首页 / 财富中文网 / 正文

马斯克的豪赌危险了:特斯拉无人驾驶出租车服务难以推广

财富中文网 2025-07-23 02:05:24

马斯克的豪赌危险了:特斯拉无人驾驶出租车服务难以推广
图片来源:Allison Robbert—AFP via Getty Images

• 近一个月前,特斯拉(Tesla)CEO埃隆·马斯克在奥斯汀高调推出了无人驾驶出租车服务。但“全自动驾驶(FSD)社区追踪器”(FSD Community Tracker)的主持人埃利亚斯·马丁内斯指出,此后发生的一切恰恰解释了该服务为何未能扩大规模。一起险些发生的事故证明,该服务的安全性远未达到无需人类安全员的地步。

就在埃隆·马斯克将其在奥斯汀推出仅三周的自动驾驶出租车服务扩展运行范围的当天,乔·泰格迈尔乘坐的特斯拉试图在火车驶近时非法穿越铁路道口。

他周一在X平台上发文称:“这辆无人驾驶出租车没有识别到(火车),安全员不得不停车等待火车通过。可见软件方面仍有些小问题需要改进,但除此之外,能见证扩展服务的运行情况有多好,这真是个绝佳的机会。”

对于这次差点丧命的事故,泰格迈尔似乎能泰然处之,随后他主张特斯拉应增加上路车辆的数量,以缩短等待时间。目前道路上仅投放了约10辆车,等待时间长达20分钟。

这一切对埃利亚斯·马丁内斯来说并不意外。作为最早参与全自动驾驶Beta测试的人员之一,他表示特斯拉的软件在过去四年里“取得了长足进步”。但他认为,所有现有证据都表明,该技术还远远无法支撑马斯克5月份提出的远大目标,即理论上首日即可支持1万辆汽车的规模。

他对《财富》杂志表示:“这些问题证明,特斯拉甚至本不该仅凭10辆车就启动服务。没错,它大部分时间都运行正常,但令我感到荒唐的是,我们仍会看到诸如FSD出租车闯红灯或在错误车道行驶等问题。这种情况不该如此频繁地发生。”

问题在于,每增加一辆车,发生碰撞的统计概率就越大。马丁内斯认为,任何无人驾驶出租车服务(包括Waymo在内)都需要做到近乎完美,才能安全地扩大规模,然而对于特斯拉而言,目前还看不到这种迹象。

对销量下滑的分心之举

前美国海军陆战队员马丁内斯主持的众包项目“FSD社区追踪器”,是目前公开可用的、针对特斯拉自动驾驶技术最复杂且最可靠的经验数据收集与分析平台。大众汽车自动驾驶出行(Volkswagen Autonomous Mobility)部门CEO克里斯蒂安·森格尔等汽车业高管高度评价其为行业标杆,甚至连马斯克也将其单独列出作为公司正在取得进展的证据。马斯克拥有内部脱管数据,但拒绝对外分享。

这款追踪器简单易用,其开发得到了一位加拿大特斯拉车主的协助:在行程中,像马丁内斯这样的FSD Beta测试员会实时将出现的问题直接记录到车载娱乐信息系统中,数据存储于此直至可上传至互联网。它通过每周表彰贡献最大的用户,激励驾驶员参与,使其变成了一种友好的竞争。

目前,其数据显示,即使是特斯拉最新的FSD版本(5月推出的13.2.9版,就在奥斯汀服务启动前几周),目前在城市和高速公路上平均每行驶约340英里(约合547公里)就会发生一次严重脱管。马丁内斯补充道:“你有时会听到埃隆说,‘我们很难发现脱管的情况。’这简直是胡说八道。”

尽管据信奥斯汀的无人驾驶出租车车队使用的是更新版本,但马丁内斯估计其表现与向公众发布的版本相差无几,因为它们暴露出相似的缺陷,例如在错误的车道上行驶等。

他认为,特斯拉更专注于不惜一切代价实现马斯克设定的6月上线时间表,而非完善实际的核心技术。由于今年上半年特斯拉电动汽车需求锐减,且Cybertruck已被证明在商业上一败涂地,马斯克需要一些东西来让投资者满意。

他表示:“这感觉像是对销量下滑的分心之举。”他补充道:“埃隆在豪赌。”

与此同时,特斯拉车主收到的最后一次重大更新是在七个月前向公众发布的13.2.1版本。

该公司未回应与本文或FSD自动驾驶技术其他相关事项有关的置评请求。

马斯克将未来押注于颠覆性技术之上

当特斯拉在周三美股收盘后举行第二季度财报电话会议时,马斯克将面临围绕其无人驾驶出租车试点路线图的一连串质询。截至发稿时,最受关注的问题是迄今为止该项服务在奥斯汀的表现情况,以及该服务能在多长时间内拓展到其他城市和更多车辆。

投资者在FSD技术上押下了重注,他们想知道奥斯汀的10辆车需要多久才能扩展到覆盖全国的数千辆。只有这样,他们才能大致判断特斯拉需要多长时间才能超越Waymo——特拉斯目前的完全无人驾驶里程数还为零,而其劲敌刚刚创下的1亿英里里程纪录。

这项技术可能成为颠覆性力量,尤其对于残障人士等边缘化群体而言。千禧一代杰西·沃林斯基是法律意义上的失明者,她在视频博客中记录了自己视力逐渐丧失的经历。她告诉加州监管机构,她很感激能成为Waymo“可信乘客计划”的一员。

她在2023年8月的一次听证会上表示:“它给了我一种前所未有的安全感。”加州不久后投票批准了这项技术。“乘坐Waymo的车辆,不仅让我能按自己的意愿到达目的地,这意义重大,而且整个过程无需担心遭到骚扰、猥亵、攻击甚至更糟的情况。”

马斯克已将公司的命运押注于无人驾驶出租车服务上,该服务现在必须产生利润,用以资助其目前正在开发的Optimus人形机器人项目。

若求信任,必先完全透明

但自动驾驶本质上是一项与统计概率息息相关的技术。在首次事故发生前,有多少辆车在同时运行?它们累计行驶了多少英里?是几千英里?几百万英里?还是更多?

对某些人而言,坐飞机看似充满危险,但实际上坐飞机是最安全的公共交通方式,因为99.9999%的航班都能安全降落。如今,尽管无人驾驶出租车面临的变量远远超过在相对空旷的天空飞行的飞机,特斯拉和Waymo等公司需要证明该技术能达到类似的可靠性水平。

这需要广泛而详细的数据——这正是马丁内斯在特斯拉社区帮助下收集的那类数据。然而,如果你向特斯拉寻求答案,你将一无所获。事实上,它的做法恰恰相反。特斯拉非但没有试图通过透明化来赢得公众信任,反而正在向联邦监管机构施压,要求封存其无人驾驶出租车的安全记录,声称出于商业原因这些数据必须保密。

马丁内斯表示:“这不该是专属商业机密。既然车辆在公共道路上行驶,数据就需要公开。他们隐瞒数据的事实本身就说明了一切。若真想赢得信任,就必须做到完全透明。”

相反,马斯克仅公布其FSD Beta计划(现称“FSD监督版”)的季度碰撞统计数据:今年前三个月,特斯拉车辆在发生事故前平均行驶了744万英里。虽然相比美国普通驾驶员平均70万英里的安全行驶里程,这个成绩相当出色,但这些并非无人驾驶出租车的里程。他们在碰撞发生前需要由驾驶员进行干预。

马丁内斯认为,即使是这些数据,在被采信前也应先由监管机构独立审核:“如果交给公司自己处理,他们就会过滤数据以符合自身的叙事立场。”

尚不具备安全扩展的条件

与此同时,特斯拉的回应似乎是对此一笑置之。周一,马斯克认为将在奥斯汀运行仅三周的无人驾驶出租车服务扩大覆盖范围,从地图上看形似巨型生殖器,这会很有趣。

这家市值万亿美元的公司写道:“更硬、更好、更快、更强”。这个一语双关的口号,指以表演机器人著称的双人乐团蠢朋克(Daft Punk)的同名合成器流行乐曲。马斯克不仅转发了这个生殖器形状的服务区域图以表赞同,还宣布将车费从之前的每次行程4.20美元上调至6.90美元。54岁的马斯克经常用这两个数字以达到滑稽效果。

简而言之,此次地理扩张更像是一场公关噱头,而非实质进展。运营收费的车辆数量似乎并未改变;特斯拉仍在限制可使用该服务的人数;车内也依然配备人类安全员。

在预测网站Polymarket上,投机者认为特斯拉能在今年剩余时间内于美国任何地点提供功能完备的无人驾驶出租车服务的概率仅为42%,较一个月前86%的高点大幅下降。

马丁内斯表示:“这表明他们尚不具备服务扩展的条件,如果贸然强行扩展,他们将会遇到问题。那样就是在置乘客于险境。与酒驾司机相比,它的风险或许确实更低,但仍然是一种风险。”(*)

译者:刘进龙

审校:汪皓

• 近一个月前,特斯拉(Tesla)CEO埃隆·马斯克在奥斯汀高调推出了无人驾驶出租车服务。但“全自动驾驶(FSD)社区追踪器”(FSD Community Tracker)的主持人埃利亚斯·马丁内斯指出,此后发生的一切恰恰解释了该服务为何未能扩大规模。一起险些发生的事故证明,该服务的安全性远未达到无需人类安全员的地步。

就在埃隆·马斯克将其在奥斯汀推出仅三周的自动驾驶出租车服务扩展运行范围的当天,乔·泰格迈尔乘坐的特斯拉试图在火车驶近时非法穿越铁路道口。

他周一在X平台上发文称:“这辆无人驾驶出租车没有识别到(火车),安全员不得不停车等待火车通过。可见软件方面仍有些小问题需要改进,但除此之外,能见证扩展服务的运行情况有多好,这真是个绝佳的机会。”

对于这次差点丧命的事故,泰格迈尔似乎能泰然处之,随后他主张特斯拉应增加上路车辆的数量,以缩短等待时间。目前道路上仅投放了约10辆车,等待时间长达20分钟。

这一切对埃利亚斯·马丁内斯来说并不意外。作为最早参与全自动驾驶Beta测试的人员之一,他表示特斯拉的软件在过去四年里“取得了长足进步”。但他认为,所有现有证据都表明,该技术还远远无法支撑马斯克5月份提出的远大目标,即理论上首日即可支持1万辆汽车的规模。

他对《财富》杂志表示:“这些问题证明,特斯拉甚至本不该仅凭10辆车就启动服务。没错,它大部分时间都运行正常,但令我感到荒唐的是,我们仍会看到诸如FSD出租车闯红灯或在错误车道行驶等问题。这种情况不该如此频繁地发生。”

问题在于,每增加一辆车,发生碰撞的统计概率就越大。马丁内斯认为,任何无人驾驶出租车服务(包括Waymo在内)都需要做到近乎完美,才能安全地扩大规模,然而对于特斯拉而言,目前还看不到这种迹象。

对销量下滑的分心之举

前美国海军陆战队员马丁内斯主持的众包项目“FSD社区追踪器”,是目前公开可用的、针对特斯拉自动驾驶技术最复杂且最可靠的经验数据收集与分析平台。大众汽车自动驾驶出行(Volkswagen Autonomous Mobility)部门CEO克里斯蒂安·森格尔等汽车业高管高度评价其为行业标杆,甚至连马斯克也将其单独列出作为公司正在取得进展的证据。马斯克拥有内部脱管数据,但拒绝对外分享。

这款追踪器简单易用,其开发得到了一位加拿大特斯拉车主的协助:在行程中,像马丁内斯这样的FSD Beta测试员会实时将出现的问题直接记录到车载娱乐信息系统中,数据存储于此直至可上传至互联网。它通过每周表彰贡献最大的用户,激励驾驶员参与,使其变成了一种友好的竞争。

目前,其数据显示,即使是特斯拉最新的FSD版本(5月推出的13.2.9版,就在奥斯汀服务启动前几周),目前在城市和高速公路上平均每行驶约340英里(约合547公里)就会发生一次严重脱管。马丁内斯补充道:“你有时会听到埃隆说,‘我们很难发现脱管的情况。’这简直是胡说八道。”

尽管据信奥斯汀的无人驾驶出租车车队使用的是更新版本,但马丁内斯估计其表现与向公众发布的版本相差无几,因为它们暴露出相似的缺陷,例如在错误的车道上行驶等。

他认为,特斯拉更专注于不惜一切代价实现马斯克设定的6月上线时间表,而非完善实际的核心技术。由于今年上半年特斯拉电动汽车需求锐减,且Cybertruck已被证明在商业上一败涂地,马斯克需要一些东西来让投资者满意。

他表示:“这感觉像是对销量下滑的分心之举。”他补充道:“埃隆在豪赌。”

与此同时,特斯拉车主收到的最后一次重大更新是在七个月前向公众发布的13.2.1版本。

该公司未回应与本文或FSD自动驾驶技术其他相关事项有关的置评请求。

马斯克将未来押注于颠覆性技术之上

当特斯拉在周三美股收盘后举行第二季度财报电话会议时,马斯克将面临围绕其无人驾驶出租车试点路线图的一连串质询。截至发稿时,最受关注的问题是迄今为止该项服务在奥斯汀的表现情况,以及该服务能在多长时间内拓展到其他城市和更多车辆。

投资者在FSD技术上押下了重注,他们想知道奥斯汀的10辆车需要多久才能扩展到覆盖全国的数千辆。只有这样,他们才能大致判断特斯拉需要多长时间才能超越Waymo——特拉斯目前的完全无人驾驶里程数还为零,而其劲敌刚刚创下的1亿英里里程纪录。

这项技术可能成为颠覆性力量,尤其对于残障人士等边缘化群体而言。千禧一代杰西·沃林斯基是法律意义上的失明者,她在视频博客中记录了自己视力逐渐丧失的经历。她告诉加州监管机构,她很感激能成为Waymo“可信乘客计划”的一员。

她在2023年8月的一次听证会上表示:“它给了我一种前所未有的安全感。”加州不久后投票批准了这项技术。“乘坐Waymo的车辆,不仅让我能按自己的意愿到达目的地,这意义重大,而且整个过程无需担心遭到骚扰、猥亵、攻击甚至更糟的情况。”

马斯克已将公司的命运押注于无人驾驶出租车服务上,该服务现在必须产生利润,用以资助其目前正在开发的Optimus人形机器人项目。

若求信任,必先完全透明

但自动驾驶本质上是一项与统计概率息息相关的技术。在首次事故发生前,有多少辆车在同时运行?它们累计行驶了多少英里?是几千英里?几百万英里?还是更多?

对某些人而言,坐飞机看似充满危险,但实际上坐飞机是最安全的公共交通方式,因为99.9999%的航班都能安全降落。如今,尽管无人驾驶出租车面临的变量远远超过在相对空旷的天空飞行的飞机,特斯拉和Waymo等公司需要证明该技术能达到类似的可靠性水平。

这需要广泛而详细的数据——这正是马丁内斯在特斯拉社区帮助下收集的那类数据。然而,如果你向特斯拉寻求答案,你将一无所获。事实上,它的做法恰恰相反。特斯拉非但没有试图通过透明化来赢得公众信任,反而正在向联邦监管机构施压,要求封存其无人驾驶出租车的安全记录,声称出于商业原因这些数据必须保密。

马丁内斯表示:“这不该是专属商业机密。既然车辆在公共道路上行驶,数据就需要公开。他们隐瞒数据的事实本身就说明了一切。若真想赢得信任,就必须做到完全透明。”

相反,马斯克仅公布其FSD Beta计划(现称“FSD监督版”)的季度碰撞统计数据:今年前三个月,特斯拉车辆在发生事故前平均行驶了744万英里。虽然相比美国普通驾驶员平均70万英里的安全行驶里程,这个成绩相当出色,但这些并非无人驾驶出租车的里程。他们在碰撞发生前需要由驾驶员进行干预。

马丁内斯认为,即使是这些数据,在被采信前也应先由监管机构独立审核:“如果交给公司自己处理,他们就会过滤数据以符合自身的叙事立场。”

尚不具备安全扩展的条件

与此同时,特斯拉的回应似乎是对此一笑置之。周一,马斯克认为将在奥斯汀运行仅三周的无人驾驶出租车服务扩大覆盖范围,从地图上看形似巨型生殖器,这会很有趣。

这家市值万亿美元的公司写道:“更硬、更好、更快、更强”。这个一语双关的口号,指以表演机器人著称的双人乐团蠢朋克(Daft Punk)的同名合成器流行乐曲。马斯克不仅转发了这个生殖器形状的服务区域图以表赞同,还宣布将车费从之前的每次行程4.20美元上调至6.90美元。54岁的马斯克经常用这两个数字以达到滑稽效果。

简而言之,此次地理扩张更像是一场公关噱头,而非实质进展。运营收费的车辆数量似乎并未改变;特斯拉仍在限制可使用该服务的人数;车内也依然配备人类安全员。

在预测网站Polymarket上,投机者认为特斯拉能在今年剩余时间内于美国任何地点提供功能完备的无人驾驶出租车服务的概率仅为42%,较一个月前86%的高点大幅下降。

马丁内斯表示:“这表明他们尚不具备服务扩展的条件,如果贸然强行扩展,他们将会遇到问题。那样就是在置乘客于险境。与酒驾司机相比,它的风险或许确实更低,但仍然是一种风险。”(*)

译者:刘进龙

审校:汪皓

• Tesla CEO Elon Musk launched his Austin robotaxi service nearly a month ago to great fanfare. But FSD Community Tracker host Elias Martinez argues everything that has happened since explains why the service has not scaled—it remains nowhere near safe enough to remove human safety monitors, as proven by a near accident.

The very day Elon Musk expanded the boundaries of his three-week-old autonomous ride hailing service in Austin, Joe Tegtmeyer’s Tesla tried to illegally run a railroad crossing just as a locomotive approached.

“The robotaxi did not see that, and the safety observer had to stop the vehicle until the train had passed. So there’s a little bit of work that still needs to be polished up with the software, but otherwise it’s been just an amazing opportunity to see how well the expanded service is working,” he said on Monday in a post on X.

Taking what might have been a life-threatening situation seemingly in stride, Tegtmeyer then argued in favor of Tesla adding more cars to the 10 or so currently on the roads to cut waiting times that had ballooned to 20 minutes.

None of this comes as a surprise to Elias Martinez. One of the earliest Full Self-Driving beta testers, he says Tesla’s software has “come a long way” over the past four years. But he argues all available evidence points to the technology being nowhere near robust enough to support the 10,000 cars Musk claimed in May were possible in theory on day one.

“These issues prove Tesla should never have launched even with just 10 vehicles,” he tells Fortune. “Yes, it works most of the time, but it blows my mind we’re still seeing issues like FSD running red lights or driving on the wrong side of the road. This shouldn’t be happening on such a regular basis.”

The problem is with each car added, the greater the statistical chance of a collision. Any robotaxi service, Waymo included, needs to be virtually flawless in order to scale the service safely—yet with Tesla there’s no sign of that, according to Martinez.

A distraction from declining sales numbers

The former U.S. Marine hosts the crowd-sourced FSD Community Tracker, the single most sophisticated and reliable form of empirical data collection and analysis on Tesla’s self-driving technology that is publicly available. Car executives like Volkswagen Autonomous Mobility CEO Christian Senger speak highly of it as a benchmark, and even Musk—who has his own internal data on disengagements that he refuses to share—singled it out as proof the company is making progress.

Developed with the help of a Canadian Tesla driver, his tracker is simple and easy to use: during a trip, FSD beta testers like Martinez catalog in real time problems that arise directly into the vehicle’s onboard infotainment system, where it’s stored until it can be uploaded to the internet. Drivers are incentivized through weekly recognition of the top contributors, turning it into something of a friendly competition.

Currently, its data shows even the latest FSD version from Tesla results in a critical disengagement roughly every 340 miles between both city and highway at present. Called 13.2.9, it rolled out in May just weeks before the Austin service launched. “You sometimes hear Elon saying, ‘we’re having a hard time finding disengagements.’ That is such BS,” Martinez adds.

Although the Austin robotaxi fleet is believed to be using a newer iteration, in Martinez’s estimation it closely approximates the performance of the version released to the public since they reveal similar shortcomings, such as driving in the wrong lane.

He believes Tesla has been more focused on meeting Musk’s June launch timetable come hell or high water than on perfecting the actual underlying technology. Since demand for his EVs dropped sharply in the first half of the year and his Cybertruck has proven to be a commercial flop, the CEO needs something to keep investors happy.

“This feels like a distraction from the declining sales numbers,” he said, adding “Elon is gambling.”

In the meantime, the last major update Tesla owners received, v13.2.1, launched to the public seven months ago.

The company did not respond to a request to comment on this or any other point related to its FSD self-driving technology.

Musk stakes future on game-changing technology

When Tesla hosts its second-quarter earnings call after the close of markets on Wednesday, Musk will face a barrage of questions around the roadmap of his robotaxi pilot. At press time, the top-ranked issue is the performance he’s seen so far in Austin and how soon the service can scale in terms of new cities and more vehicles.

Investors have a lot of money riding on FSD, and will want answers as to how soon 10 cars in Austin can grow to thousands across the country. Only then will they get a feeling for how long it will take Tesla to leapfrog Waymo, going from zero unsupervised miles currently to the 100 million just recorded by its archrival.

The technology could prove a game changer, especially for marginalized communities like the handicapped. Jessie Wolinsky, a legally blind millennial who video blogs about her experience slowly losing her eyesight, told California regulators she was grateful for being part of Waymo’s trusted rider program.

“It has provided me with a feeling of safety that I’ve never had before.,” she said at an August 2023 hearing shortly before the state voted to greenlight the technology. “I get into a Waymo vehicle, not only am I able to get to where I need to be on my own terms, which is huge, but I am able to do so without the fear of being harassed, groped, assaulted, attacked or potentially worse.”

Musk staked the company’s fortune on the robotaxi service, which now must generate the profits needed to fund his Optimus robot program currently under development.

If you want trust, you need full transparency

But autonomous driving at its heart is a technology steeped in statistical eventualities. How many cars are operating at the same time and how many miles do they collectively log before the first accident occurs—thousands? Millions? More?

Flying may seem like a dangerous endeavor to some, but there is no form of mass transportation safer since 99.9999% of flights land without incident. Companies like Tesla and Waymo now need to demonstrate a similar level of reliability despite variables far exceeding a plane flying through a relatively less crowded sky.

For that you need extensive, detailed data — the kind that Martinez collects with the help of the Tesla community. If you ask the company for answers, though, you’ll get none — just the opposite in fact. Instead of attempting to gain public trust through transparency, Musk’s company is currently pressing federal regulators to bury its robotaxi safety record, claiming the data must remain confidential for business reasons.

“This shouldn’t be proprietary. You’re driving on public roads so the data needs to be made available,” he said. “The fact that they’re hiding data should tell you everything you need to know. If you really want trust, you have to have full transparency.”

Instead, Musk only releases a quarterly crash statistic for his FSD beta program, now called FSD Supervised: for the first three months of this year Teslas drove 7.44 million miles before an accident. While this is a sterling result compared to the 700,000 miles for the average American driver, these are not robotaxi miles—they rely on drivers intervening before a collision ensues.

And even these figures, Martinez argues, should be vetted independently by regulators before being taken as credible: “If you leave it to a company, they will filter it to fit their narrative.”

Not ready to scale safely

Meanwhile, Tesla’s response seems to laugh it all off. On Monday, Musk thought it would be funny to expand the area covered by its three-week-old Austin robotaxi service to resemble a giant penis when seen on a map.

“Harder, better, faster, stronger,” the $1 trillion company wrote on Monday, a double entendre referencing the synth pop track of the same name by Daft Punk, a duo appropriately known for performing as robots. Musk approvingly reposted the phallus-shaped service map, adding the fare would now be hiked to $6.90 per ride from $4.20 previously, both numbers the 54-year old often employs for comical effect.

In short, the geographic expansion seemed more like a PR stunt more than anything else. The number of cars collecting fares has not appeared to change; Tesla continues to limit the number of people that can use the service; and human safety monitors still sit in the vehicle.

On the prediction site Polymarket, speculators have put the probability Tesla will have a fully functioning robotaxi service anywhere in the country at anytime during the rest of this year at just 42%, down from a high of 86% one month ago.

“It shows they’re not ready to scale, and if they did try to prematurely scale, they’re going to run into problems,” Martinez says. “Then you’re putting people at risk. Yes, maybe it’s a lower risk compared to a drunk driver, but it’s still a risk.”

*