\During Tesla’s quarterly earnings call in October, CEO Elon Musk reiterated a striking projection: the company is on the verge of releasing its long-promised Full Self-Driving package.

“While it’s going to be tight,” Musk said, “It still does appear that we will be . . . in early access release of a feature-complete Full Self-Driving feature this year.”

Tesla has not clarified what an “early access” release would entail, and the electric carmaker is notorious for flexible timelines. But the end-of-year goal has held steady since at least February—and with the end of 2019 fast approaching, that would mean at least some Tesla owners could get Full Self-Driving within a matter of weeks.

Any release of Full Self-Driving (FSD) this year would be a massive public relations coup, giving Tesla at least some claim to be the winner of the decade-plus race to create a self-driving car. As a sizable bonus, the company says it would also allow nearly $500 million in revenue from pre-orders of the self-driving features, held off the books for years, to be recognized in quarterly earnings statements. This has the potential to transform Tesla’s near-term financial outlook.

Tesla did not respond to multiple requests to comment for this story. But it has admitted that cars will not be fully autonomous when Full Self-Driving is released. “There’s the car being able to be autonomous, but requiring supervision and intervention at times. That’s feature complete,” Musk said on the October earnings call. “And it doesn’t mean like every scenario, everywhere on earth, including every corner case, it just means most of the time.”

There is ample evidence, however, that Tesla drivers routinely use existing features such as Autopilot without following the company’s guidelines for driver supervision. And the kind of scenarios Musk says FSD won’t be able to handle on release have already led to accidents of varying severity for users of other autonomous features.

Under these circumstances, the imminent release of FSD suggests at least the shadow of a conflict of interest, with Tesla’s bottom line and stock price in the balance against public safety.

The gold at the end of the self-driving rainbow

The half-billion dollars in deferred revenue from Full Self-Driving came from Tesla buyers who have pre-ordered the feature since 2016 for a surcharge of as much as $7,000. Customers who bought the feature got an onboard computer and all the hardware required for the car to drive itself—but without the software, which remains in development. When the software is declared complete, properly equipped Teslas will receive it via an over-the-air software update to those pre-installed hardware units. Tesla has not yet said whether it will roll it out to all drivers at once, or test it with smaller subsets.

Roughly half a billion dollars in presales would then be recognized as revenue, which Tesla said in June it would do over 12 months. A further portion of deferred FSD revenue, previously stated by the company as $316 million, would be recognized over the 8 year lifespan of each vehicle.

The $500 million referenced in October, spread evenly over the next four quarters, would add roughly $125 million per quarter to the bottom line. Though Tesla’s balance sheet shows big changes from quarter to quarter, some sense of the impact can be gained from looking at prior quarters. Adding $125 million in revenue to Q3 2019 results, for instance, would have increased operating profit by nearly 50%. Adding it to Q2 2019 would have cut operating losses by 74%. And because Tesla is considered a growth company, those changes could have an exaggerated impact on analyst’s projections of the company’s future growth.

“Feature complete” vs. completely functional

To be sure, Tesla has previously been aggressive about releasing features in semi-functional “beta” form, and the company is already warning that FSD will need further refinement after release.

On the October call, Musk specifically said FSD may not be prepared to deal with “corner cases” on release, making driver intervention necessary. These cases—better known as “edge cases”—include the truly unpredictable events common on mid-speed surface streets, from construction changing the flow of traffic, to large objects in the road, to pedestrians and cyclists.

“Edge cases” may have already misled existing Tesla autonomy features, such as the Autopilot feature for highways. One particular edge case, semi trucks turning across oncoming highway traffic, has allegedly contributed to two fatal accidents.

“Edge cases” are the main reason most manufacturers have slowed down the rollout of autonomous vehicles, leading expectations across the auto industry to slow sharply over the last year. Legacy carmakers like Ford and GM have moved away from any clear timeline for the retail sale of self-driving vehicles. Uber suspended its autonomous vehicle research last year after one of its vehicles killed a pedestrian. Waymo has released autonomous taxis, but only in one city, Phoenix, noted for its light traffic and good weather—that is, conditions where edge cases are at a minimum.

Tesla, by contrast, is aiming to release Full Self-Driving for owners worldwide, to potentially be used under a vast array of circumstances. But it is explicitly not guaranteeing that the technology will work under all circumstances on release.

Tesla’s language on that point may be misleading to the layperson. When Musk describes the coming FSD release as “feature-complete,” he’s referring to a list of discrete capabilities, including those already released—highway Autopilot, self-park, and, most recently, the parking lot feature Smart Summon. The additional features still needed to reach “feature complete” status primarily involve mid-speed city driving, such as recognizing stop signs and traffic lights. Musk described this part of the package in October as connecting the highway and parking lot features already released.

But ticking all those feature boxes—features which Tesla has defined for itself—is not the same as guaranteeing that Tesla owners can safely read the newspaper while their car takes them to the grocery store. In October, Musk said that “By feature complete I mean the car is able to drive from one’s house to work, most likely without intervention.”

That “most likely” caveat speaks volumes. As it has with Autopilot, Tesla will continue to officially require drivers to keep their hands on the wheel and monitor their vehicle to help the FSD system deal with unexpected edge cases.

This was foreshadowed in Tesla’s September 26 release of Smart Summon, a feature in the Full Self-Driving suite that’s designed to let a car pick up its driver in parking lots. The feature immediately ran into minor trouble, including some frightening and property-damaging mishaps documented on YouTube. The feature seemed particularly flummoxed by other cars and pedestrians. Consumer Reports described the feature as “glitchy” and “without a lot of obvious benefits for consumers.”

Yet Tesla added $30 million in self-driving revenue to its Q3 revenue on the basis of the Smart Summon release, making a small contribution to the company’s surprising third-quarter profit.

Failing before they succeed

Tesla leadership and supporters, though, have said that releasing software like Smart Summon with rough edges is part of the path to creating an autonomous vehicle that actually works. The most extreme expression of this mindset may have been Musk’s statements in 2016 that media highlighting the present-day risk of crashes involving self-driving technology would actually cost future lives:

“If, in writing some article that’s negative, you effectively dissuade people from using autonomous vehicles, you’re killing people,” the CEO said.

This stance is justified by the fact that, as with all forms of machine learning, autonomous driving algorithms depend on massive amounts of example data. Tesla’s aggressive approach, with all the risk it entails, has given the company a major edge in data gathering.

Teslas on the road today gather data about decisions made by their drivers, and test their autonomy systems in so-called “shadow mode”—recording the systems’ decisions for further review, without ever giving over control of the car. With a fleet of nearly 800,000 autopilot-equipped cars on the road, Tesla has recorded data from, by one analyst’s estimate, 1.8 billion miles of driving. Waymo has far more actual autonomous miles driven than Tesla or any other competitor, but Tesla appears to have more raw real-world data about driving conditions to work with.

With Full Self Driving actually available to owners, Tesla could quickly surpass Waymo’s tally of real-world autonomous miles driven, if it counts miles with a driver present but not intervening. Musk said in October that after its initial release, Full Self Driving would steadily improve to the point that “from a Tesla standpoint, we think the car is safe enough to be driven without supervision.” Presumably, those improvements will be partly enabled by additional data gathered with FSD active in the real world, though Tesla did not reply to clarifying questions on the topic.

Tesla has significant discretion both on technological and financial decisions related to Full Self-Driving. According to the nonprofit Center for Auto Safety, there are currently no federal performance standards or other regulations for advanced driver assistance technologies like FSD, or for fully-autonomous vehicles. So Tesla alone decides when FSD is ready for release. Tesla did not reply to inquiries from Fortune regarding its internal metrics for deciding when FSD will be safe for public use.

The decision of when to recognize held-back revenue is also made by Tesla, in collaboration with its accounting partners. This situation is not unique. “Lots of companies struggle with this very sticky question of what you can recognize as revenue.” says Joe Osha, a Tesla analyst with JMP Securities.

The key question, then, may be simply how much the public trusts Elon Musk and his company to decide when the technology is not a public safety risk, and to remove any financial considerations from that decision.

Jason K. Levine, Executive Director of the Center for Auto Safety, does not believe that trust is warranted. “Based on Tesla’s cavalier attitude up until now when it comes to quality control, the safety of their [systems], or educating consumers regarding the limitations of their technology,” he says, “There is nothing that should lead anyone to believe ‘entirely safe fully self-driving’ vehicle software is being released by Tesla anytime soon. All car manufacturers have a conflict of interest when it comes to putting safety before revenue. Most aren’t quite as publicly obvious about which one they are choosing as Tesla has been.”

Other observers believe Tesla has pure motives for the aggressive FSD timetable. “To be honest, that is less about the financials than it is just their desire to push the envelope in terms of the technology,” says Osha. “I agree with criticisms that they get out over their skis sometimes. But I don’t think that’s motivated by desire to move the gross margin. It’s their DNA.”

And despite cautious notes from Tesla itself, some who have experienced FSD in its current state are already impressed. Matt Joyce, a generally bullish independent Tesla analyst, got a sneak peek of Tesla’s autonomous city driving, and reports that it already works, though not flawlessly. “It was raw, but still super, super impressive. It literally went 15 minutes driving itself,” he said, through a variety of complex, real-world driving tasks on the streets of Fremont.

Even Joyce, though, isn’t confident Tesla will meet its nominal end-of-year target. “I think they are going to surprise people. It’s just not going to be on Elon’s timeline.”

Will Musk apply the brakes or speed ahead? With the end of 2019 rapidly approaching, investors and drivers alike are surely eager to find out.

在今年10月特斯拉的季度财报电话会议上，特斯拉的首席执行官埃隆·马斯克确认了一个惊人的猜测——特斯拉承诺已久的全自动驾驶汽车套件终于要来了。

马斯克表示：“虽然时间很紧张……但我们年内很有可能率先发布功能完整的全自动驾驶套件。”

特斯拉尚未就“率先发布”的内容给出任何具体说明，而且特斯拉爱跳票是出了名的。不过自从今年2月开始，特斯拉就一口咬定了年底这个时间节点。看来至少有少部分的特斯拉车主是有望在几周内体验到全自动驾驶功能的。

各大汽车厂商在无人驾驶领域已经竞争了十几年了，到目前为止，还没有人能够实现真正意义上的全自动无人驾驶（Full Self-Driving，以下简称为FSD）。如果特斯拉在今年确实如约发布了全自动驾驶套件，这将是公关上的一场巨大胜利，至少让特斯拉在名义上成为了无人驾驶领域的赢家。特斯拉表示，作为奖励，该公司将从这几年未入账的预付款中拿出5亿美元计入季度收益项，这至少将在短期内改变特斯拉的财务状况。

特斯拉并没有回复我们的多次置评请求。不过特斯拉承认，即便FSD功能发布了，车辆也无法实现全天候的全自动无人驾驶。马斯克在10月的财报会议上表示：“汽车虽然可以全自动驾驶，但它有时仍然需要人的管理和干预。我们只是说它在功能上完整了。而且它不意味着在任何情形、任何地方、任何极端状况下都可以实现全自动无人驾驶，它只是已经适合了大部分情况。”

不过现在已经有充分证据表明，特斯拉车主在使用现有的无人驾驶功能时，经常不按照操作指南对车辆进行必要的主动干预——当然，从马斯克的话来看，即便是在FSD功能发布后，也有一些情形对FSD来说是不可控的。而且这种情形已经造成了多起不同程度的交通事故。

在这种情况下，即将发布的FSD功能很有可能埋下利益冲突的种子。一方面，它对特斯拉的财务状况和股价无疑是一个利好消息，另一方面，它也给公共安全埋下了一定的隐患。

彩虹尽头的黄金

这5亿美元的递延收益是怎么来的呢？俗话说“羊毛出在羊身上”，它当然还是特斯拉的车主们。2016年以后购车的车主们可以选择多缴纳7000美元的定金，对这项功能进行预订。这样他们的车里就会预装全自动驾驶所需的车载电脑和其他相关硬件，只是它们没有软件——因为软件还在开发当中。一旦软件开发完毕，这些车辆就能够通过无线软件更新功能，将软件下载到预先安装好的硬件中。不过第一批FSD软件究竟是会推送给所有车主，还是只在小范围内测试，特斯拉并没有明确表态。

在FSD功能推出后，特斯拉就会将5亿美元的预售金额计为收益。特斯拉在今年6月表示，它将在12个月内实现这个目标。此外还有一部分与FSD相关的递延收益，将在每台车子8年的生命周期里被逐渐计入收益项，据特斯拉称，这部分的金额将达到3.16亿美元。

特斯拉在今年10月公布的这5亿美元收益将平均计入接下来的4个季度，也就是每季度新增1.25亿美元收益。尽管特斯拉的资产负债表在每个季度都有不小的波动，但我们只需要代入前几个季度的数据，就能够大概看出这笔收益对它的财务状况的影响。比如，如果在2019年第三季度的营收中增加1.25亿美元，那么特斯拉的运营利润将增加近50%。如果将这笔钱增加第2019年第二季度的营收中，它的运营亏损将下降74%。由于人们普遍认为特斯拉是一家增长型公司，这些变化必然会极大影响到分析师对特斯拉未来增长前景的预测。

“功能完整”VS完全自主

首先需要提醒的是，此前特斯拉就在以“内测”的形式，积极推广半自主化的无人驾驶功能。而且特斯拉也已经警告称，即便是在FSD功能发布后，它也需要进一步的优化和完善。

在10月的电话会议上，马斯克明确表示，特斯拉的初代FSD功能尚不能够应对一些“极端情况”，因此驾驶员的干预仍然是必要的。所谓“极端情况”，是指街面上发生的那些确实无法预测的事件，比如修路、路面上突然出现的大型物体，以及神出鬼没的行人和自行车等等。

一些“极端情况”可能已经误导过特斯拉现有的自动驾驶功能，比如它针对高速公路的Autopilot功能。其中的一种极端情况就是高速公路上的重型拖车。据报道，目前至少已经发生两起无人驾驶状态下的特斯拉因为撞上高速公路上的重型拖车而导致人员死亡的案例。

“极端情况”也是大多数汽车厂商决定暂缓推出无人驾驶汽车的一个主要原因。由于各大厂商对这些“极端情况”迟迟拿不出解决方案，在过去的一年里，整个行业对无人驾驶技术的预期也大幅放缓。福特和通用汽车等老牌厂商已经放弃了就无人驾驶汽车何时上市给出任何明确的时间表。去年，Uber的一辆无人驾驶汽车撞死一名行人后，Uber也暂停了它对无人驾驶汽车的研究。Waymo虽然已经发布了它的无人驾驶出租车，但范围却仅限于菲尼克斯这一个城市，因为这里的车流量不大，气候也以晴好天气为主——换句话说，在这种条件下，可以把“极端情况”出现的概率压到最低。

相比之下，特斯拉的目标是为全球车主提供全工况、全自动的无人驾驶功能。不过它并未明确表示这项技术在发布后可以立即在所有条件下使用。

对于外行来说，特斯拉在这个问题上的语言可能存在一定的误导性。马斯克在描述该公司即将发布的FSD功能时使用了“功能完整”这个词，他在这里指的是一系列独立的功能的集合，包括那些此前已经发布的功能——如高速公路自动驾驶、自动驻车，以及最近发布的“Smart Summon”停车场自动召唤功能等等。而要达到“功能完整”状态，首先需要解决的是中等车速下的市内道路驾驶问题，比如识别停车标志和红绿灯等。马斯克在10月表示，这部分功能将把已经发布的“高速公路自动驾驶”和“停车场自动驻车”功能连接起来。

即便这些功能都具备了，也不等于特斯拉的车主在开车去超市的路上可以悠闲地看报纸了。马斯克今年10月的原话是：“我所说的‘功能完整’，指的是车辆从家开到工作单位的过程中，很可能不需要人工干预了。”

“很有可能”这句话隐含着不少潜台词。就像它的Autopilot功能一样，特斯拉会继续正式要求车主在使用FSD功能时，将双手放在方向盘上，并对车辆进行监控，以帮助FSD系统处理任何极端情况。

FSD究竟有多靠谱？这一点从特斯拉9月26日发布的Smart Summon智能召唤功能中可见端倪。Smart Summon也是特斯拉的FSD套件的一部分，它可以将车辆从停车场中自动“召唤”至车主面前。不过该功能一经发布便遇到了不少的小麻烦，比如一些吓人的小事故和财产损失等，YouTube上就有不少相关视频。而它“鬼车”般的骚操作也吓坏了不少行人和其他车辆。《消费者报告》（Consumer Reports）认为这是一项“有缺陷”的功能，而且“对消费者并没有太多明显的好处”。

然而随着Smart Summon功能的发布，特斯拉又在其第三季度营收中计入了3000万美元的无人驾驶收入，为该公司令人惊讶的第三季度利润做了不大不小的贡献。

失败是成功之母

不过，特斯拉的领导层和支持者都认为，像Smart Summon这样的软件虽然存在一定瑕疵，却是在打造真正意义上的无人驾驶汽车的过程中不可或缺的一步。关于这一点，最极端的言论当属马斯克在2016年一段话，他认为，媒体一味夸大无人驾驶技术的撞车风险，这样做的代价，是会使很多人将来失去生命。

马斯克表示：“如果你写了一些负面的文章，成功地劝阻了人们对无人驾驶汽车敬而远之，那你就是在杀人。”

马斯克的这番激进言论，在逻辑上其实是自洽的。与所有形式的机器学习一样，无人驾驶技术的算法也依赖于大量的案例数据。特斯拉在无人驾驶上的高歌猛进，虽然会带来一定的风险，但也在数据收集上让该公司具备了重大的优势。

你可能不知道的是，在特斯拉汽车上路行驶的过程中，车辆会在所谓的“影子模式”下悄悄记录车主的驾驶决策数据，并对无人驾驶系统进行测试。不过它只是记录系统决策以供进一步分析，即便这时车辆并不处于自动驾驶状态。现在，特斯拉已经卖出了80万辆具备无人驾驶功能的汽车，有分析师估计，特斯拉已经记录了18亿英里的行驶里程数据。当然，如果只看实际的无人驾驶里程的话，Waymo的成绩要好于特斯拉或者其他任何一个竞争者，但特斯拉却记录了关于各种驾驶条件的更多原始数据资料。

在FSD功能发布后，如果算上有车主在场但未实施干预的里程，那么特斯拉很快就会在实际无人驾驶总里程上超越Waymo。今年10月，马斯克曾经表示，在FSD功能发布后，随着该功能的稳步改进，“从特斯拉的角度来看，我们认为总有一天，它可以达到没有车主监督也能够安全驾驶的地步。”可以想见，要实施这些改进，特斯拉就必然要通过FSD功能在现实世界中收集一些额外的数据。不过特斯拉并未回应我们的请求，对该问题做进一步澄清。

在FSD功能的有关技术和财务问题上，特斯拉是有很大的自由裁量权的。据非营利组织汽车安全中心（Center for Auto Safety）介绍，美国在联邦层面对于全自动无人驾驶或FSD这种高级驾驶辅助技术尚无统一的性能标准，也没有可以参考的规定。所以特斯拉可以自己决定什么时候发布FSD。关于特斯拉就FSD何时可供公众安全使用有无内部标准，特斯拉并未回复《财富》杂志的询问。

至于何时将递延收入计入营收，这个决定也是由特斯拉与它的会计伙伴共同做出的。这种情况也不鲜见。JMP Securities公司的特斯拉问题分析师乔·奥沙表示：“很多公司都在为一个非常棘手的问题感到头疼，那就是你能够把哪些东西计入收益。”

目前来说，最关键的问题就是，公众对马斯克和他的特斯拉公司有多信任？如果特斯拉认为它的技术不会给公众带来安全风险，并表示这个决定没有任何财务方面的考虑，公众会相信他们的这种说法吗？

汽车安全中心的执行主任杰森·莱文认为，特斯拉不见得一定能够赢得消费者的这种信任。他表示：“到目前为止，特斯拉在质量控制、系统安全性，以及在教育消费者他们的技术的局限性上，都表现出了比较傲慢的态度。所以人们没有理由相信，特斯拉短期内会推出‘完全安全的全自动驾驶’软件。在公共安全和企业收益的矛盾上，所有汽车厂商都会面临利益冲突，但大多数厂商都没有像特斯拉这样做出如此明显的选择。”

还有一些观察人士认为，特斯拉为FSD制订如此激进的时间表，其动机是很纯粹的。奥沙表示：“坦率地说，这与财务方面关系不大，主要还是由于他们具有推动技术极限的强烈欲望。我同意一些人的批评，他们的技术有时也会‘翻车’，但我并不认为他们这样做是为了提高利润。对技术的追求已经融入了他们的DNA里。”

虽然特斯拉自己对FSD的表态还算谨慎，但一些已经体验过FSD的人却实实在在地被惊艳了一把。马特·乔伊斯是一名总体上较为看好特斯拉的独立分析师，他最近抢先体验了特斯拉无人驾驶技术在城市道路上的表现。他表示，特斯拉的FSD功能已经相当厉害了，虽然它并非完美无缺。“它虽然比较原始，但已经非常、非常惊艳了。它实现了15分钟的全自动无人驾驶。”而且还在弗里蒙特市的街道上完成了一系列真实场景中的复杂的驾驶任务。

不过，就连乔伊斯也不相信特斯拉能够实现年底前发布FSD的目标。“我认为，他们会让人们大吃一惊，但马斯克的时间表则未必能够兑现。”

在这个问题上，马斯克的脚踩的是油门还是刹车呢？年底很快就要到了，无论是投资者还是特斯拉的车主都在期待着答案。（财富中文网）

译者：朴成奎