Elon Musk Predicts Tesla Will Have Level 5 Full Self Driving Ready in 2021

Elon Musk is highly confident Tesla will have level 5 full self-driving ready in 2021 and will release it to Tesla customers in 2021. Some jurisdictions will allow full self-driving in 2021.

By 2030, about 70%+ of new cars will be electric and all will be full self-driving. It will take about 20 years to replace the entire fleet.

SOURCES – Axel Springer Award, Elon Musk, Business Insider Deutschland
Written by Brian Wang, Nextbigfuture.com

44 thoughts on “Elon Musk Predicts Tesla Will Have Level 5 Full Self Driving Ready in 2021”

  1. It still requires supervision at all times, and the driver is entirely responsible. That's level 2, not level 3.

    When the FSD beta allows you (legally) to take a nap in the driver's seat, then we can call it L3.

  2. humans mistake things at initial glance all the time. Even if current computer vision systems are wrong more often from the start, they can make up for it by reaction time.

    eventually though, computer vision will be as good or better, and then it'll be right first and faster in reaction.

  3. I'm not so sure the claims of massive increases in taxi's ride shares etc. are going to happen, or rather be as abundant as projected by their proponents, an increase in these services? for sure, but large amounts of society switching to non ownership? No. It's cheaper to own and you don't have strangers leaving trash in your unsupervised car/rental or scheduling conflicts if you own it solely for your own use.

    one good benefit will be a reduction in family car fleet sizes, American family driveways are currently clogged with vehicles now, often 2 or more (usually 3+ when your children are old enough to drive) but with self driving, you can reduce that number since your car can go drop off mom to the grocery store then drop the kids off to soccer practice then pick up dad from work etc.

  4. Several Musk time lines for full self driving have come and gone. It will happen, but guessing when is a fruitless exercise.

  5. As lousy(relatively) as Tesla's Autopilot is, it would make a competent human a much much safer driver. Unfortunately, most humans are not competent.

  6. That's if you think Waymo's approach is brittle or not. It seems they can be successful, but they rely on high precision LIDAR mapping. As soon as something changes, they need to remap. It makes it hard for them to scale.

    Beyond that, if you rely on LIDAR, can you work in inclement weather (rain, snow) that blocks LIDAR?

    Further, does the sensor and compute package cost more than the entire car when it comes to a Tesla Model 3 (~$40k). How much power does the Waymo sensor and compute array require? Tesla's uses a few hundred watts.

  7. Wildy optimistic,why would anyone wait for Tesla when Waymo and GM will be automated for years before Tesla makes any advances?

  8. Everything is an edge case to Tesla,do not expect fsd to be activated in 2021,they will continue to collect money for it.

  9. You'd have to make that mistake on every frame of the video feed. Those types of errors tend to not be stable as the scene changes. So if the car saw a dog last frame, and all of a sudden sees a aeroplane in the next frame, then back to dog, the 'driver' will not plow into it as if its not there.

    Here is a video of Tesla's beta FSD system handling deer at night on a country lane.
    https://www.youtube.com/watch?v=pWJoe8hwu_I&list=UUaBOIBVD_f6eyRRRBRrCz8w

  10. Except there are jurisdictions that have the regulations in place already. Business and voters will be clamouring for it when the benefits of it become obvious. Not to say all jurisdictions will adopt it right away, but those that don't are going to be left in the dust, economically.

  11. That seems rather pessimistic. They have cars that can drive on city streets for long periods without interventions right now, in the wild. You can watch it on youtube.

  12. I think people don't realize that if it is safer, even in a very litigious society, it makes sense to use FSD. As long as the penalties for at fault accidents are not more punitive if done by FSD than a human driver. Tesla may need to collect 'insurance' premiums from users to accept that liability, or bundle with their insurance product (since a user using FSD should ideally result in fewer claims).

    Companies are not afraid of getting sued. That is a fact of life for big business. They are worried about the liabilities stemming from litigation, but that can be mitigated.

  13. That rate is superhuman, but humans with their lower accuracy rates dont have the same kinds of failure modes. A human will not look at a puddle in the road when the sun is hitting it at just the right angle and see Jeffrey the giraffe, and with a high confidence level.

    Even with this flaw, the bots will still kill and maim less people than human drivers because humans have a short attention span. I no longer fear the bots will be held to a higher standard and sidelined until they prove they are perfect.

    Humans make mistakes all the time, the 1.3million people killed on the world's road every year are classified as accidents, that makes them sort of ok. Let's hope the bots are afforded the same level of freedom to maim and kill..

  14. 98.7%?

    That's ridiculous. You will encounter maybe 1000 vehicles/people/dogs etc. on even a short drive through a city. So you'll misidentify 3 of those. I presume that "misidentify" means something like "thought a dog was an aeroplane" not "thought a large car was a small truck". So a mistake that could cause a crash.

    That's a fail.

  15. If you don't have to drive on your daily commute then you will be happy to live out in the suburbs, where you can get a place 10x the size for the same price with double lockup garages.

  16. Following Waymo's lead, AutoX is China’s first self-driving co to remove safety drivers from robotaxis.

    It remains to be seen if Tesla's choice to go with the cheaper vision based navigation was correct. Although top 5 state-of-the-art image classification accuracy on ImageNet is @98.7%, it still occasionally fails spectacularly while being highly confident its identifications are correct.

  17. The USA is VERY litigious. The lone advantage of that is practice can precede laws or regulation as long as you can avoid being cratered by lawsuits.
    So in the USA it really just needs to have a better average safety record than humans.
    Insurance + excessive litigation = no laws necessary

  18. Elon Musk is highly confident Tesla will have level 5 full self-driving ready in 2021

    It could be just another "Full Self-Driving" release. Does "level 5 full self-driving" mean the same thing to Musk as it does to everyone else?

  19. Their system will go around objects in the way if it's been waiting awhile. These situations are already being handled in some cases.

  20. Level 5 is not what you think it means Elon. Tesla's FSD Beta system is only level 2. They are nowhere close to a L5 system and won't be for at least a decade. Maybe if they work really hard, they can reach L3 in a couple years.

  21. True, and that just makes it even slower.

    1. Develop SDV
    2. Get enough of them on the road in one location to see how they function in the real world.
    3. Demonstrate that they are better
    4. Now there is political pressure to adopt them
    5. Now other places start to do the work to adapt all their various transport rules, road rules, liability legislation, rules, regulations, work practices etc
    6. Then, when that's all in place, NOW you will start to sell SDVs
    7. Then SDVs, if they work as well as the rose tinted optimists hope, grows to 100% of the market

    That's a lot to get done by 2030. We appear to be part way through step 1.

  22. The thing is that, without an actual, honest-to-God real self-driving car to light a fire under their butts the proposals and studies will never even get off the ground. So the tech still needs to come out for the lawyers to begin to understand the legal ramifications of their existence.

  23. I can't see self driving reaching 100% before EVs do.

    Avery big proportion of all car buyers would get an EV tomorrow if they had the money.
    The number of people who want a self driving vehicle isn't THAT big. The number of governments happy with self driving vehicles isn't big at all.
    The number of legal systems that could adapt to self driving ability by 2030 even if they started today isn't that big.

    I mean, what if this morning a bunch of lawyers and traffic experts were told to start organising the work group to prepare the preliminary documentation for a proposal to do a study required to draft legislation to adapt the current road rules and regulations for self driving cars? How far through the process would they be by 2030?

    And don't interpret that as me criticising the legal process here. They SHOULD be doing a whole series of proposals and studies and carefully drafted changes to the law.

  24. Reduced accident rates give insurance companies LOW profits, not high profits.

    Insurance is a highly competitive market (outside of a couple of niches such as health insurance in highly regulated markets). So premiums tend to closely match the expected value of the payouts that will be made.

    Insurance companies make almost all their money on the "float". You are paying premiums now in return for a payout that will occur (on average, over the whole customer base) years from now. So you are effectively giving the company an interest free loan.

    The loan is invested in bonds and things and makes a market return. That's how modern insurance works.

    Now, drastically cut the accident rate. Insurance payouts fall. Within a couple of cycles the insurance premiums fall to match it. Now there is much less float, and hence much lower profits.

    Now of course the companies will do everything they can to keep the premiums up. But unless they can get the government to ban other financial companies from moving into their market, anyone earning excess premiums will be undercut by someone else.
    Note that this works especially well for insurance, because all you need to start an insurance company in a local market is a website and 50 million dollars. And the world is FULL of financial companies with money and the ability to put together an insurance company in any particular local market that is currently charging above the market value premiums.

  25. Drivers that give way when they have right of way are a menace to everyone and in the glorious future will be fined as though they were breaking the road rules.

    Everyone knew who was going to do what, and then this "nice guy" stops, even though he should have gone, and waves through someone who had already stopped because they knew they had to give way.

    So now what? Do we break the road rules to take advantage of this guy giving way? What if he changes his mind? What if he was just waving at his friend who was behind you and wasn't signalling to you at all?

    By the time everyone works out the "new rules" that apply to just this situation, it takes 5 times as long as if they'd just done the normal, expected, thing.

  26. If they're self driving, they can drive themselves to the charging station while you sleep. Hopefully we won't have so many cars parked on the street since many will shift to taxis.

  27. I think the missing angle here that Im not seeing better addressed is the charging infrastructure for urban dwellers that find it difficult to charge because they park on streets.

  28. robo taxi is such a change in practice from the existing procedures. It will take years of fledgling before an actual flight.

  29. there will be edge cases. A broken down car at an intersection that has row. A driver that waves you forward. Eye contact with pedestrians.

  30. I'd like a story on how the insurance companies intend on handling this. I can bankruptcy or (more likely) insanely high profits as accidents per mile decrease in the population.

  31. There are two parts of this question: 1. When it will be "ready" and more importantly is 2. When the government will actually allow it to be activated.

  32. Yeah, he has a reputation for always being a few years late in delivering miracles. The salient point is that’s still a reputation for delivering miracles.

  33. What’s new is FSD is already out in Beta that’s going to wide release in 2020. He has a pretty good idea how well that’s working under real world conditions already. Tesla Network is likely also near release.

  34. By Xmas 2021, expect Tesla share price to be well over $2,000 if Musk L5 prediction is accurate. I do however believe L5 is years away.

  35. The turnover of fleet increases as TAS takes hold through better capital utilisation. I suspect you'll see faster fleet replacement as a consequence.

  36. Hm, but what does this mean? Does it really mean that they could roll out a robotaxi network given government approval? Seems like an anwful lot of improvement to cover in such a short time, but I hope he is right..

Comments are closed.