A Million Zombie Taxis By 2020? It’s Not Going To Happen

The tech world has a love for Messianic figures, usually high-profile CEOs of darling companies whose words are hung upon and combed through for hidden meaning, as though they had arrived from above to our venture-capital-backed prophet on tablets of stone. In the past it has been Steve Jobs or Bill Gates, now it seems to be Elon Musk who has received this treatment. Whether his companies are launching a used car into space, shooting things down tubes in the desert, or synchronised-landing used booster rockets, everybody’s talking about him. He’s a showman whose many pronouncements are always soon eclipsed by bigger ones to keep his public on the edge of their seats, and now we’ve been suckered in too, which puts us on the spot, doesn’t it.

Your Johnny Cab is almost here

The latest pearl of Muskology came in a late April presentation: that by 2020 there would be a million Tesla electric self-driving taxis on the road. It involves a little slight-of-hand in assuming that a fleet of existing Teslas will be software upgraded to be autonomous-capable and that some of them will somehow be abandoned by their current owners and end up as taxis, but it’s still a bold claim by any standard.

Here at Hackaday, we want to believe, but we’re not so sure. It’s time to have a little think about it all. It’s the start of May, so 2020 is about 7 months away. December 2020 is about 18 months away, so let’s give Tesla that timescale. 18 months to put a million self-driving taxis on the road. Can the company do it? Let’s find out.

Elon Musk with a very shiny new Tesla. Maurizio Pesce [CC BY 2.0]
Elon Musk with a very shiny new Tesla. Maurizio Pesce [CC BY 2.0]

Oh, the Red Tape

The first hurdle he would have to overcome in order to achieve the goal is not a technological or capacity one, but a legal one. Is it legal to have a robot taxi service? The simple answer to that question is almost certainly no with laws in their current form as in the vast majority of territories such things are still only authorised for use in trials rather than commercial service. But let’s suspend that belief for a while and imagine he could change it with some legal wizardry. The answer then becomes extremely complex, depending upon where in the world you live and how you define a driverless taxi.

The USA for example — which we would imagine to be his largest potential market — enacts this legislation at state level, so his market becomes a checkerboard of different requirements. In all but a few states the taxi cannot even be truly driverless, instead a safety driver has to sit behind the wheel ready to take over in the event of a malfunction. Hardly a driverless car when you still have to pay a driver.

A drivers-eye view of a Tesla Model X in Autopilot. Ian Maddox [CC BY-SA 4.0]
A drivers-eye view of a Tesla Model X in Autopilot. Ian Maddox [CC BY-SA 4.0]

Are the Cars Up to the Task?

Assuming Tesla can find enough places in the world to deploy a million robo-cabs legally, can they do the job? We’ll take a big gamble because we’ve done the same above in giving legal authorisation to run them and assume that for the purposes of Musk’s 18-month timeline they can, but it’s safe to say that self-driving technology can still go wrong and cause accidents. There have been reports of Teslas on Autopilot mode failing to see trucks and attempting to drive straight through them, and also incidents involving pedestrians. Autonomous vehicles have caused fatalities, and it would be foolish to imagine that there would be no more following the introduction of a million more cars.

Can They Make That Many Cars In That Time?

Teslas on the production line in Fremont. Maurizio Pesce [CC BY 2.0].
Teslas on the production line in Fremont. Maurizio Pesce [CC BY 2.0].
So we’ve made a couple of leaps of faith, and arrived at a point where autonomous taxis can legally and effectively ply their trade. We need to place up to a million of them on the tarmac depending on how many existing cars upgrade, and we only have 18 months. Can that be done?

The Tesla company has several plants including the famous Gigafactories producing batteries, Powerwalls, and solar panels, but the cars are produced in a former General Motors plant in Fremont, California. Their total vehicle production from 2012 until the end of 2018 is reported as being in the region of 530,000 cars, with between 80,000 and 90,000 cars per quarter produced in the final two quarters of 2018. If production remains at that figure then they can expect to produce another 540,000 cars in the 18-month taxi timeline, so to produce their million they should expect to at the very least double their output.

It does not sound impossible to double production were there investment to expand their factory to that level, but of course it fails to account for any Teslas that would be sold as ordinary customer cars. To avoid taking their cars off the market and producing only taxis they would have to at least triple production simply to maintain their late-2018 sales, and increase it further were they to expect to sell any more. Meanwhile their inventory of unsold taxis would not be bringing in the usual sales revenue, so would need to be put to work with paying taxi passengers in short order. We’re not auto industry experts here at Hackaday, but to us those figures sound optimistic at best. Perhaps they could instead put their drivetrains in a car from another manufacturer such as those from Mercedes which are already subject to a technology agreement with Tesla, but that manufacturer would also have to ramp up production at short notice with little immediate cash return.

Our conclusion therefore is that unless Elon Musk has managed to secure the services of Harry Potter to magically enhance his production process (and a lawyer a bit better than Lionel Hutz to ease his way in the minefield of autonomous vehicle regulation) we won’t be seeing a million self-driving Tesla taxis by the end of next year. It’s very likely that self-driving taxis will eventually feature in our futures (after all, not far from where this is being written they’ve even been tested in public) and they may well have Tesla badges on them. But “one million” rolls off the tongue we thing that’s why this is another piece of Muskology. Let’s go back to watching SpaceX boosters landing perfectly on their pad at Cape Canaveral while we wait out a more reasonable timeline on those taxis.

52 thoughts on “A Million Zombie Taxis By 2020? It’s Not Going To Happen

  1. Zombie Taxis? I guess they are Zombie just because they are driverless?

    This one was so disappointing. I thought I was going to read something about driverless taxis that somehow just keep going, servicing customers all on their own long after their owning company has gone out of business just because nobody bothered to turn them off on their way out the door. Or maybe driverless taxis that lost their network connection due to malfunction and so were lost by their parent companies but they just keep going. Or something like that.

    That just seemed so much more interesting. I’m not used to equating automation with “zombie”. Is that a thing?

    1. if they try that here in the NY capitol region (Albany, Troy, Schenectady), 100% of them will never get anywhere beside the interstates. and that doesn’t include I787 or I890, since they’re snaking, narrow monstrosities of 1950’s highway design. as for the local roads, open street view and look at the nightmare that is driving around the Empire State Plaza.

  2. yeah, he is a master at Misdirection. He has figured out that as long as he keeps making outlandish claims, he can distract people from the claims that he doesnt achieve on time or the ones he doesnt achieve at all. Having one or two of his claims come true (even if not on the timeline he provides) just serves as confirmation bias for people that believe he can do no wrong.

    He is a showman and a business man, for that he has my utmost respect. He knows how to get people to open up their bank accounts to fund his business ventures and it certainly helps that some of the big ones have succeeded.

    1. It also helps that many of the big claims that have succeeded are fantastically difficult. Sure, he does things occasionally that detract from his accomplishments… but the accomplishments are absolutely amazing. Personally, I think if you have the technical competence to understand what the successful accomplishments are, you can’t help but be a fan. Unless, that is, your world view requires you to be one of the best in the world yourself.. in which case he’s obviously got to be a charlatan because for him to be otherwise would mean you’re probably not that special. Unfortunately, this last point seems to cover a lot of people.

      This is like an eccentric researcher claiming they’ll cure cancer by the end of 2019, they miss the deadline and cure cancer in 2021… and the news is all about how they’ve failed to meet the deadline and they’re some shady snake oil salesman. Hello!?! Cured cancer here! WTF.

      He’s bad with timelines. He probably could do better, but he obviously prefers to set optimistic goals rather than conservative ones. How the hell does that matter as much as people seem to make out? It’s crazy.

      Celebrate the actual accomplishments. They’re really quite fantastic at this point.

      1. I think you hot the nail on the head. “This is like an eccentric researcher claiming they’ll cure cancer by the end of 2019, they miss the deadline and cure cancer in 2021… and the news is all about how they’ve failed to meet the deadline and they’re some shady snake oil salesman. Hello!?! Cured cancer here! WTF.”

        Can’t launch a rocket and return it to earth. SpaceX Nailed it, they missed the promise date but still did it.

        Can’t create a new car company based on only EV’s and survive. 500,000 cars later Tesla is the most popular luxury car maker on the planet. It took way longer then estimated but got it done.

        Can’t just unify the internet around a single payment platform. When he sold Paypal they were pretty much the top dog. I’m sure he promised this to be finished faster then it does.

        At this point, he is over promising and under delivering to sell cars and try and get Tesla into the green. What it really comes down to is, will self driving be ready when the Army of cars come off lease.

        1. “Can’t just unify the internet around a single payment platform. When he sold Paypal they were pretty much the top dog. I’m sure he promised this to be finished faster then it does.”

          You should do some reading about paypal’s history, in the same month that Elon decided to terminate X.com’s other online banking operations to focus on Paypal he was replaced by Peter Theil. Paypal was a money transfer service that was owned by coinfinity which merged with X.com in march of 2000. Musk was replaced as CEO in october of 2000 and the company wasnt re-named paypal until 2001.

          To say that Paypal’s sucess was because of Elon, or that he made any claims of the sort is not a good logical argument to be bringing up as that is a very complicated and contentious point (as he clearly had other people influencing the company to get it there). Also, claiming that the internet is unified around a single payment platform is factually wrong. Sure paypal may be a dominant payment processor but they are not the only one, and thankfully so! (no one wants one company processing all of the payments as it could easily lead to monopolistic behavior).

          Over-promising and under-delivering is a big problem that he has and it is not something that should just be swept away because of his accomplishments. Analogy: You are a manager who has a rockstar employee, said employee has had some great accomplishments but yet cannot give an accurate estimate of completion of the work and even then not all of the times does the work get completed. Now tell me, is this really a position as a manager that you want to be in? I mean sure you could try and adjust for the advanced deadlines but then you do not have a guarantee that the work is going to be completed.

        2. The point of Tesla wasn’t just to make a new car company – those things come and go.

          Their mission was to a) prove that electric cars are just as good (failed – still aren’t), b) make an affordable electric vehicle in three steps: 1) roadster funds 2) luxury car funds 3) people’s car.

          What actually happened was, the Roadster was plagued with problems and was eventually a gimmick that didn’t make any profit. The Model S was half the car they promised and too expensive, failed to make any profit, the Model X wasn’t even part of the plan and it too was 50% more expensive than promised, again failed to make any profit, and then finally the Model 3 was again much much more expensive than promised and far away from being affordable in a market where the average car of its class (subcompact) sells for half the price and the difference in price buys you the car’s lifetime in gasoline.

          Tesla has been a total failure on all its promises – it was never supposed to spin-off into electric semi-trucks or SUVs, or self-driving vehicles. It had a simple mission: make electric cars for the masses to make transportation clean for the 21st century. It failed.

      2. Oh i am a fan, and i do celebrate his accomplishments but i also dont believe that you need to be an elon hater or lover and that like light, everything exists on a spectrum.

        The problem with confirmation bias is that it is essentially people throwing logic out the door for something they believe in. Not only is it a bad logical fallacy but when the subject making the claims believes it themselves it can lead to some pretty bad destructive behavior.

        Both sides of the Elon musk coin suffer from confirmation bias, the detractors focus on his missed deadlines and failures while the positive side insists on forgetting about his claims and focusing on his accomplishments only.

        The rational way of thinking about the claims he makes is to evaluate each one of his claims individually based on facts pertaining to that claim. In regards to the man himself, well, by making the assumption that he is a rationally functioning individual then we must also assume that he is able to see his missing set deadlines as a trend and therefor he either is doing it consciously (as a strategy to keep Tesla and his name in the public’s mind) or he is unable to adjust his deadlines to better fit the reality of the situation.

        As for how does it matter as much as people make it out to be: wallstreet. The market cares, and why should we care if the market cares? because everyone is invested in the market in some way shape or form (pensions, unions, governments, etc etc.). As a society we have handed the control of our economy to a bunch of “wealth managers” and what they care about is stability and people doing what they say they are going to do. Both of those things are needs when it comes to predictions regarding the future and for those wealth managers to make more money, Elon missing his deadlines while having some successes is a dangerous situation for them. They cant determine if something he says is going to be a failure, or a late success or even worse and early success (that they couldnt profit on).

        Any which way, it should be possible for people on either side of the Elon love/hate dichotomy to have an actual discussion with out resorting to hyperbolic claims or debasing the other side. Not only should we celebrate the accomplishments but we should also analyze his failures.

        We should be able to ask questions like: If him missing deadlines is a trend then why cant he analyze the source of the problem and correct to give more accurate deadlines?

        1. Because he’s buried himself with so many empty promises that his business would crumble apart instantly if he said “Sorry, we couldn’t do it”.

          He has to keep shifting the goalposts and lampshading his past promises in order to keep the house of cards up. He already ran his mouth so far ahead of his feet that he’s in the process of falling onto his face no matter how fast he tries to run. It’s just a question of when.

          Like the question of successes: what successes? He’s got an ailing car company that can’t actually make the cars he told it would, that is constantly on the brink of bankruptcy, a space company that’s doing cool things but don’t ask about the finances or who’s actually paying how much, a solar power company that is doing what exactly now?… etc. What exactly do you count as a “success”? Landing a rocket on its tail? Yeah, but wasn’t the point to make it profitable by saving a significant chunk of money per flight, rather than just do a party trick?

          1. Particularly for Tesla – it’s difficult to understand what objective merits it has, when all it has ever done is take off-the-shelf parts that are available for everyone, and put them together into cars that aren’t competitive with regular cars on the market and exist merely as hyped up toys for smug middle class pricks who like to pretend they’re saving the earth by buying a $100k car.

            The reason why nobody did electric cars after the 90’s is exactly the reason why Tesla hasn’t made a year of profit in its whole existence: the battery technology for electric cars just isn’t good enough to displace the petrol engine. It wasn’t when Tesla started, and it still isn’t, and Tesla couldn’t squeeze blood from a stone regardless of how much they tooted their own horn. Elon made the promise of making electric cars for the masses, and in that venture he has been a resounding failure on all counts.

          2. Or, SpaceX – woohoo, a successful private space company… that was founded on massive techology transfers and hand-downs from NASA after Elon failed to buy rockets from Russia. Its biggest customer with the largest investments, plus loans, plus subsidies, is the US government, plus the fact that it isn’t required to maintain launch readiness for the military which avoids all sorts of costs against its biggest competitors…

            What exactly is the success we’re counting?

            For example, where did SpaceX get its engines? From Tom Mueller

            >The TR-106 or low-cost pintle engine (LCPE) was a developmental rocket engine designed by TRW under the Space Launch Initiative to reduce the cost of launch services and space flight. Operating on LOX/LH2 the engine had a thrust of 2892 kN, or 650,000 pounds, making it one of the most powerful engines ever constructed

            Elon nabbed the lead engineer who was already working on reducing the cost of space access under a NASA and U.S. Department of Defense contract. The “innovation” that Elon Musk did was to spin a government funded program into a private company that was and is still being basically funded by the government. Better yet, it hasn’t actually managed to demonstrate any actual drop in the cost of space access, because the re-usable rocket program requires so many flights to turn any savings and they simply haven’t done enough yet. The whole thing is still completely unproven.

  3. We’re not even close to having an actually fully autonomous car so 1 million autonomous taxis in 2020 is simply never going to happen. Musk (and Tesla) is imho also way too optimistic about being able to do everything with camera’s and (limited) radar already installed in the car.

    1. One thing I will give Elon is that his claims of not needing LIDAR are not that far fetched, there are researchers that claim to be getting LIDAR levels of depth accuracy with cameras alone. https://www.youtube.com/watch?v=0FPQdVOYoAU&feature=youtu.be

      Then there is the question of how far along other aspects of the tech are. Nvidia only just started shipping their new SoC that has a heavy self driving focus. Their new Drive PX Pegasus has yet to ship. Tesla only just announced their self driving silicon as well.

      Why will these new chips and systems make any difference?
      A self driving car is actually a network of DNNs. The video link above is just one that uses a pair of cameras to extrapolate depth information that could be used for a self driving vehicle (assuming those artifacts on the left of the images is worked out)
      The video feed and possibly the depth information will also be used by other DNNs for recognizing specific objects or classes of objects. Others will be fed data from these networks to do things like try and figure out the speed and heading of objects in the scene, then still more will be used to orchestrate all this data with path planning and generating vehicle control inputs. Then there will be diagnostic DNNs to consider, things that can determine if a sensor is impaired, or at least judge the quality of the data from a given sensor.

      There is also a lot of work being done in simulation. Companies are not blind to the media, they can see how uncomfortable many members of the public are. And int he wake of certain high profile accidents, they don’t want their efforts out in the public yet. While they can feed data into the DNNs faster this way, they are limited to the training data they have gathered.

      If you are in an area where you see cars loaded up with sensors, the car is not necessarily driving itself. These might be drives that are being used purely to gather data to feed into the DNNs used in the simulators.

      TL;DR I don’t think it’ll happen by the end of 2020 but there are a lot of pieces people aren’t aware of because they are not happening in public. I think it will happen soon enough that Elon won’t be totally wrong, but it’s a bit more like his other predictions.

      1. I’m kinda agnostic on the LIDAR vs cameras debate, and I’m _still_ not sure why it has to be “versus” rather than a sensor fusion type solution. ANNs are practically built for that type of input.

        From what I’ve seen, LIDAR is still winning — the Waymo and Cruise cars are outperforming what little we know about the Teslas in fully auto mode. But Tesla isn’t betting on today, instead tomorrow.

        Musk’s point (a human driver is a stereo camera on a slow gimbal) is right at face value. But it also ignores the incredible difference in computation power between a human and the best of our current tech. It’s not unreasonable to help the poor computer out by giving it as much sensor input as it can process until the computer can catch up.

        TL;DR: I think the LIDAR/vision thing is mostly about time. LIDAR is ahead now, but given plausible computational (and algorithmic?) advances, it’s not impossible that computer vision could get the job done cheaper.

        But by 2020? I wouldn’t be the future of my company on it.

        (Edit: unless my company was bleeding money at such a rate that I had to make promises of future performance seem much closer than they really are.)

        1. The issue is that your computer isn’t helped by adding more and different types of sensor input – it merely gets swamped by data that it cannot effectively process. Another issue is that even with perfect data, it’s still operating more or less as a “checklist program” of conditional operations because of the difference in how the AI works versus how a real neural network operates.

          “IF A THEN B ELSE C” doesn’t constitute intelligence – even when you come up with the list by reinforcement learning and stuff it in a black box that has “ANN” printed on top. It’s still going to fail in the same catastrophic ways, such as rigidly prioritizing one input over another based on the particular rule it has learned.

          For example, the classic case where the image recognition NN learned to distinguish between people and buildings by the fact that photographs of buildings are always taken outdoors, so there are some telltale things like the blue sky. In the end, the NN went by the least amount of information it could to pass the test. Likewise, when you try “sensory fusion” for an artificial neural network, it picks up on the most consistent cue and ignores all the others because it passes the test anyways. That means for some scenarios it looks at the radar, for some at the camera, for some at the sonar, but it is very unlikely to cross-check them at all, not even by accident, because the NN is frozen when it is actually operating in the car to stop it from drifting away from the learned solution.

          1. That said, if Tesla really does “flip the switch” in 2020 and turn on full automatic driving based on reinforcement learning on the shadow driving data, we can all witness some very interesting bugs crop up.

            Given that the system is fundamentally collecting a bunch of IF THEN rules to pass a synthetic test where it is taught to react like humans would, it will pick up the least information that allows it to pass the tests, and that information needs only to correlate with the event, not actually relate to it in any way.

            Essentially, they’re teaching the cars to ape humans without understanding the reasons behind their actions. Being unable to perceive most of the triggers of these actions – such as non-verbal communication through gaze: people don’t look at the car’s cameras like they would into other people’s eyes – the AI has to pick up some other likely cue to trigger the same action. That will allow it to pass the training test, because it has to pass the test and it will pass the test eventually, but will make it fail in real traffic because it most likely has learned a false rule.

            This is the equivalent of acing your course exams not because you understand the problems, but because you took the exam so many times (or rehearsed so many sample exams) that you’ve learned the correct answers by memory. People actually do that – AIs do it even better because they’re absolutely relentless and its basically their only mode of operation.

            So you’ll be left with a whole bunch of baffled programmers trying to solve why Tesla cars kill people, but only on particular tuesdays.

        2. “Musk’s point (a human driver is a stereo camera on a slow gimbal) is right at face value.”

          A human also has a pair of gyroscopes, a pair of microphones, a number of vibration and force sensing elements which also work as accelerometers…

          Imagine if you had to drive a real car by remote, by only a pair of cameras with no force feedback or even sound. To make the situation more even, the cameras would be mounted on a well-balanced steadicam because the AI doesn’t know how to infer forces from the shaking of the picture – so you can’t guess that you just drove over the curbside because the image jolted.

          It’s not just the amount of processing that you can do, although it certainly helps. Other bits of information, like where the horizon appears and where you feel is “down” gives you constant feedback about your current sideways motion for an example. Without that sensation, it would be very difficult and inaccurate to sense lateral movement by the parallax of distant objects – which chances relatively little – and the objects close by are just a blur in your vision because you’re moving.

          1. Besides, the “slow gimbal” is inaccurate as well. Saccades are one of the fastest movements produced by the human body. Your eyes can turn at 900 degrees per second, and take between 20-200 ms to redirect your gaze from anywhere to anywhere in your visual field which spans almost 180 degrees. That’s fixing both eyes on the same target.

            A camera can only best that if you’re using a fisheye lens so you don’t have to physically turn it, but then you would get exceedingly poor angular resolution. You could use a tiny cellphone camera sensor and lens so you could move it as fast, but then your sensitivity will be poor and the camera will become slow otherwise (i.e. motion blur). Human retinas have vastly superior dynamic range compared to just about any camera.

            For example, a 1024×1024 sensor would give you 0.176 degrees per pixel, which means the camera couldn’t tell a white soccer ball from 100 yards away because it would be the size of a single pixel. A person has no problems identifying the soccer ball, because they are able to resolve about 1 arc minutes, which is over ten times better and would only be matched by a fixed camera of around 105 Megapixels. Some sources say 125 Mpix. Give or take.

            So not only are the AIs a bunch of hooey, the sensors you pair them with are actually also a pile of manure. The AI car just can’t see properly.

  4. A million times this. There is NO chance he will be able to do that. Not happening.
    Not only is the tech not even remotely there, the legal hurdles are a nightmare to get around even in fairly liberal areas open to innovation and change. Not happening. It’d take several years minimum to even get the go-ahead.
    The tech, on that note, is downright dangerous right now. Any company “brave” enough to put those cars on the road now or the next 2 DECADES is downright retarded ad suicidal. They’ll kill their company the instant it rams in to someone. Billions lost.
    Image recognition in every one of these systems, hell, every system that exists right now, are horribly flawed.
    Animals, including humans, have several hundred million years of image recognition behind them to refine those models.
    Meanwhile these dumb algorithms have a few carefully selected and constructed roads to drive through, with the people training them usually opting out of taking those dodgy side roads which is completely and utterly DUMB beyond belief!
    When a piece of black tape can convert stop signs to something else entirely, you know things are not good.
    Then there was that team recently that managed to screw with that Yolo… something or other algorithm by using a picture hanging from their body.

    These algorithms are horribly brute-forced. They have no intelligence behind them AT ALL.
    Until they do, it should be completely avoided for any mission critical or life threatening devices. It’s fine in carefully controlled environments like it is currently used, but the instant any uncontrolled values come in it all goes to hell very quickly.
    We need an entirely new substrate for processors to get to the core counts needed for _decent_ image recognition!
    Whether it is graphene, optical or some other weird exotic design, silicon absolutely cannot work for image recognition at these scales needed.
    The only reasonable workaround would be if 5g or even a potential “6g” level of networking was everywhere and you could offload image recognition to massive supercomputers because that’s what is needed. Same reason Siri and the like use “the cloud”, a 50 cars couldn’t fit the computing power needed for decent image recognition.
    The you get on to adapting your algorithms as they get exposed to the varied and massively different world out there.

    Not everything is nice clean lines and easy to read signs, in fact that is a majority of roads in most countries.
    Not to mention that large parts of the worlds drivers break the road laws all the time! Even Google said they would need to make their algorithms break the law to prevent crashing many times during their training.
    Loads of areas have their own local ways of driving on roads which have evolved over time. Regular drivers get used to them and don’t even question them despite a large number of these methods actually being against the law.
    It usually ends up confusing the hell out of someone new coming in and wondering what the hell the drivers are doing. (which can end up being funny or deadly)

    Self-driving cars? Not even in a decade, if not double that! Not without someone being killed. (again)

    1. What are you talking about?

      Here in Manitoba we already have driverless taxis! Sure, we are nowhere near 1-million yet and they run over the occasional foreigner that doesn’t understand our road rules but we’re getting there and that’s no great loss.

      I knew the rest of the world was behind but is it really so far behind as to not even have a driverless taxi?

    2. The legal hurdles are very country specific. If China decides to be the frontrunner, they are not going to take years to decide. Tesla will also never make the switch to full self driving. Instead, they will keep adding self-driving features until the car can drive itself. Legally, the driver is only required to be present and pay attention, both of these affect the driver not that car. As long as Tesla tells there customers they need to pay attention to the road, they can automate everything. People can then vote that they want to drop the legal requirement to pay attention, and then self driving cars will exist without any technical modification.

  5. The main problem with this article is it’s based on a falsehood, and the comments are happily repeating this falsehood.

    Musk never actually said there would be a million zombie taxis by 2020.

    1. Generally I am not a fan of theverge for it’s sub par reporting some times but I find it hard to believe that a reporter from a mildly respected journal would outright forge quotes.

      “We will have more than one million robotaxis on the road,” Musk said. “A year from now, we’ll have over a million cars with full self-driving, software… everything.”

      Sooo… what falsehood do we have? Because unless this quote is outright fabricated Musk absolutely said this ridiculous thing (just like every other ridiculous thing he says) and you just sound like someone in the Musk fan club who believes he can do no wrong while he consistently makes outlandish claims with no sense of reality. You ready for that Mars colony by 2028? I’m sure logistics just magically won’t be a problem in 9 years.

      1. “We will have more than one million robotaxis on the road,”
        No timeframe given for this.

        “A year from now, we’ll have over a million cars with full self-driving, software… everything.”
        He seems to define “full self-driving” as something like the current navigate on autopilot, that can drive a route without user input, unless there are exceptional conditions like construction or bad weather.

        So, a million self driving cars by next year does not equal a million robotaxis by next year.

      2. Could you probably check the original statement and not cite the verge? They are notorious for taking part of sentences out of context. And why would you cite some article anyway, if you could watch direct speech of a man you are talking about and try to understand the statement he made.

      3. That includes most of the existing Tesla’s, the don’t need to produce one million cars after finishing the self-driving software. The latest self-driving computer has the same physical size and power requirements, and most importantly does not use any new sensors. That’s not an accident, the plan is to enable this for most of the existing fleet.

    2. I know, right! It seems like they wrote this piece of text even without watching Musk’s announcement, and was using only headlines from WSJ and CNBC. Where is dislike button?!

  6. Dodgy showman rule #1. Make sure someone else is stopping you.

    No way were the legalities ever going to be sorted by 2020. But if you can do a demo of ‘capability’, no matter how far from production, you can claim you are being stifled or undermined…

  7. If self driving cars can’t figure out how to retreat to safety when their batteries get low… Then things like the buggy “delivery-drone” in the following video would happen:

    That stalled “drone” in the road could’ve caused an accident.

    On the subject of outlandish claims… I’m sure the UK will not have reached the Net-zero Emissions by 2025, Tesla’s claims not fruiting being one of millions of reasons.

  8. Uh, it’s very aggressive target and of course it will slip like virtually all “Elon time” targets do.

    So what? It’s coming, and it’s amazing to watch how the world is changing in our lifetimes. My 6 year old is probably not going to learn how to drive. My 12 year old might not either, it’s hard to tell. They’re going to live in a world where people don’t (commonly) drive cars anymore. Not next year, probably not the next few… but in 10 years probably and 15-20 years almost certainly.

    What I don’t get is why you seem to think they still need to produce 1 million cars to meet the 1 million robotaxi target. Let’ss take your numbers and say they’ve produced 530,000 cars so far. Only the very first cars won’t be able to work as robotaxis. With 6 quarters to go, and 80-90K per quarter CURRENT production… they don’t have to increase production at all to hit 1 million cars CAPABLE of being robotaxis. They don’t have to double production, where do you get that?

    As for how many people will actually add their car to the robotaxi network… I’d guess around 60-80% will at least try it out once.

    So, it’s reasonably possible that nearly 1 million cars could be REGISTERED on the Tesla taxi network if it was open for business. As for 1 million ACTIVE cars, yeah… that’ll be a bit further off. Probably not more than another year, though.

    The real limiting factors are not production, but how well they’re able to solve the long tail problems and convince regulatory bodies it’s safe enough to allow it to more forward. I’d normally be very skeptical that it’d be allowed any time soon… but then I’ve been surprised at how quickly some states have allowed fairly open ended autonomous testing… so… who knows.

    1. Testing doesn’t equal fit or safe for the public.

      A quarter million Model 3’s have been made, current production is about 24k per month.

      At that rate it will take 3 years to produce a million.

      But they are not fully autonomous vehicles.

      Show me where a driverless Model 3 can handle driving down the road at 70mph weaving in and out of traffic or handling dirt roads and unmarked paved roads. Driving in snow storms, dust storms and heavy rain with low visibility or deal with kids and animals on surface streets or the black clad cyclist driving on the road at night.

      Oh wait you can’t. Even serious automakers cannot make such claims.

      AFIAK Tesla is not doing real world autonomous vehicle testing. Right now Teslas are just for rich snobs who are into virtue signalling.

      1. Your comment implies that the autopilot must be able to drive the same way a human does, why?

        Personally, as long as I get there safely I don’t care as much about how fast I get there. Especially if I don’t have to stare at the tailights in front of me and pump the brake petal. If autopilot pulls in behind a semi and travels at 55 the whole way, who cares as long as I adequately budgeted time for it?

        At best I do my paperwork, at worst I watch YouTube videos the whole way.

  9. Fully autonomous transport of parcels and people is the missing final link needed to create an industrial feedback loop that launches us into massive exponential expansion. Electronic money transfer and automated manufacturing are already in place and deployment/undeployment of equipment can be automated well enough too. We are in for a serious explosion of automation that will make the original industrial revolution look like a fidget toy. Some day a machine will notice damage to a device or just a panel on a wall. It will call for its removal and replacement while sending the original off for repair/recycling/re-manufacturing eventually leading to it becoming usable again in another location or the same after being stored in a fully automated inventory system. Robotics is going “organic” at break-neck speed to the point of becoming indistinguishable from basic life in its dynamics and that’s a good thing. We need our machines to be be self repairing/building especially with us aspiring to expand into space, a place so hostile that only fully automated systems can keep up.

  10. Too much emphasis or faith is being placed in the idea of autonomous cars that will function (and solve all our traffic problems) on today’s streets. That’s dumb. Where driverless automation is most successful and dependable is in applications like container-ports, where not just the vehicles but the infrastructure has been adapted as well – with sensors and tracks and a central system communicating with each vehicle.

    Cities need better urban design, to REDUCE the number of personal vehicles that circulate, improve infrastructure for walking, biking and accessibility… and better mass transit. Self driving cars will have useful roles, but they’re not as big a part of the future as some people and companies fantasize.

    1. But this is where the autonomous vehicle proponents cannot go. The advantage of self-driving individual cars over good mass transit is that the latter involves a ton of expensive, centralized buildout. So, even though it makes sense, you can’t propose the same infrastructure for autonomous single-owner vehicles.

      To be fair, Tesla’s car-sharing/taxi plan is a step in the right direction. If/when it works.

  11. While this is obviously outlandish, I thought the idea wasn’t about minting a million new cars. Instead existing owners, after a patch, could turn on Elon’s Uber mode and your own car would go out at night driving people around. So if that figure is based on the number of existing Tesla’s in the world with the right hardware installed to support that, it might not be such a stretch to say there will be a million “taxi capable” cars by the end of next year.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.