The Future of Self-driving Cars

I have researched the major self driving car companies.

We know Argo AI failed. They were ranked 5th in California based upon disengagement miles.
Disengagement miles is a misleading statistic. It is only somewhat predictive.

Argo had $4 billion or so of funding and 2100 employees.

Ford after pulling the plug said Argo was over 5 years from Level 4. Level 4 is when it will be safe to remove human drivers for virtually all conditions. Level 3 is when it is safe to remove the human driver for limited conditions.

Are the others dependent upon Lidar that much further ahead? Waymo, Cruise, Apollo are the leaders. Apollo has 500 vehicles in China and are looking to scale to 3000 in 2023.

Up to this month, San Francisco Cruise customers could order Cruise’s transport service from 10:00pm to 5:00am from about 70 active Cruise vehicles in San Francisco. Cruise is just now getting permission to operate in daylight hours. Yes, there was no driver but being able to operate in hypermapped locations at limited times was not level 4 that seems like level 3.

Cruise self-driving cars have been in nine hit-and-run accidents this year in San Francisco. Almost all nine of Cruise’s 2022 self-driving accidents have led to the collision’s human driver fleeing the area. There have been other accidents where the Cruise was hit for being too tentative.

Waymo can now provide driverless rides to passengers in San Francisco. Two days ago (November 18, 2022) the California Public Utilities Commission (CPUC) said it granted the Alphabet-owned Waymo a permit to participate in its driverless pilot program. Waymo still isn’t allowed to charge for these rides, though. The state of California requires companies to obtain a series of incremental permits from both the CPUC and the Department of Motor Vehicles before they’re able to charge for driverless rides. Waymo has been testing its vehicles in San Francisco for several years and started charging for rides with a human safety driver present in March. They are approved to driving with passengers in Los Angeles. They are still working to expand fully driverless robotaxi services in Phoenix.

Waymo, the self-driving company spun out of Google, drove 2.3 million in 2021. In 2020, they drove 2020, with 628,838 miles.In 2019, Waymo drove 1.45 million miles.

All of the dozens of self driving car companies operating in California drove a combined 4 million miles in 2021 in California. There are others operating in Arizona and in China.

Tesla’s level full self driving system still has customers operating them but they have over 35 million miles driving in the first 9 months of 2022. Tesla is switching highway driving from autopilot to the full self driving system. 160,000 Tesla owners are using the full self driving beta. Most are barely using it or not using it at all despite having access. However, the usage is about ten times higher than all of the other self driving car miles in the US combined. Tesla should have 15 million more miles in the fourth quarter and should have over 50 million in just the first quarter of 2023.

I think best case 3 years to get good enough to truly and successfully start scaling Lidar robotaxi. It is more likely to take over 5 years.

Shortage of Lidar

Only 200k auto grade Lidar units made in 2021. This is from 80 companies. The one or two suppliers used by a leading Lidar self driving needs to scale to 8 million units if they use 8 units per car for 1 million robotaxi.

1 million robotaxi needed to get to about current Uber scale. Need to get to 30 billion miles of paid robotaxi to make about Uber levels of money of about $20-25 billion per year.

Tesla is already profitable with autopilot and FSD in spite of hardware costs. Tesla chose to add self driving hardware on all cars in 2017 even if the owner chose not to purchase autopilot or FSD..

TeslaFSD Version 11 uses FSD for highway miles and not navigate on autopilot. This should get Tesla to at least 50 million miles in Q1 2023. Tesla needs to get to 1 billon FSD miles per quarter as an indication they are on the brink of robotaxi.

If Tesla has robotaxi at Uber scale in 2024 or 2025, they will have 5 million cars in 2024 and 8 million cars in 2025 if they have the $25k-ish car ramping as well as Model Y in china in early 2024. Robotaxi making $20 billion of profit would only be 30% or so of the earnings.

Tesla would get to 20 million robotaxi network by around 2028f. I expect more customers to get FSD to and join the robotaxi network. They will operate with dedicated robotaxis built and owned by Tesla. 25-50% of all Tesla could be dedicated or customer part time robotaxi. 20 million robotaxi under heavy 30-50K per car usage will be $600 billion to $1 trillion per year in net income.

Tesla alone is capable of that hard a ramp. Others need the Lidar, millimeter wave radar and other sensors to scale.

If the others are ten years behind then they are dead.

Uber had a 30 month lead on Lyft for 75% market share.

13 thoughts on “The Future of Self-driving Cars”

    • A better robotaxi or a really good full self-driving with human driver overseeing will have more miles driven. It should converge towards at least 50%+ of the miles driven for the car and eventually 98+% of the miles. Right now the Waymo cars are down around 1500 miles per car per year. This is 10% of the miles for an average person but only 3% of the miles for a taxi (taxi drives 50k+ miles per year). Tesla is at 160k possible people for FSD but only 30% or less are using it and no highway miles count. At 40M+ miles per year. Also at about 1000 miles per year per car actually using it. FSD V11 will have highway under FSD and not autopilot. Should get to 3000-5000+ miles per car by Q2 2023.

  1. I need a self driving car too come out soon because I’m disabled, If this doesn’t happen then I will have to walk on the street the rest of my life.

  2. Tele-operated vehicles is the near term future, IMO. And latency wouldn’t be a problem if aerostats were utilized instead of orbiting satellites.

    Such vehicles could also be programed to be robotically operated in emergencies– if communications is somehow cut off from the remote operator.

    • Latency isn’t an issue with satellites, either, so long as they’re not way out in geosych orbit. Starlink is running at around 20ms latency, with is too little to notice for teleoperation.

      I’d be more concerned about dropouts than latency.

  3. Meh.
    I’d rather be in a restaurant that is fully staffed by Optimus servers, bussers, hosts, and bar-tenders. How is this not a thing and how is Musk not making this a selling/ promoting point?
    Then of course…
    A fully Optimus bot retail outlet – especially something high-interaction like clothing…
    A fully Optimus retail mall…
    A fully Optimus hotel, with pool…
    A fully Optimus cruise…
    A fully Optimus amusement park – though I have flashbacks of that Simpson’s episode.

    • It’s not a thing yet because they’re still in the prototype stage. Won’t be a thing for a while after that, because they plan on proving them out in their own factories.

      • Agreed.
        We have to recall that Musk is a monied/prima-donna industrialist-engineer-type rather than a thinker, researcher, or pure engineer-type. His products/ services are cutting edge but highly standardized at a competitive price point, meaning a level of quality that is above average, out the door early, but not focussed on very detailed, bespoke applications. Perhaps then a future Apple thing?
        Must be my subconcious pining for that Westworld series that ‘jumped the shark’ oh-so many seasons ago.

        • Musk is kind of at that sweet spot where you’re enough of an engineer to advance things, and enough of a businessman to not go broke in the process. More of an Edison than a Tesla. We could use more people like him.

          • Perhaps.
            But I often yearn for further examples of the tight, brilliant team, reinforcing and vetting each other’s ideas, typically given great autonomy to pursue projects in a development-friendly businesses, as with Xerox PARC of the early 70s, Texas Instruments in the decade before, the whole Skunkworks concept (which is said to have created the SR-71, Macintosh computer, etc., at other similar places). Everett Rogers even created a list of the criteria for facilitating such a team, environment, and the likely results. Masterful.

  4. “hypermapped locations at limited times was not level 4 that seems like level 3”

    To you. Because that is not how the Musk want it to be. More miles does not translate to a pure AI driving. You cannot even see that. Only human conciousness can to that and one way to bring it in is with Autonomy maps!

  5. I think the level definitions are slightly different. Level 3 is when the user must always be at the wheel, ready to take control immediately if anything goes wrong. Level 4 is when the car can only drive itself in some conditions, but it knows whether it’s in such a condition, and if it reaches a point that it can’t drive anymore, then it can safely pull over and park until the human resumes control. So the human doesn’t have to monitor it all the time.

    So level 3 isn’t terribly useful: the user has to be at the wheel the whole time. But level 4 would be great. Even if it could only drive itself during daylight in good weather on well-mapped streets, that would still let you take a nap or work on a laptop during those conditions.

Comments are closed.