this environment and then of course this this

2:14:10we can use this for supervision for the network so we just source a lot of this moon fleet we train a neural network on the on those trajectories and then the neural predicts paths just from that data so really what this is referred to typically is called imitation learning we're taking human trajectories from the real world I'm just trying to imitate how people drive in real worlds and we can also apply the same data engine crank to all of this and make this work over time so here's an example of path

2:14:37prediction going through a kind of a complicated environment so what you're seeing here is a video and we are overlaying the predictions of the network so this is a path that the network would follow in green and some yeah maybe the crazy thing is the network is predicting paths it can't even see with incredibly high accuracy they can't see around the corner but it would but it's saying the probability of that curve is extremely high so that's the path and it nails it you will see

2:15:11that in the cars today but we're gonna turn on augmented vision so you can see the the lane lines and the path predictions of the cars overlaid on the video yeah there's actually more going on under to hood that you can even tell you scary and of course there's a lot of details I'm skipping over you might not want to annotate all the drivers you might annotate just you might want to just imitate the better drivers and there's many technical ways that we actually slice and dice that data but the interesting thing here is that this prediction is actually a 3d prediction that we project back to the image here so the path here forward is a three-dimensional thing that we're just

2:15:46rendering in 2d but we know about the slope of the ground from all this and that's actually extremely valuable for driving so math prediction actually is live in a fleet today by the way so if you're driving clover Leafs if you're in a clover leaf on the highway until maybe five months ago or so your car would not be able to do clover leaf now it can that's a prediction running live on your cars we've shipped this a while ago and today you are going to get to experience this for traversing intersections a large component of how we go through intersections in your drives today is all sourced from a prediction from automatic labels so I talked about so

2:16:20far is really the three key components of how we iterate on the predictions of the network and how we make it work over time you require large varied and real data set we can really achieve that here at Tesla and we do that through the scale to fleet the data engine shipping things in shadow mode iterating that cycle and potentially even using fleet learning where no human annotators are harmed in the process and just using data automatically and we can really do that at scale so in the next section of

2:16:48my talk I'm going to especially talk about depth perception using vision only so you might be familiar that there are at least two sensors in the car one is vision cameras just getting pixels and the other is lidar that a lot of a lot of companies also use and lidar gives you these point measurements of distance around you now one one thing

2:17:08I'd like to point out first of all is you all came here you drove here many of you and you used your your neural net and vision you were not shooting lasers out of your eyes and you still ended up here we might have so clearly the human neural net derives distance and all the measurements in the 3d understanding of the world just from vision it actually uses multiple keys to do so I'll just briefly go over some of them just to give you a sense of roughly what's going on and inside as an example we have two ice pointed out so you get two independent measurement every single time step of the role ahead of you and your brain stitches this information together to arrive at some

2:17:45depth estimation because you can triangulate any points across those two viewpoints a lot of animals instead have eyes that are positioned on the sides so they have very little overlap in their visual fields so they will typically use structure from motion and the idea is that they Bob their heads and because of the movement they actually get multiple observations of the world and you can triangulate again depths and even with one eye closed and completely motionless you can still have some sense of depth perception if you did this I don't think you would notice me coming two meters towards you or a hundred miles back and that's because there are a lot of very

2:18:18strong monocular cues that your brain also takes into account this is an example of a pretty common visual illusion where you have you know these two blue bars are identical but your brain the way it stitches up the scene is it just expects one of them to be larger than the other because of the vanishing lines of this image so your brain does a lot of this automatically and a neural Nets artificial neural net scan as well so let me give you three examples of how you can arrive at depth perception from vision alone a classical approach and two that rely on neural networks so here's a video going down I think this is San Francisco of a Tesla so this is these are our cameras are

2:18:54sensing and we're looking at all I'm only showing the main camera but all the cameras are turned on the eight cameras on the autopilot and if you just have this six second clip what you can do is you can stitch up this environment in 3d using multi-view stereo techniques so this oops this is supposed to be a video

2:19:16is it not a video although I know it so there we go so this is the 3d reconstruction of those six seconds of that car driving through that path and you can see that this information is purely is it's very well recoverable from just videos and roughly that's through process of triangulation and as I mentioned multi-view Syria and we've applied similar techniques on a slightly more sparse and approximate also in the car so it's remarkable all that

2:19:45information is really there in the sensor and just a matter of extracting it the other project that I want to briefly talk about is as I mentioned there nothing about neural network neural networks are very powerful visual recognition engines and if you want them to predict depth then you need to for example look for labels of depth and then they can actually do that extremely well so there's nothing limiting networks from predicting this monocular depth except for label data so one example project that we've actually looked at internally is we use the forward-facing radar which is shown in blue and that radar is looking out and measuring depths of objects and we use that radar

2:20:19to annotate the what vision is seeing the bounding boxes that come out of the neural networks so instead of human annotators telling you okay this this car and this bounding box is roughly 25 meters away you can annotate that data much better using sensors so you sensor annotation so as an example radar is quite good at that distance you can annotate that and then you can train your lab work on it and if you just have enough data of it this neural network is very good at predicting those patterns so here's an example of predictions of that so in circles I'm showing radar objects and in and the keyboards that are coming out or here are purely from

2:20:54vision so the keyboards here are just coming out of vision and the depth of those cuboids is learned by a sensor annotation from the radar so if this is working very well then you would see that the circles in the top-down view would agree with the keyboards and they do and that's because you know levers are very competent at predicting depths they can learn the different sizes of vehicles internally and they know how big those vehicles are and you can actually derive depth from that quite accurately the last mechanism I will talk about very briefly is a slightly more fancy it's a bit more technical but it is a mechanism that has recently there's a few papers basically over the

2:21:30last year or two on this approach it's called self supervision so what you do in a lot of these papers is you only feed raw videos into neural networks with no labels whatsoever and you can still learn you can still get a neural networks to learn depth and it's a bit a little bit technical so I can't go into the full details but the idea is that neural network predicts depth at every single frame of that video and then there are no explicit targets that the neural network is supposed to regress to with the labels but instead the objective for the network is to be consistent over time so whatever depth you predict should be consistent over the duration of that video and the only

2:22:03way to be consistent is to be right as the network automatically predicts the that's for all the pixels and we've reproduced some of these results internally so this also works quite well so in summary people drive with vision only no no lasers are involved that seems to work quite well the point that I'd like to make is that visual recognition and very powerful really recognition is is absolutely necessary for autonomy it's not a nice to have like we must have neural networks that actually really understand the environment around you and and lidar points are much less information rich environment so vision really understands the full details just a few points

2:22:39around are much there's much less information in those so as an example on the left here is that a plastic bag or is that attire well lidar might just give you a few points on that but vision can tell you which one of those two is true and that impacts your control is that person who is slightly looking backwards are they trying to merge into your lane on the bike or are they just or are they just going forward in the construction sites what do those signs say how should I behave in this world the entire infrastructure that we have built up for roads is all designed for human visual consumption so all the signs all the traffic lights everything is designed

2:23:15for vision and so that's where all that information is and so you need that ability is that person distracted and on their phone are they going to work walk into your lane those answers to all these questions are only found in vision and are necessary for level 4 level 5 autonomy and that is the capability that we are developing at Tesla and through this is done through combination of large scale on your lower training through data engine and getting that to work over time and using the power of the fleet and so in this sense lidar is really a shortcut it's sidesteps the fundamental problems the important problem of visual recognition that is necessary for

2:23:52autonomy and so it gives a false sense of progress and is ultimately ultimately a crutch it does give like really fast demos so if I was to summarize the entire my entire talk in one slide it would be this every all of autonomy because you want level 4 level 5 systems that can handle all the possible situations in in 99.99% of the cases and chasing some of the last few nights is going to be there tricky and very difficult and is going to require a very powerful visual system so I'm showing you some images of what you might encounter in any one slice of that nine so in the beginning you just have very simple cars going forward then those cars start to look a little bit

2:24:33funny then maybe you have black stone cars then maybe have cars and cars they maybe you start to get into really rare events like cars turned over or even cars airborne we see a lot of things coming from the fleet and we see them at some rate like a really good rate compared to all of our competitors and so the rate of progress at which you can actually address these problems iterate on the software and really feed the neural hours with the right data that rate of progress is really just proportional to how often you encounter these situations in the wild and we encountered them significantly more frequently than anywhere else which is why we're going to do extremely well

2:25:05thank you [Applause] good it's all super impressive thank you so much how much data how many pictures are you collecting on average from each car per period of time and then it sounds like the new hardware with the dual dual active active computers gives you some really interesting opportunities to run in full simulation one copy of the neural net while you're running the other one the only the other one drive the car and compare the results to do Quality Assurance and then I was also wondering if there are other opportunities to use the computers for training when they're parked in the garage for the 90% of the time that I'm not driving my Tesla around thank you

2:25:59very much yep so for the first question how much data do we get from the fleet so it's really important to point out it's not just a scale of the data set it really is the variety of that data set that matters if you just have lots of images of something going forward on the highway at some point a neurologist gets it you don't need that data so we are really strategic and how we can choose and the trigger infrastructure that we've built up is is quite sophisticated analysis to get just the data that we need right now and so it's not a massive amount of data it's just very well picked data for the second question with respect to redundancy absolutely you can

2:26:29run basically the copy of the network on both and that is actually how it's designed to achieve a little 405 system that is redundant so that's absolutely the case and your last question I'm sorry I did not training the car is an inference optimized computer we do have a major program at Tesla which we don't

2:26:54have enough time to talk about today called dojo that's a super powerful training computer the Gulf Georgia will be to be able to take in vast amounts of data and train at a video level and do unsupervised massive training of vast amounts of video with the dojo program dojo computer but that's for another day

2:27:24test pilot in a way because I Drive the four five ten and all these really tricky really long tail things happen every day but the one challenge that I'm curious to how you're gonna solve is changing lanes because whenever I try to get into a lane with traffic everybody cuts you off and so human behavior is very irrational when you're driving in LA and the car just wants to do it safely and you almost have to do it unsafely so I was wondering how you're gonna solve that problem yeah so one thing I will point out is I spoke about the data engine as iterating on neural networks but we do the exact same thing on level of software and all the hyper parameters

2:28:03that go into the choices of when we actually link change how aggressive we are we're always changing those potentially running in shadow mode and seeing how well they work and so to tune our heuristics around when it's okay to lane change we would also potentially utilize the data engine and a shadow mode and so on ultimately actually designing all the different heuristics for when it's okay to lane change is actually a little bit intractable I think in the general case and so ideally you actually want to use fleet learning to to guide those decisions so when do humans lane change in what scenarios and when do they feel it's not safe to lane

2:28:33change and let's just look at a lot of the data and train machine learning classifiers for distinguishing when it is too safe to do so and those machine learning classifiers can can write much better code than humans because they have the massive amount of data backing that so they can really tune all the right thresholds and agree with humans and do something safe well we will probably have a mode that goes beyond Mad Max mode to LA traffic mode yeah well you know Mad Max would have a hard time in LA traffic I think yeah so it really it's a trade-off like you don't want to create unsafe situations but you want to be assertive but that little

2:29:02dance of how you make that work as a human is actually very complicated it was very hard to write in code but I think we really do you it really does seem like machine learning approach is kind of like the right way to go about it where we just look at a lot of ways that people do this and try to imitate that we're just being like more conservative right now and then as we gain higher higher confidence will allow users to select a more aggressive mode that'll be up to the user but in in the

2:29:30more aggressive modes in trying to merge in traffic there is a slight you an over no matter how many new there's a slight chance of like a fender bender or not a serious accident but you basically will have a choice if do you want to have a nonzero chance of a fender bender on freeway traffic which is unfortunately the only way to navigate la traffic yes yeah yes yes yes yes yes

2:30:02and it was nice with like la story that's going on yeah it will have more aggressive options over time that will be user-specified yes madam x+ exactly yes hello hi Jed or Hummer from Canaccord Genuity thank you and congratulations on everything that you've developed when we look at the alpha zero project it was a very defined

2:30:42and limited variable in terms of the parameters on that which allowed for the learning curve to be so quick the risk or what what you're trying to do here is almost developed consciousness in the cars through the neural network and so I guess the challenge is how do you not create a circular reference in terms of the pulling from the centralized model of the fleet to that handoff where the

2:31:13car has enough information where is that line I guess in terms of the the point of the learning process to handing it off where there's enough information in the car and not having to pull from the the fleet well look at the car can operate if it's completely disconnected from the fleet it just it just it

2:31:40uploads that the training that's you know better and better as the free fleet gets better and better so simply if you're just connected from the fleet from that point on it would stop getting better but it was so function fine in the heart the previous version and talked about a lot of the power benefits of not storing a lot of the images and so in this portion you're talking about the learning that's going on by pulling from the fleet I guess I'm having a hard time reconciling how if there was a situation where I'm

2:32:19driving up the hill as you showed and I'm predicting where the road is going to go that's coming from all of the other fleet variables that led to that that intelligence how I'm not how I'm getting the benefit of the the low power using the cameras with the neural network that's where I'm losing the the - maybe it's just me but I guess that's I mean the compute power in the full self-driving computer is incredible and

2:32:59it maybe we should mentioned that if it had never seen that road before it would still have made those predictions provided it was a road in the United States in the case of lidar the march of nines isn't there an example it won't just get to your slam on lidar because it's pretty clear you don't like light are in this latter flame lighter is name that isn't there like a case where at some point nine nine nine nine nine down the road we're actually lidar may be helpful and why not have it as some sort of a redundancy or backup sets my first question and the second so you can still have your focus on computer vision but just have it as a redundant my second

2:33:55question is if that is true what happens to the rest of the industry that's building their autonomy solutions on lidar they're all gonna dump lidar that's my prediction mark my words I should point out that I don't actually super hate light or as much as may sound but at SpaceX its basics dragon uses

2:34:16lidar to navigate to the space station or dock normally that we SpaceX developed its own lidar from scratch to do that and I spearheaded that effort personally because in that scenario lidar makes sense and it carves it's friggin stupid it's expensive and unnecessary and as Larry was saying once you solve vision it it worthless so you have expensive hardware that's worthless on the car the we do have a four word radar which which is low-cost and is helpful especially for occlusion situations so if there's like fog or dust or you know snow the radar can see through that if you're going to use active photon generation don't use visible wavelength because once you with

2:35:01passive optical you've taken care of all visible wavelength stuff you want to if you want to use a wavelength that is occlusion penetrating like radar so so what Lana is just active photon generation in the visual spectrum if you're gonna do active photon generation do it outside the visual spectrum in the radars in the radar spectrum so like at twenty point eight millimeters versus 400 to 700 nanometers you're gonna be a much better occlusion penetration and that's why we have a forward radar and then we also have I'll just twelve ultrasonics for for near-field information in addition to the eight cameras and and the ford radar young need the radar in the four

2:35:43direction because that's the only direction you're going real fast so it's I mean we've got over this multiple times like always sure we have the right-sized the sweet should we add anything more no high so right right here so you had mentioned that you asked the fleet for the information that you're looking for for some of the vision and I have two questions about that well it sounds like the cars are doing some computation to determine what kind of information to send back to you is that is that a correct assumption and are they doing that in real-time or are they doing based on stored information so they absolutely do do computation in in

2:36:29real-time on the car over there and we will wait to basically specify condition that we're interested in and then those cars do that competition there if they did not then we'd have to send all the data and do that offline in our back-end we don't want to do that so all that computation have us on the car so it's based on that question it sounds like you guys are in a really good position to have currently half a million cars in the future potentially millions of cars that are essentially computers representing free almost free data centers for you yes you to do computational is is that a huge future opportunity for Tesla car it is hearts a current opportunity and and that's not

2:37:07really factored in for anything yet that's incredible thank you we have four hundred twenty five thousand cars with hardware two and beyond which is means they've got all eight cameras the right of the radar in ultrasonics and they've got at least a the nvidia computer which is enough to essentially figure out what information is important what is not compress the

2:37:32information that is important to the most salient elements and upload it to the network for training so it's a massive compression of real-world data

2:37:49you have these sort of network of millions of computers which is like massive data centers essentially that are distributed data centers for computational capacity do you see it use being used for other things besides self-driving in the future I suppose it could possibly be used for something besides self-driving we're going to were focused on self-driving so you know as we as we get that really nailed maybe there's gonna be some other use for you know millions and then tens of millions of computers with Hardware three or four so traffic computer yeah maybe there

2:38:27would be it could be it could be maybe this like some sort of a AWS angle here it's possible hello hi you on mat choice loop ventures I own a model three in Minnesota where it snows a lot since camera and radar cannot see road markings through snow what is your technical strategy to solve this challenge is it involve high precision GPS at all yeah so actually

2:39:04like today actually auto pal will do a decent decent job in snow even when lane markings are covered even when Alana markings are faded covered or when there's lots of rain on them we still seem to drive a relatively well we didn't specifically go after snow yet with our data engine but I actually think this is this is completely tractable because in a lot of those images even when things are snowy when you ask a human annotator where are the lane lines they actually could tell you they actually are micro literally consistent in the rain those line lines as long as the annotators are consistent on your data then I have there's the neural network will pick up on those

2:39:36patterns and will do just fine so it's really just about is the signal there even for the human annotator if that is the answer to that is yes then the neural network can do it just fine there's actually there are number important signals as under saying so Lane lines are one of those things but one of them are the most important signals is drive space so what is drivable space and what is not drivable space and what actually really matters the most is is driveable space more than main lines and the and the prediction of drivable space is extremely good and I think especially after this upcoming winter will be incredible it's it's it's like it will be like how could it possibly be

2:40:16that good that's crazy the other thing to point out is maybe it's not even only about human annotators as long as you as a human can drive through that impairment fleet learning we actually know the path you took and you obviously use vision to guide you through that path you did not just use the lane line markings you use the entire geometry of the entire scene so you see like you know you see how the world is roughly curling you see how the cars are positioned around you you know that work will pick up on all those patterns automatically inside it if you just have enough of the data people traversing those environments yeah it's actually extremely important that things

2:40:48not be rigidly tied to GPS because GPS error can vary quite a bit and the if the the actual situation for a road can vary quite a bit so the reconstruction that could be a detour and if the car is is using GPS as primary this is a real bad situation as asking for trouble it's fine to use GPS for like tips and tricks so it's like you can drive your your home neighborhood better than a neighborhood enough in like some other country or some other part of the country so you know your our neighborhood well and you use kind of like the knowledge of your neighborhood to drive with more confidence to maybe have counterintuitive shortcuts and that kind of thing but you you it's the GPS

2:41:40overlay data should only be helpful but never primary if it's ever primary its problem so question back here in the back corner I just wanted to follow up partially on that because several of your competitors in the space over the past few years have made you know I've talked about how they are augmenting all of their perception and path planning capabilities that are kind of on the car platform with high-definition maps of the areas that our driving does that play a role in your system do you see it adding any value are there areas where you would like to get more data that is not collected from the fleet but is more kind of mapping style data I think the

2:42:24high precision of the high type high precision GPS maps and and lanes are a really bad idea the system becomes extremely brittle so any change like this this might any change to the system makes it it can't adapt so if it if it locks on to GPS and high precision Lane lines and and does not allow vision or

2:42:51override in fact great vision should be the thing that that does everything it's and then like Lane lines are a guideline but but they're they're not the main main thing we we briefly barked up the tree of high precision Lane lines and then realized that was a huge mistake and and reversed it out it's not good

2:43:20so this is very helpful for understanding annotation where the objects are and how the car drives but what about the negotiation aspect for parking and roundabouts and other things where there are other cars on the road that are human driven where it's more art than science it is pretty good

2:43:39actually look with cut ends and stuff it's doing really well yeah so I'll Commission we're using a lot of machine learning right now in terms of predicting kind of creating an explicit representation of what the world looks like and then there's an explicit planner and a controller and talk about representation and there's a lot of heuristics for how to traverse and negotiate and so on there is a long tail just like in what the visual environs look like there's a long tail in just those negotiations and a little game of chicken that you play with other people and so on and so I think we have a lot of confidence that eventually there must be some kind of a fleet learning

2:44:10component to how you actually do that because writing all those rules by hand is going to it's going to quickly vato I think yeah we've dealt with this issue with cut-ins and it's like will allow gradually more aggressive behavior on the part of the user they can do just dial the setting up and say be more aggressive be less aggressive you know Drive easy chill mode aggressive yeah

2:44:37incredible progress phenomenal two questions first in terms of platooning do you think the system is geared because somebody asked about when there is snow on the road but if you have large plat winning feature you can just follow the car in front does your system is your system capable of doing that and I have two follow-ups so you're asking about platooning so I think like we could absolutely build those features but again if you just use if you just train your own networks for example on imitating humans humans already what followed the car ahead and so that neural neural network actually incorporates those patterns internally it's just it figures out that there's a

2:45:18correlation between the way the car ahead of you faces and the path that you are going to take but that's all done internally in the net so you're just concerned with getting enough data and the tricky data and the neural training process actually is quite magical does all the other stuff automatically so you turn all the different problems into just a one problem just collect your data set and using your lopper training yeah this there's three steps to self driving you know this being future complete then there's being future complete to the degree that where we think that the the person in the car does not need to pay attention and then there's being at a

2:45:53reliability level we've also convinced regulators that that is true so it's got there's kind of like three levels we expect to be so feature complete in self-driving this year and we expect to be confident enough from our standpoint to say that we think people do not need to touch the wheel look out of the window sometime probably around I don't know second quarter of next year and then we start to expect to get regulatory approval at least in some jurisdictions for that towards the end of next year what is that that's a roughly the timeline that I expect things to go on and probably for for

2:46:29trucks the platooning will be approved by regulators before anything else and you can have like maybe if you're a long-haul doing long haul Freight you can have one driver in the front and then have four semis trailing behind in a platooning manner and I think that probably the regulators will be quicker to approve that than other things of

2:46:55course you don't have to convince us technology in my opinion which has an answer looking for a question probably dead I mean this is very impressive what we saw today and probably demo could show something more I'm just wondering what is the maximum dimension of a matrix that you may be having in your training or in your deep learning pipeline ballpark figure nice information of the matrix so you're doing you know what matrix multiply operations inside you know network you're asking about them there's many different ways to answer that question but I'm not percent sure they're they're useful they're useful answers these neural hours were typically have like I

2:47:32mentioned about tens to hundreds of millions of neurons each of them are on average have about a thousand connections to neurons below so these are the typical scales that are kind of used across T in this train also that that we reduce as well yeah I've been actually very impressed by the rate of improvement on autopilot the past year on my model three the two scenarios I wanted your feedback on last week first scenario was I was on the right-hand most lane of the freeway and there was a highway on-ramp and then my model three actually was able to detect two cars on the side slow down and let the car go in front of me and one car go behind me and I was like oh my gosh this

2:48:15is like insane like I didn't think my Model T could do that so that was like super impressive but the same week another scenario which is I was on the right-hand lane again but my right-hand Lane was merging with the left lane and it wasn't a on ramp it's just a normal highway freeway Lane and my Model T wasn't able to detect really that situation and I wasn't able to slow down or or speed up and I had to intervene kind of so can you from your perspective kind of share kind of the background on

2:48:51how a neural net would how Tesla might adjust for that and you know like how that could be improved over the Union over time yeah so like I mentioned we have a very sophisticated trigger infrastructure if you have intervened it's actually potentially likely that we received that clip and that we can actually analyze it and see what happened and and tune the system so it probably enters some statistics over okay at what rate are we are we correctly merging the traffic and we look at those numbers and we look at the clips and we see what's wrong and we try to fix those clips and make progress against those benchmarks so

2:49:29yes so we would potentially go through a phase of categorization and then we look at some of the biggest kind of categories that actually seemed to semantically be related to a simple through the same problem and then we would look at some of those and then try to develop software against that okay we

2:49:46do have one more presentation which is the the software's there's like essentially the the autopilot Hardware with with Stewart there's the the sort of neural net vision with Andre and then there's the software engineering at scale that's a compute presenter by Stewart so likes and will be opportunity afterwards to ask questions so yeah thanks [Applause]

2:50:13I just wanted to very briefly say if you have an early flight and you want to do a test ride with our latest development software if you could please speak to my colleague and/or drop her an email and we can take you out for a test ride and Stuart over to you

2:50:40all right so that's actually from a clip of a longer than 30 minute uninterrupted drive with no interventions navigate an auto pod on the highway system which is in production today on hundreds of thousands of cars so I'm Stewart and I'm here to talk about how we build so many systems at scale just like a really short induction I'm kind of where I'm coming from and I do so I've been in a couple companies or less I've been writing software profession about twelve years the thing that excites me most and I'm really passionate about is taking the cutting edge of machine learning and actually connecting that with customers through a bustah sand scale so at

2:51:12Facebook I worked initially inside of our ads infrastructure to build some of the Machine lore it's really really smart people and we she tried to build into a single platform that we could Zdenek scale to all the other aspects of the business from how we rank the newsfeed to how we deliver search results to how we make every recommendation across the platform and that became the applied machine learning group that's something he was incredibly proud of and a lot of that wasn't just the core algorithm some of the really important improvements that happened there those that matters a lot of actually the engineering practices to build these systems at scale on the same

2:51:40thing was true at snap where I went where we were really really excited to sort of actually help to monetize this product but the hardest part we were using Google at the time and they were effectively you know running us and a fairly small scale and we wanted to build that same infrastructure we take understanding of these users connect that with cutting-edge machine learning build that at a massive scale and hender billions and then trillions of both predictions and auctions every day in which is really robust and so when the opportunity came to come to Tesla that's something I'm just incredibly excited to do which is specifically take the

2:52:11amazing things that are happening both in the hardware side and the computer vision and AI side and actually package that together with all the planning that controls the testing the kernel patching of the operating system all of our continuous integration our simulation and actually build that into a product we get onto people's cars in production today and so I want to talk about the timeline for how we did that with navigate on autopilot and how we're gonna do that as we get a navigator on pot off the highway and onto city streets so we're at 770 million miles

2:52:40already for navigating on autopilot is something really really really cool and I think one thing that is worth kind of calling out on this is that we're continuing to accelerate and keep learning from this data like Andriy talked about this data engine as this accelerates up we actually do make more and more assertive lane changes we are learning from these cases where we will intervene either because they fail to detect emerge correctly or because they wanted the car to be a little more peppy in different environments and we just want to keep making that progress so to start all of this we begin with trying to understand the world around us and we talked about

2:53:13the different sensors in the vehicle but I want to like dig in a little bit more here we have eight cameras but then we also have additionally 12 ultrasonic sensors or radar an inertial measurement unit GPS and then one thing we forget about also the pail and steering actions so not only can we look at what's happening around the vehicle we can look at how humans chose to interact with that environment and so I'll talk to this clip right now this basically is showing what's happening today in the car and we'll continuing to push this forward so we start with the single neural network we see the detection surround it we then build all that together from multiple neural networks

2:53:45in multiple seductions we bring in the other sensors and we convert that into Alan calls of vector space an understanding of the world around us and this is something where as we continue to get better and better at this we're moving more and more of this logic into the neural networks themselves and the obvious endgame here is that the neural network looks across all the cars brings in all the information together and just ultimately outputs a source of truth for the world around us and this is actually not like a hardest rendering in many senses this is actually the output of one of the debugging tools that we use on the team every day to understand what

2:54:14the world looks like around us so another thing that I think is really really exciting to me I think when I do hear about sensors like lidar a common question is around just having extra sensory modalities like why not have some redundancy on the vehicle and I want to dig in on one thing that it's not it's not always obvious with neural networks themselves so we have a neural network running on our say wide fisheye camera that neural network is not making one prediction about the world it's making many separate predictions some of which actually thought of each other so it's a real example we have the ability to detect a pedestrian that's something

2:54:45we train very very carefully on and put a lot of work into but we also have the ability to detect obstacles in the roadway and a pedestrian is an obstacle and it's shown differently to the neural network it says oh there's a thing I can't drive through and these together combine to give us an increased sense of what we can and can't do in front of the vehicle and how to plan for that we then do this across multiple cameras because we have overlapping fields of view and many around the vehicle in front we have a particularly large number of overlapping fields of view lastly we can combine that if things like the radar and the ultrasonic stabilities extremely precise

2:55:15understandings which happen in front of the car we can use that both to learn future behaviors that are very accurate we can also build very accurate predictions of how things will continue to happen in front of us so one example I think is really exciting is we can actually look at bicyclists and people and not just ask where are you now but where are you going and this is actually the heart of ordinary part of our next generation automatic emergency braking system which will not just stop for people in your path but all software people are going to be in your path and that's running in shadow mode right now we'll go out to the fleet this quarter

2:55:46I'll talk about shadow mode in a second so when you want to start a feature like this for navigate on autopilot on the highway system you can start by learning from data and you can just look at how humans do things today what is their assertiveness profile how do they change lanes what causes them to either abort or change it like their manoeuvres and you can see things that are not immediately obvious like oh yes I'm will taneous merging is rare but very complicated and very important and you can start to build opinions about different scenarios such as a fast overtaking vehicle so this is what we do when you initially have some algorithm you want to try out we can put them on

2:56:19the fleet and we can see what they would have done in a real-world scenario such as this car that's overtaking us very quickly this is taken from our actual simulation environment showing different paths that we have considered taking and how those overlay on the real-world behavior of a user when you get those algorithms tuned up and you feel good about them specifically and this is really taking that out for the neural network putting it in that vector space and building and tuning these parameters on top of it ultimately I think we can do through more and more machine learning you go into a controlled deployment which for us is our early access program and this is you get this

2:56:49out to a couple thousand people who are really excited to give you highly vigilant but useful feedback about house behaves not an open-loop it in a closed-loop way in the real world and you watch their interventions and we talked about like when somebody takes over we can actually get that clip try to understand what happens and one thing we can really do is we can actually play this back again an open-loop way and ask as we build our software are we getting closer or further from how humans behave in the real world and what they were just super cool with the full self-driving computers we're actually building our own racks and infrastructure so he

2:57:21basically can pit four or full self-driving computers fully wrapped up build these into our own cluster and actually run this very sophisticated data infrastructure to actually understand over time as we tune in fix these algorithms are getting closer and closer to humans behave and ultimately we get can we exceed their capabilities and so once we had this we were really good about it we wanted to do our wide rollout but to start we actually asked everybody to confirm the cars behavior via a stock confirm and so we started making lots and lots of predictions about how we should be navigating the highway we asked people to tell us is this right or is this wrong and this is

2:57:55again a chance to turn that data engine and we did spot some really tricky and interesting long tales of in this case I think what really fun example like they're these very interesting cases of simultaneous merging where you start going and then somebody moves either behind or before you not noticing you and what is the appropriate behavior here and what are the tunings of the neural network we need to do to be super precise about the appropriate behaviors here we worked we tuned these in the background we made them better and over the course of time we got 9 million successfully accepted lane changes and we use these again with our continuous

2:58:28integration infrastructure to actually understand how do we think we're ready and this is one thing we're full self-driving is also really exciting to me since we own the entire software stack straight from the kernel patching all the way to the I suspect that the tuning on the image signal processor we can start to collect even more data that is even more accurate and this allows us to do even better and better tuning these faster iteration cycles and so earlier this month we were kind of thought we're ready to employ an even more seamless version of navigate on autopilot on the highway system and that seamless version does not require a stock confirm so you can sit there relax

2:59:02put your hand on the wheel and just oversee what the car is doing and in this case we're actually seeing over a hundred thousand automated lane changes every single day on the highway system and this is something is just like super cool to us to deploy at scale and the thing that I kind of most excited about from all this is the actual lifecycle of this and how we actually able to turn that data engine crank faster and faster and faster with time and I think one thing it's really becoming very clear is the combination of the infrastructure we have built the tooling we built on top of that combined power of the full self-driving computer I believe we can do this even

2:59:36faster as we move now to be an anon appellate from the highway system onto city streets and so yeah with that I'll hand off the only yes I mean to the best of my knowledge all those lane changes have occurred with zero accidents that is correct yeah I watch every single accident so it's conservative obviously but but it's to have hundreds of thousands going to millions of lane changes and zero accidents is I think a great achievement by the as a team yeah thank you [Applause]

3:00:12so let's see you know a few other things that are familiar with mentioning in order to have a self-driving car or Robo taxi you really need redundancy throughout the vehicle at the hardware level so starting in every was October 2016 all cars made by Tesla have redundant power steering so we're done at motors on the power steering so any one failure of the if the motor fails the car can still steer all of the power and data lines have redundancy so you can sever any given power line or any data line and the call will keep driving the auxilary power system even if the main pack you lose complete power in the main pack the car is capable of steering

3:00:55and braking using the axillary power system so you can completely lose the main pack and this the the car is safe the the whole system it from a hardware standpoint has been signed to for to be a Robo taxi since basically October 2016 so when we were all that Hardware autopilot version two what we do not expect to upgrade cars made before that we think it would actually cost more to make a new car than to upgrade the cars just to give you a sense of how hard it is to do this unless it's designed in yesterday it's not worth it so we've gone through the future of self-driving where it's clear it's it's hardware it's vision and then there's a lot of

3:01:46software and there's that the software problem here should not be minimizing some two massive software problem that that yeah managing vast amounts of data training against the data how do you control the car based on the vision it's a very difficult software problem so going after a guy over just like Tesla Tesla master plan obviously made a bunch of forward-looking statements as they call it and but let's go through some of our

3:02:22look forward looking statements that we've made no way back when we created the company we sit at both Tesla Roadster they said it was impossible and then and that even if we did build it nobody would buy it this was like universal opinion was that building an electric car was extremely dumb and would fail I agree with him that probability of failure was high but but that this was important so we built Tesla Roadster got a bunch of rupture in 2008 and shipping that car it's not collector's item they were build a more affordable car with the the Model S we did that again we were told that's impossible I was called a fraud and a liar there's not gonna happen this is

3:03:09all untrue okay famous last words now is we're into production with the Model S in 2012 exceeded all expectations there is still in 2019 no car that can compete with Model S of 2012 it's seven years later still waiting an affordable car maybe highly affordable it's affordable more affordable with the model 3 we've bought the model 3 we're in production I said we'd get over five thousand cars we've model 3 at this point five thousand cars week is is a walk in the park for us it's not even hard so we do large-scale solar which we

3:03:58did through the souls to the acquisition and that we're developing to play solar roof which is going really well we're now on version 3 of the solar tile roof and we expect this will a production of the solar tower roof significantly later this year I have it on on my house and it's great and like I

3:04:20sort of make the power wall and the power pack we made the power wind power pack in fact the power pack is now deployed in massive grid scale utility systems around the world including the largest operating battery projects in the world that at above 100 megawatts and in the next or probably my next you next year cheered at the most we expect to have a gaggle gigawatt scale battery project I completed so all these things I said we

3:04:51would do them we did it say we do it we did it we're gonna do the rover taxi thing to only criticism and it's a fair one and sometimes I'm not on time but I get it done and the tesla team gets it done so what we're gonna do this year is we're gonna reach combined production of 10,000 a week between air six and three feel very confident about that and we feel very confident about being future complete with self-driving next year

3:05:26we'll expand the park line with model why and semi and we expect to have the first operating Robo taxis next year with no one in them next year it's always difficult to like when things aren't on exponential at an exponential rate of improvement it's very difficult to correct one's mind around it because we're used to extrapolating on a linear basis but when you've got massive amounts of like as the hardware massive amounts of hardware on the road that the the the cumulative data is increasing exponentially the software is getting better at an exponential rate I feel very confident

3:06:12predicting autonomous Rover taxis for Tesla next year not an order of state or jurisdictions because we won't have regulatory approval everywhere but I'm confident we'll have least regulatory approval somewhere literally next year so any customer will be able to add or remove their car to the Tesla Network so expect us to operate it somewhat it's sort of like a combination of maybe the uber and airbnb model so if you own the car you can add or subtract it to the Tesla Network and tells the would take 25 or 30 percent of the revenue and and

3:06:50then in places where there aren't enough people sharing their cars we would just have dedicated Tesla vehicles so when you use the car we'll show you our ride-sharing app do you label just be able to summon the car from the parking lot get in and go for a drive it's really simple you just take the same Tesla app that you currently have we'll just worlds update the app and add a summin summin Tesla or or can make your car to the fleet so see the summin summin your car or at so many Tesla or add your add or subtract your part of the fleet you'll be able to do that from your phone

3:07:41so we see potential for smoothing out the demand distribution curve and having a car operates at a much higher utility than an old car it operates so like typically the use of a car is about 10 to 12 hours a week so most people will drive one and a half to two hours a day typically 10 to 12 hours a week of total driving but if you have a a car that can operate autonomously then most likely you could probably you most likely have that car operate for if a third of the week or longer so they're 168 hours in a week so probably you've got something on the order of 55 to 60 hours a week of operation maybe a bit longer so the the fundamental utility of the vehicle

3:08:30increases by a factor of five so you look at this from a macroeconomic standpoint and say just if this was like some if we were operating some big simulation if you could upgrade your simulation to increase the utility of cars by a factor of five that would be a massive increase in the economic efficiency of the simulation just gigantic so we'll do model 3 SAS 3 and X as taxis

3:09:01but we've made an important change to our leases so if you lease a model 3 you don't have the option of buying it at the end of the lease we want them back if you buy the car you can keep you can keep it but if you lease it you have to grid back and as I said we're in any locations where there's not enough to supply for sharing will Tesla will just make its own cars and add them to the network in that place

3:09:34so the current cost of roto model three Robo taxi is less than $38,000 we expect that number to improve over time and redesigning the cars the cars currently being built are all designed for a million miles of operation the drive units the design design and testing validated for a million million miles of operation the current battery pack is about maybe 300 to 500 thousand miles of the new battery pack that probably going to production next year is designed explicitly for a million miles of operation the entire vehicle battery pack inclusive well well it's designed

3:10:13to operate for a million miles with minimal maintenance maintenance so will actually be adjusting tire design and really optimizing the car for a hyper efficient Robo taxi and at some point you won't need steering wheels or pedals and will just delete those so as as as these things become less and less important we'll just leave parts just what they won't be there if you say like probably two years from now we make a

3:10:44car that has no steering wheels and pedals and if we need to accelerate that time we can always just delete parts easy and probably say a long term three years rubber taxis with with eliminated parts it may be it ends up being $25,000 or less and you want a super efficient car so the illustrate electricity consumption is very low so we're currently at four-and-a-half miles per kilowatt hour but we can we'll improve that to five and beyond and there's just

3:11:23really no company that has the full stack integration we've got the the vehicle design and manufacturing but the computer hardware in-house we've got the in-house Soccer Development the at and AI and we've got by far the biggest suite it's extremely difficult not impossible perhaps but extremely difficult to catch up when Tesla has a hundred times more miles per day than everyone else mind this is today's this is the cost of

3:11:52running a gasoline car the average cost of running a car in the u.s. is taken from triple-a so it's currently about a 62 cents a mile the inter 13 a half thousand miles for 15 million vehicles adds up to two trillion a year these literally just taken from the triple-a website cost of ride-sharing is

3:12:15according to her left is two to three dollars a mile the cost to run a rover taxi we think less than 18 cents a mile and and dropping like this is car this would be current this current cost future costs will be lower if you say what would be the probable gross profit from a single Robo taxi we think probably something on the order of $30,000 per year and expect that word

3:12:54literally designing we're designing the cars the same way that commercial semi-trailer a semi trucks are designed commercial semi trucks were all designed for a million mile life and we're designing the cars for a million mile life as well so no nominal fellows that would be you know a little over three hundred thousand dollars over the course of 11 years maybe higher I think these consumption is actually relatively conservative and this assumes that 50 percent of the miles driven are art there's nothing or not useful so this is only at 50% utility

3:13:36by the middle of next year we'll have over a million Tesla cars on the road with full self-driving Hardware feature complete at a reliably level that we would consider that no one needs to pay attention meaning you could go to sleep in your from our standpoint if you fast for a year should look maybe a year maybe a year in three months but next year for sure we will have over

3:14:03a million Robo taxis on the road the fleet wakes up with an over-the-air update that's all it takes you say what what is the net present value of a rover taxi probably on the order of a couple hundred thousand dollars so buying a Model 3 is good deal questions well I mean in our own fleets I don't

3:14:49know I guess long term we have probably on the order of 10 million vehicles I mean our production rates generally if you look at a compound annual production rate since 2012 which is like the that's our first full year of Model Model S production we went from 23,000 vehicles produced in 2013 to around 250,000

3:15:15vehicles produced last year so in the course of five years we increased output by a factor of 10 I would expect that something similar occurs over the next five or six years as for sharing sharing buses I don't know but the nice thing is that essentially customers are fronting us the money for the car it's great so

3:15:40in terms of the one thing is the snake charger I'm curious about that and how did you determine the pricing looks like you're undercutting the average lyft or uber ride by about 50 percent so I'm curious if you could talk a little bit about that the pricing strategy I'm sure we expect to solving the solving for the

3:16:00snake charger is it's pretty straightforward it's from a vision prop standpoint it's like a known situation any kind of known situation with with vision is like like a charge port it's a trivial so so yeah the car would just automatically Park but and automatically plug in there would be no one no human

3:16:21supervision required yeah no sorry what was a pricing huh yeah we just threw some numbers on there I mean I think it's like definitely plug in whatever pricing you think makes sense we just kind of randomly said okay maybe a dollar and things like it's theirs like on the order of two billion cars and trucks in the world so Robo taxis will be an extremely high demand for a very long time and from my observation thus far is that the order industry is very slow to adapt I mean I said there's still not a car in the road that you can buy today that is as good as the Model S was in 2012 so that

3:17:06suggests a pretty slow rate of adaptation for the car industry and so probably a dolla is conservative for the next 10 years because I keep people sort of think like there's like actually not enough appreciation for the difficulty of manufacturing manufacturing is insanely difficult but a lot of people I talk to think like if you just have the right design you can like instantly make as much of that thing as the world wants this is not true it's extremely hard to design a new

3:17:47manufacturing system for new technology I mean outies having major problems may affectionate ron and they are extremely good at manufacturing and if they're having problems what what about others so the you know there's this on the order of two billion cars trucks in the world on the order of about a hundred million units per year of production capacity of vehicles but but only of the old design it will take a very long time

3:18:19to convert all of that to full self-driving cars and they really need to be electric because the cost of operation of a gasoline diesel car is much higher than an electric car and any any any robotex that is an electric will absolutely not be competitive Elin it's a Colin rush from Oppenheimer over here you know obviously we appreciate that the customers are running some of the cash for this this fleet getting built up but it sounds like a massive balance sheet commitment from the organization over the course of time can you talk a little bit about what that looks like what your expectations are in terms of financing over the next call it three

3:18:57years three four years for building up this fleet and and storing monetize it with with your you know customer base well we're aiming to be approximately cash flow neutral during the fleet build-up phase and then I respect the extremely cashflow positive once the Robo taxis are enabled but I don't wanna talk about financing around so be difficult talk about financing rounds up in this in this venue but it

3:19:27well I think we'll make the right moves oh wait I think I'll make the move so you think think we should main I have a question if if I'm Ober why wouldn't I just buy all your cars you know why would I let you put me out of business there's a there's a close that we put into our cars I think it was about three or four years ago they can only be using the Tesla Network so even a private person like if I would go out and buy ten model threes I can't the network

3:19:59that's a business now right you're only it already used it doesn't network right but if I use the test the network in theory I could run a car sharing Robo taxi business with my ten model threes yes but it's like the App Store the the the you can only you can add only add or remove them through the Tesla Network and then tells it gets a revenue share but but similar to Airbnb though in that I have this home my car and now I can just rent them out so I can make an extra income from owning multiple cars and just renting them out like I have a model three I aspire to get this roadster here next when you build it and I'm gonna just rent my model three owl

3:20:37why would I give it back to you you know I guess you could operate a rental car fleet but I think this is very unwieldy yeah I don't it seems easy okay try it in order to operate a robo taxi now Orkut it sounds like you have to solve certain problems like for example autopilot today if you over steer it it lets you take over but if it's you know if it's a ride-sharing product that someone else is getting in the passenger seat like moving the steering can't let that person take over the car for example because they might not even be in the driver's seat so is the hardware already there for it to be a robo taxi and it might get into situations such as

3:21:34a cop pulling it over where some human might need to intervene like using central fleet of operators that remotely sort of interact with humans or I mean it's all of that type of an infrastructure already built into each of the cars does that make sense I think there will be sort of a phone home thing where if the car gets stuck it'll just burn home - to Tesla and ask for a solution things like being pulled over for you know by a police officer that's that's easy for us to program in that's not a problem the it will be possible for somebody to take over using a steering wheel or at least for some period of time and then probably down the road we'll just cap

3:22:19the steering wheel so there's no steering control well just take steering wheel off put a cap on and if you're in long you know give it like a couple years hardware modification to the to the car in order for it to enable that or yeah we're literally just unbolt the steering wheel and put a cap on where the steering wheel handle Carlita's but but that that is a like future car that you would put out but what about today's cars where the steering wheel is a mechanism to take over autopilot like it so if it's in a robo taxi mode would someone be able to take it over by just simply moving the steering wheel type yes I think it'll be a transition period

3:23:00where people will take over and should be able to take over from the rover taxi and then once regulators are comfortable with us not having steering wheel we'll just delete that and for cars that are on the that are in the fleet you know obviously with the permission of the owner if it's owned by somebody else we would just take the steering wheel off and put a cap where the steering wheel currently attaches so there might be like two phases to Robo taxi one where the the service is provided and you come in as the driver but could potentially take over and then in the future there might not be a driver option is that how you see it as well or like in the future that will in

3:23:41future will the probability of the steering wheel being taken away in the future is one hundred percent people consumers will demand it but but initially you would follow this is not clear this is not me prescribing a point of view about the world this isn't me predicting what consumers will demand yeah a consumers will demand in the future that people are not allowed to drive these two-ton death machines I don't but totally agree with that yeah but but in order for a model 3 today to be part of the Robo taxi Network when you call it you would then get into the driver's seat essentially because yeah just to be on the safe okay that's right sense thanks like you know there were

3:24:20amphibians you know but then pretty much that things just become like land creatures little bit a little bit of civilian phase hi sorry okay yes the strategy we've heard from other players in the Robo taxi space is to select a certain municipal area to create geo-fenced self-driving that way you're

3:24:50using an HD map to have a more confined area with a bit more safety hey we didn't hear much today around the importance of HD maps to what extent is an HD map necessary for you in a second we also didn't hear much about deploying this into specific municipalities where you're working with the municipality to get the buy-in from them and you're also getting a more defined area so what's the importance of HD maps and to what extent are you looking at specific municipalities for rollout I think HTM F SAR a mistake we actually had age after a while actually can't can that because you either need HD maps in which case if anything changes about the

3:25:32environment the car will will break down or you don't need HTM apps in which case why you're wasting your time during HD maps so the HD maps thing is like that the two main crutches that offer that should not be used and wilt in retro speak just retrospect the obviously false and foolish on lidar and HD maps

3:25:58hello if you need a geofence to area you don't have real self-driving just it sounds like maybe battery supply could be the only bottleneck left towards this vision and also could you just clarify how you get the battery packs to last a million miles I think cells will be a constraint that's all that's that's a subject for a whole separate there's a whole separate subject and I think we're actually gonna

3:26:35want to push out sort of standard range plus battery more than our long range battery because the energy content in the long range pack is 50% higher kilowatt hours so essentially you can make you know 1/3 more cars if you if you just if if they all sort of standard range plus instead of the long range pack the ones like we're on 50 kilowatt hours the other ones around 75 kilowatt hours so we're actually probably in a bias our sales intentionally towards the smaller battery pack in order to have a higher volume of what basically want to eat but the obvious max think thing to do is to maximize the number of autonomous units or the number of maximize the output that will

3:27:21subsequently result in the biggest autonomous leap down the road so we're making during a number of things in that regard but it's just on for today's meeting the million mile life is basically just about getting these the cycle life the that the pack - you know you need basically you know on the order like I say you've got a basic math if you've got a 250 mile range pack you know you need four thousand cycles so very cheap

3:27:58well we already do that with our stationary storage so much they say stationary storage solutions like power pack we're ready to ready to play power pack with 4000 cycle life capability yeah can I ask sorry yeah it's like we're tourism I a it's obviously significant very constructive margin implications to the extent you can drive the Tatra it's much higher of the full self driving option I just be curious if we can level set kind of where you are in terms of those attach rates and how you expect to educate consumers about the Robotech scenario so that attach rates do materially improve over time sorry a bit hard to hear your question yeah just curious where where we are

3:28:45today in terms of full self-driving at at rates in terms of the financial implications I think it's hugely beneficial if those attached rates materially increase because of the higher gross margin dollar that that flow through to the extent people do sign up for full FST just curious how you see that ramping or what the attach

3:29:07rates are today versus you know when do you expect how do you expect to educate consumers and get them aware that they should be attaching FSD to to their vehicle purchases we ramped that up massively after today yeah I mean if the fundamental really fundamental message that consumers should be taking today is that it's financially insane to buy anything other than a Tesla they will be like owning a

3:29:38horse in three years I mean fine if you're known a horse but you should go into it with that expectation if you buy a car that it does not that does not have the hardware necessary for full self-driving it was like buying a horse and the only car that has the hardness or acceptable self-driving Tesla like

3:29:59people should really think about their approaches any any other any other vehicle it's basically crazy to buy any other car than Tesla yeah we need to make that convey that argument clearly and we will have today thanks for bringing the future to present various rock informational time today I was wondering like you did not talk much about Tesla pickup and let me give a context for that I could be wrong but the way I am looking at test on network it will as an early adopter and something as a test bread I think Tesla's pick up maybe the first phase of putting the vehicles in network because the utility of Tesla pickup would be pretty much people who are

3:30:55either loading a lot of stuff or are in the profession of construction or little here and there odd items like picking up stuff from Home Depot sure I would say that you know maybe it needs to have a two-stage process pick up drops exclusively for tests on network as a starting point then people like me can buy them later but what are your thoughts on that well

3:31:23today was really just about autonomy there's there's a lot that we could talk about such as cell production pickup truck and future vehicle vehicles but today was just to focus on autonomy but I agree it's a major thing I'm very excited for the tells a pickup truck unveil later this year it's gonna be great Colin Lang and UB s just so we understand the definitions we need to refer to feature complete self-driving it sounds like you're talking level five no geofence is that what's expected by the end of year just so and then the regulatory process I mean have you talked to regulators about this this seems quite an aggressive timeline from what other people have put out

3:32:06there I mean are they you know what are the hurdles that are needed and what is the timeline to get approval and do you need things like in California know they're tracking miles that you know what's an operator behind that do you need those things but what is that process going to look like yeah we talk to regulators around the world all the time as we introduce you know additional

3:32:27features like a navigator an auto pilot we you know this requires like in regulatory approval on a jurisdiction basis so but I think fundamentally regulators my experience are convinced by data so if you have a massive amount of data that shows that autonomy is safe they listen to it they may take a they may take time to digest the information that process may take about a bit of time but they have always come to the right conclusion from what I've seen

3:33:13oh I have a question over here like it says I've got lights in my eyes and a pillar okay I just wanted just to you know some of the work we've done trying to better understand the ride Hale market it looks like it's very concentrated in major dense urban centers so is the way to think about this that the Robo taxis would probably deploy more into that area and the additional fault full self

3:33:41driving for personally owned vehicles would be in the suburban areas I think like probably yeah like Tesla owned Robo taxis would be in the intimate areas along with customer vehicles and then as you get to medium and lower density areas it would tend to be more that people own the car and occasionally lend it out

3:34:19yeah there are a lot of edge cases in Manhattan and say downtown San Francisco but those are you know and there are various cities around the world that have a challenging urban environments but we do not expect this to be a significant issue when I say future complete I mean it will work in downtown San Francisco and downtown Manhattan this year hi I have a neural net

3:34:51architecture question done do you use different models for say path planning and perception or different types of AI and sort of how do you split up that problem across the different pieces of autonomy well essentially the right now AI renewal that's used really for object recognition and we're still basically just using it as still frames so identifying objects that still frames and tying it together in a perception path planning layer they're after the but but what's happening is steadily is that the neural net is kind of eating into the software base more and more and so over time we expect the neural net to do more and more now from a computational cost standpoint there are

3:35:45some things that are very simple for a for a heuristic and very difficult for a neural net and so it probably makes sense to maintain some level of heuristics in the system because they're just computationally a thousand times easier than a neural net I can do that it's like a cruise missile and if you're trying to swat a fly just use a flyswatter or not a cruise missile so a bit over time to expect that it moves really to just training it on against video and then a video in car is

3:36:20steering and pedals out well basically it's video and that lateral longitudinal acceleration out almost entirely that's that's what we're gonna use the dojo system for there's no system that can currently do that maybe over here just going back to the sensor suite discussion you on the the one area I'd like to talk about is is a lack of side radars an instance situation where you have an intersection with a stop sign where there's maybe a 35 40 mile-per-hour cross traffic are you comfortable with the sensor suite the side cameras being able to handle that we talked about that yeah their problem essentially the cause going to do kind of what a human would do I think you may

3:37:07be human is like basically a camera on a slow gimbal and if it's a quite remarkable that people are able to drive the car in the way that they are because if you know what you can't look in all directions at once the car can literally look in all directions at once with multiple cameras so humans are able to

3:37:25drive just by sort of sort of looking this way looking that way there are stuck in their driver's seat they can't really get out of the driver's seat so it's like kind of one camera on a gimbal and is able to drive a conscientious driver to drive with very high safety the cameras in the cars have a better vantage point than the person so they're like up in the up and the b-pillar or in front of the rearview mirror they've got the really got a great vantage point so if you're turning onto a road that's got a lot of a lot of high-speed traffic you can just do what person is just like graduate like turn a little bit don't go fully into the road

3:38:03that the cameras see what's going on and if things look good and then the rear cameras don't show any oncoming traffic or if you go and if it looks sketchy you can just pull back a little bit just like a person if the behaviors like remarkably it starts to become remarkably lifelike it's like quite eerie actually that's a car just starts behaving like a person over here you're going then trouble

3:38:29course right here okay given all the value you're creating in your auto business by wrapping all of this technology around yourselves I guess I'm curious as to why you still be taking some of your cell capacity and putting it into power wall and power pack wouldn't it make sense to put every single you know unit you can make into this part of your business we're already stolen almost all the cell

3:39:00lines for that were meant to go to power world power pack and and use them for Model 3 I mean last year in order to make our model three production and not be sell stock we had to convert all of the 2170 lines at the gigafactory two to two car sells the and it's our actual output in in total gigawatt hours of stationary storage compared to vehicles is an order of magnitude different and for stationary storage we can we can basically use a whole bunch of miscellaneous cells out there so we can just gather cells of from multiple suppliers all around the world and you know you don't have a homologation issue or a safety issue like you have with

3:39:46cars so that's basically our stationary battery business has been just kind of feeding off scraps for for quite a while yeah so like we really think of like the production as being a there are many many constraints of a massive constrictive production system it's like a kid but like the degree to which manufacturing a supply chain is underappreciated is amazing they're a whole series of constraints and what is the constraint in one week may not be the constraint in another week it's

3:40:28insanely difficult to make a car especially one which is rapidly evolving so yeah but I'll just take a few more questions and then I think wishes break for so you can try out the cars all right anyone Adam Jonas questions on safety what what data can you share with us today how safe this technology is which would obviously be important in a regulatory or insurance well we published the accidents per mile every quarter and what we see right now is that autopilot is about twice as safe as a normal you know normal driver on average and we expect that to increase quite a bit over time it like said in

3:41:14the future it will be consumers will want to outlaw you don't think they will succeed or am I saying I agree with this position but in the future consumers will want to outlaw people driving their own cars because there is unsafe if you think of like elevators elevators used to be operated on a big lever like up and down the floor and there's like a big relay and yet elevator operators but then periodically they would get tired or drunk or something and then they turn

3:41:43the lever at the wrong time and sever somebody in half so now you do not have televator operators and you could be quite alarming if you went into an elevator that had a big lever that could just move between floors arbitrarily so there's just buttons and in the long term again not a value judgment well it's saying I want the world to be this way I'm saying consumers will most likely demanded that the people are not allowed to drive cars follow up can you share with us how much Tesla's spending on auto pilot or autonomous technology by order of magnitude on annual basis thank you it's basically our entire expense structure

3:42:31question on the economics of the Tesla Network just so I understand it look like so you get a model three off lease $25,000 goes on the balance sheet would be an asset and then you it would cashflow $30,000 a year roughly is that the way to think about yeah something like that yeah and then just in terms of the fan cing of it there's a question earlier you mentioned you would do it is it cash flow neutral to the Robo taxi program or cash flow neutral to Tesla as a whole

3:43:03sorry the cash flow here in terms of he asked a question about financing the Robo tax yet it looks to me like they're self financing but yeah you mentioned they would be basically cash flow neutral is that what you're referring to now I'm just saying between now and when the Robo taxis are fully deployed throughout the world the sensible thing for us is to maximize rate and drive the company to cash flow in your troll once that once the Robo taxi fleet is active I would expect to be extremely cashflow positive in this so you were talking about production yeah did to produce '''l okay thanks maximize the number of autonomous units made thank you maybe

3:43:42one last question yeah if I if I add my Tesla to the Robo taxi Network who is liable for an accident is it test lower than me if the vehicle has an accident and harms probably Tesla Tesla yeah I could the right thing to is to make sure there are very very few accidents all right thanks everyone please enjoy the price thank you [Applause] [Music]

资源与出处

相关视频

同频道、同主题或相邻年份