-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Waymo
- Thread starter Daniel in SD
- Start date
Alphabet might distribute their shares to existing shareholders as dividend. If Waymo needs more cash, Alphabet can give it to them, like they have in the last 10+ years. Google makes $20 billion in profit per quarter...
Perhaps an IPO makes sense at some point, but not quite yet.
You can buy $GOOG. Waymo, DeepMind with spinoffs like Reimagining Drug Discovery Process with AI - Isomorphic Labs. I got a fair bit of stock in GOOG. My largest investment and up 55% last 12 months or so...
AlanSubie4Life
Efficiency Obsessed Member
Seems like it depends on how much of their R&D cost goes into expanding the service areas. If that scales up rapidly at some point, then maybe. It’s unclear to me how they are scaling it right now and whether it is “simple” brute force.
Seems like they are also just one unfortunate accident (and a bit of dissembling) away from failure so they have to be really careful.
Regulatory hurdles could also increase.
Hard to know.
diplomat33
Average guy who loves autonomous vehicles
Waymo planning expansion of autonomous taxis to L.A.
Waymo already offers for-hire cars in San Francisco and Phoenix. It now wants to expand into Los Angeles and some suburbs in Northern and Southern California.
www.nbcnews.com
diplomat33
Average guy who loves autonomous vehicles
I am not sure what you mean by brute force. I think Waymo has been clear on how they scale. They start with testing with safety drivers. When they reach a certain safety benchmark, they allow employees to take driverless rides. When they reach their next benchmark, they open up early access to the non-employees on a waitlist. When they reach the next benchmark, they open driverless up to more people on the waitlist. And finally, they open up driverless to the general public with no waitlist. At the same, they've also started in a limited ODD. As they reach certain safety benchmarks, they've expanded the geofence and the ODD (ex: adding rain, adding highways). I would not really call that brute force.
AlanSubie4Life
Efficiency Obsessed Member
It is labor intensive. The question is whether or not the effort to expand the ODD is proportional to the increase in ODD.
It’s not like at each step of the process they are just waiting to see what happens - they’re mapping, fixing safety-critical and non-critical bugs, etc. It’s a massive effort.
I am just saying it seems like progress is slow to increase total ODD area. It has probably sped up - but is that due to greater investment (that would be pure brute force) or building on prior work (or both)? How much more will it speed up?
Just in the context of profitability, this is important.
I am not offering an opinion one way or another on how it works.
But exactly how it scales impacts when profitability will occur.
diplomat33
Average guy who loves autonomous vehicles
I would argue that the effort to expand the ODD is decreasing compared to the increase in ODD. Think about a few years back when Waymo was only doing Chandler and they had to spend a massive effort to validate the new 5th Gen and train new NN to handle SF driving. They spent months of driving around with safety drivers in SF before they could even start doing driverless. Now, they are expanding to new areas with much less effort as the 5th Gen is already validated and it is able to handle the driving in the new area with perhaps only a small amount of tweaking. Waymo was able to expand the Phoenix area quicker. They were able to launch driverless testing in LA quicker. Certainly, based on the maps proposed in their latest CPUC application, it would seem Waymo is aiming to scale much faster. How it will impact profitability is unknown at this point but I think the scaling is speeding up.
Doggydogworld
Active Member
They said something last spring about growing 10x in a year. So maybe 7m trips in 2024. Or maybe they slow down with Cruise no longer there to push them. Either way it's still a tiny business. I'd guess something like a billion rides a year for breakeven. And that may be optimistic.
This is like saying a normal taxi company would need "billion rides a year for breakeven". That's just silly.
A normal taxi company can be profitable with one car. They have to get the car investment back over three years, pay for upkeep and pay the driver.
The same goes for Waymo, except they don't need to pay the driver. They can have something 0.1 human per car instead (remote ops and service vehicles). This number will go down as the Driver gets better.
R&D cost will go down to almost zero for existing service areas and so will the car hardware at scale. The main question is if Waymo want to be an operator or just license the Driver when it reaches a certain maturity.
Doggydogworld
Active Member
Normal Taxi, Inc. doesn't have 2b annual R&D. Normal taxis aren't tricked out with expensive, finicky sensors that need cleaning, calibration and maintenance.
They're not at 0.1. Cruise's 1.5 was overstated -- it was a total number, not a constant 1.5 people. It compares to 3 full time taxi drivers rotating shifts on the same car. So a 50% labor savings. And Waymo may be better than Cruise. But they still have a huge operations staff supporting very few on-road vehicles. Remote monitors, Roadside Assistance, Customer Support, Vehicle Maintenance and all the infrastructure, overhead and management to support those people and their tools.
"Zero for existing areas" is a far cry from zero, since they'll constantly expand areas. And update their vehicles. And sensor suites. And improve the s/w. And.... well, you get the idea. Total R&D is more likely to grow than shrink. You need a lot of profitable rides to offset that R&D. Assuming the rides are actually profitable at the unit level, which I strongly doubt is true today.
A billion rides a year is maybe 100k cars with good utilization. (Today's poor utilization is a non-starter for breakeven.) A fully cost-optimized podcar doesn't really make sense for the first ~100k. That's why Waymo's next car is a modified consumer minivan. That plus Gen 6 sensors will be their workhorse from 2025 through 2028 and (hopefully) 1++ billion rides a year. It will cost a lot less than Jaguar + Gen 5, but still 2-3x as much as a normal taxi.
diplomat33
Average guy who loves autonomous vehicles
You seem to be assuming they will need to keep spending $2B a year on R&D. I would argue that R&D costs will likely go down over time because they won't need to develop the hardware and software from scratch. Remember, Waymo spent billions in R&D because they had to basically invent autonomous driving from scratch. This involved a ton of research to develop new ML, train perception, prediction and planning NN from scratch etc... Not to mention all the costs in validating the 5th Gen. But as the Waymo Driver continues to mature, they won't need to train NN from scratch or build new perception modules etc... This is not to say that R&D will be zero. Of course not. There will always be R&D because they will want to continue to improve the tech. I just think the cost will go down because the R&D going forward will be more about perfecting the existing tech, rather than building new tech from zero. And the cost of the hardware has being going down too. for example, we know the 5th Gen costs half of what the 4th Gen cost.
Now, there will be operational costs. Increasing their fleet size will cost a lot, buying and retrofitting thousands of new vehicles. But a lot of that will be one time costs. Furthermore, as the Waymo Driver matures, they won't need as many remote assistance personnel per car. That cost won't be zero but it will go down over time IMO.
All of this to say that I think the rides needed to break even won't be 1B/yr. I think that number is too high. It might be closer to 500k rides per year IMO. That's just a guess of course since I don't have the actual numbers to make an exact calculation.
diplomat33
Average guy who loves autonomous vehicles
I would clarify that it might be more accurate to say that Waymo needs to reduce R&D costs rather than they will reduce R&D costs. If you are correct that it will take Waymo 1B rides per year to break even, that would not be a realistic path to profitability IMO. Waymo needs to reduce that number in order to achieve profitability sooner. I am very optimistic that they can do it. But, they will need to reduce R&D costs IMO.
I think it is doable for the reasons I mentioned. Once the Waymo Driver is mature enough and it works "everywhere", they can shift more to perfecting the tech, rather than developing new tech from zero. So I think in the future, when the Waymo Driver can do driverless "everywhere", they can reduce R&D costs by reducing staff and focusing more on improving the existing tech, rather than developing new tech. Also, they can focus the R&D on reducing the cost of the hardware while maintaining the same performance. Also, remember that as they shift more and more to unsupervised ML, that will reduce the need for human involvement. They will be able to do more ML training with less human involvement. So they can reduce staff there and that will reduce costs. In fact, we already see the need for human labelers is much lower now as more and more ML training is automated.
Basically, I would argue that the goals of the R&D will shift. In the early days, the goal of R&D was to build perception/prediction/planning modules from zero that could drive the car. Then the goal of R&D was develop new and better perception/prediction/planning (like ML-first planner) in order to handle more difficult ODD and edge cases (like SF, rain and fog). At some point, the Waymo Driver will be good enough to scale everywhere and the goal of the R&D can be more to reduce costs and tweak performance to handle nuances in driving (ie they won't need to do major rewrites).
Last edited:
Waymo doesn't have $2B in R&D. The "Other bets" operating margin was -6B in '22. That would include the army of robotics engineers at Everyday Robots, Deep Mind, Calico and Wing + Waymo. If the cost of running Waymo in '22 was -2-3B in total, most of that would be from operations and not R&D. Perhaps 20% is R&D.
I'd argue the bulk of the R&D cost is behind Waymo and it will likely get lower from here. The sensor hardware is invented, the Driver is pretty awesome.
I agree they're not at 0.1 yet, but I think if you include all roles (excl mgmt) they might be at 0.4-0.5 operational persons per vehicle all inclusive.
Expansion also means more revenue, and the expansion cost is what it is, but that's not R&D. Scale matters a lot, obviously.
In CA alone there was 1.37 million drivers who performed at least one ride or delivery across the DoorDash, Instacart, Lyft, and Uber platforms in California. Gross driver pay, including tips, totaled $1.35 billion in Q3'21 alone. So perhaps 5B for the full year.
If Waymo can get their hands on a 20-50% of that and expand in AZ, TX and a few other sunny states like FL I think they'll do just fine while scaling out to the rest of the US.
AlanSubie4Life
Efficiency Obsessed Member
Makes one realize how far away broad rollout of driverless is. Going to be a while!
diplomat33
Average guy who loves autonomous vehicles
Why would you say this? The Waymo apparently makes an illegal move but one illegal move does not not necessarily mean that driverless is far from widespread. And Waymo may have already fixed this issue. This video is certainly not emblematic of widespread issues. Waymo does thousands of perfectly good rides every week. Furthermore, Waymo has driverless that is very safe in a growing ODD (multiple cities, all roads including highways, parking lots, airport terminals, 24/7, rain and fog). We need to look at the total picture and not just one video. When we look at the big picture, Waymo's driverless is very capable and scaling very well. I would say broad rollout of driverless is closer than we think. It is certainly not going to be a long while. Lastly, remember that driverless does not need to be perfect to be rolled out broadly. In fact, we will likely see a broad rollout of driverless before it is perfect.
AlanSubie4Life
Efficiency Obsessed Member
I don’t care about the illegal move (though why would that be programmed to do that around a vehicle that is not stationary?).
It’s the interaction of autonomous vehicles. Seems like a potential problem.
Similar threads
- Article
