Tesla Roadmap Change is FSD and Teslabot Focused Roadmap

If people listened and believed what Elon said in the 2021 earnings call then they would hear the message that the tipping point for full self driving is this year. The annual report says there are 60,000 FSD beta users. This is out of about 300,000 US FSD beta purchasers. Elon says he would be shocked if they do not get FSD safer than human this year.

Elon said Optimus the Teslabot is the most important product they are working upon. It will be bigger than cars, energy and robotaxi.

If believed then there will 300,000 US FSD users this year. The general release of FSD beta to all FSD purchasers sounds like Q2 or Q3. Nextbigfuture expects this would drive FSD adoptions up and other countries would get their versions of FSD late this year or next year. This would likely get to a highly useful robotaxi level next year. Robotaxi likely needs to be at least twice as safe as human in order to show statistically to regulators that it is clearly statistically safer. Four to ten times safer for two to three standard deviations beyond human.

They are already making Tesla Semis. A safer than human level FSD would mean highway and city semi truck driving FSD. This would be something where the use case of safer than human, twice as safe as human and four times safer than human would have high economic value.

The full self driving AI would be used as starting point for navigation and control to guide the Teslabot. A useful robot product could be 2025-2027 in this scenario. 2030s the robots as become as common as droids in Star Wars. Elon and his team said it over and over but the belief was not in the listeners (stock analysts),

Tesla and Elon are so confident that they are saying .. Tesla will not lower prices but we will still sell every car they make. Every car at $50-60k per car. Value of the cars will go up over 5X. You can drive 85 hours per week instead of 17 hours per week.

Tesla roadmap change is FSD and Teslabot focused roadmap. This will replace labor and make the world economy unlimited.

SOURCES- Tesla, Elon Musk, David Lee Investing
Written By Brian Wang, Nextbigfuture.com (Brian has shares of Tesla)

20 thoughts on “Tesla Roadmap Change is FSD and Teslabot Focused Roadmap”

  1. Tesla is frittering away its advantage by ignoring radar and lidar. Perhaps cameras alone can work in good driving conditions but humans don't do well in dark, rainy or foggy situations. So why not do better than humans with instrument flying like airplanes.

  2. I think the Teslabot is Elon’s answer to the population decline problem. Robots do the work, we all get a universal income and we all live hundreds of years, with any surplus population going to Mars.

  3. For the 3d point cloud, sort of seems like that's what Spot from Boston Dynamics does. Which maybe explains Teslabot – Musk knew he'd have to replicate that ability anyway.

  4. I have wondered if people would feel more comfortable getting into an Uber/Taxi if a friendly humanoid robot were sitting behind the wheel. Even if it wasn't really doing anything in terms of driving the car.

  5. FSD Semi is nice, but cargo loading/unloading would be why humans would still ride. For conventional cargo though, a Teslabot could do the loading/unloading (though in most contexts a forklift/Teslabot hybrid would be moving pallets at the destinations). Loose parcel cargo would be the dream of a teslabot and electric cargo van combo for full autonomous delivery. Strapping down large cargo on flatbed trailers seems like a hard problem for teslabot though.

  6. That may be true, but a Teslabot is "universal" due to the bot's own sensors. Add to that that it is probably the most expensive component, so having a universal thing that is removable for reinstallation in any another vehicle is also a big advantage. It might even allow parlaying vehicle interfaces and equipment so that a Teslabot can hardwire into a followon of OBD2 ports, taking over much of the advanced functionality of current ECU's (potentially allowing manufacturers the ability to make their ECU's dumber/cheaper), and having access to vehicle sensors to augment it's own.

    Though a teslabot driver is a harder problem, as you reduce the sensor capabilities to the same as a human by default (not 360 views, blind spots, vehicle interface quirks/delays/sloppiness)

  7. Kind of looking forward to getting a good nap in once these bots are available. Maybe with folded hands, even.

  8. About the 3D mapping of the surroundings. This is a enormously computationally intensive task, which is probably why they have not implemented this in the current version of FSD beta. Is 72 Terra-ops – the current capability of the FSD computer v. 3.0 – sufficient to perform this calculation in real time? The jury is out on this one..

    Conventional methods would need far more calculations than 72 Terra-ops, but ANN's has a nack for finding "simplified" ways of solving calculations that are one to two orders of magnitude smaller in terms of flops. So it might be possible, but it is by no means certain, IMHO.

    And if not, then FSD computer v4 would be necessary for full FSD…

  9. It is impressive. One thought. Rob Mauer has said that "you kind of know when the FSD will run into problems in advance", which I interpret as most users only engaging FSD when they are confident it will work without intervention. So, the (global) metric interventions per mile for the FSD beta will be skewed because of this to a number that is more optimistic than reality.

  10. Sorry, FSD safer than humans this year.. I'd wish, but I don't believe. Call it 2025 instead?

    The perception of cars is still flickering in and out of existence when the cars are hidden from view, even through the spatial memory should have solved that. Furthermore, even given perfect perception the car still makes stupid mistakes.

    These mistakes includes not driving when it should (intersections), driving onto things that are obvious physical barriers. Generally, the FSD perception seems to recognize objects, but not structures. Meaning, if it does not "know" what a thing is – lane mark, sign, car, truck – it will see absolutely nothing in that space. This must be improved so that it can understand the 3D geometry of the surroundings even if it does not recognize the specific objects. A continously updated 3D map of the surroundings, irrespective of actual catogorization of the objects coupled with an ANN that governs driving commands.

    If these fundamental improvements are implemented, I think that we might see a step like improvement, but right now my impression is that FSD progress is slowing down to an assymptotical that is way below human capabilities. Say, converging towards one intervention every 5-10 miles?

  11. The problem with that scenario is that a humanoid robot would add mechanical reaction delay to the system if they physically move their body to control the car, making them less safe.

    And if they would 'plug in' to use the car's senses and directly control servomotors in the car, there'd be little point in using a humanoid body in a taxi, taking up a seat that could otherwise hold a passenger.

    Possibly a rich person who collects old "dumb" cars might have a humanoid robot do their driving, if robots are expensive enough to be status symbols?

  12. There's about 1.5B cars in the world, but new car sales to replace those are about 70M/year. EVs might survive longer – that'd reduce new car sales, but global demand for car mobility will increase, so call those a wash.

    Assume over the next few decades private cars drop to ~100M in service that (new and used) survive about ~20 years, meaning ~5M new private cars a year.

    With some increase in demand, maybe 350M robotaxis could serve the rest of us who would otherwise have private cars. Recycling them after 10 years (ICE taxis get about 5 years service, though some have an after service life) would mean ~35M new robotaxis sold a year.

    So about 40M total new cars a year by around 2040 when most currently-in-service cars are gone.

    By 2030 the other (surviving) car makers will have essentially caught up to Tesla on EVs and self-driving, and Tesla could be producing ~8M cars/year. Tesla holding onto 20% of the global car market through 2040 wouldn't be bad at all.

  13. Before 2030 under your assumptions it would be obvious that every one of 2 billion non-autonomous vehicles would need to be functionally replaced ASAP. That’s likely a lot more demand than 4M-8M per year.

  14. 60,000 of 300,000 FSD beta drivers is impressive. The top scoring 20% have it now. Dojo is presumably nearly operational and ready for the data. By later 2022 all 300,000 will be in the pool and Tesla’s take rate for FSD will be soaring along with it’s price. Tesla will be able to book most of the existing FSD payments.

    I’m a bit curious what happens with HW4 which was supposed to debut in CyberTruck. That’s probably an example of a supply chain issue. It seems likely they will start putting it in production cars as soon as it’s available. Model S & X? Easier to guarantee enough for lower volume models, highest priced. Semi of course.

  15. I'd guess Tesla FSD may start robotaxi tests in 2023, but it'll take them longer than they'd like to get approval for driverless operation, even in a relatively easy to drive area like suburban Phoenix where Waymo has been operating. (Maybe they'll offer Uber and Lyft drivers a great deal on Teslas if they'll agree to operate a certain number of hours as self-driving safety drivers, training FSD on unique ways a taxi operates.)

    Call it 2026 for first driverless robotaxi service approval in one or two cities, and by 2028 maybe they'll have proven FSD safety in 'difficult' cities and so can start getting approved for use in large urban regions around the world.

    There are ~20M taxis in the world. Assuming double or quadruple that if robotaxis cut into private ownership, and a 10 year life (double an ICE taxi, may require interior renewal in that time), demand for new Tesla robotaxis could eventually be around 4M to 8M a year. Tesla should be able to handle that by 2030.

  16. The idea of Optimus Teslabots driving ordinary vehicles isn’t silly. It’s something that seemed like silly retro SciFi that a RoboTaxi might be an ordinary Taxi that has a humanoid robot behind the wheel. In 2030 most vehicles in service will still be mindless and Teslabots may well be in mass production.

    The most complex proven and valuable skill available for Teslabots will be driving – they are built around the hardware and NN training system for FSD. That skill will be low hanging fruit.

Comments are closed.