> Apple came out with a revolutionary design in the iphone
The first iPhone was revolutionary and ahead of its time, but copying a user experience, a grid of square icons, fancy animations and double touch is a lot easier than copying Waymo. You can't just copy Waymo by looking at it like you could by looking at an iPhone. Waymo's code is all in a black box of compiled code and how it works really is anyone's guess besides neural network and all the captors and millions of hours of simulations and drives etc.
Once a self-driving car has a breakout success, everyone will quickly know what kind of sensor package it has, what kinds of compute hardware, and general features of the machine learning approach and tradeoffs like mapping vs on-the-fly interpretation. Knowing these things will make it much quicker to catch up compared to when everyone is still exploring for a solution.
The hardware/software demands are well known. Best practices have been established. The secret sauce is in getting the whole orchestra to harmonize, and in optimizing to get through the insane workload.
The biggest hurdles are at the start, but any company that can demo an autonomous vehicle that can, on average, drive a few miles in urban traffic without getting hung up is off and running. After those initial big hurdles, it's small hurdles stretching off further than the eye can see.
Uber had never really learned the first big hurdles. They scaled too big, too fast, and the whole operation was a clusterfuck. The rumour is that Uber was having trouble getting simulation working for them. So early on with Uber, driving around Philly, they were disengaging every block or two, and 2 years later they weren't doing much better.
In the wake of their accident they've had some time to reevaluate everything, and we'll see if they've actually managed to sort out their problems.
Hi Fricken, I'm a reporter doing some research on self-driving cars, in particular some of the lingering technical and safety challenges to commercial deployment. I've enjoyed many of your posts on the topic. I was hoping to talk with you on-background, and I'd really appreciate your help. Happy to explain more if you wouldn't mind connecting. heather.somerville@thomsonreuters.com. Thank you.
Seems to me that the data collected in the miles driven is an essential component in addition to the sensor package and other hardware. Creating a hardware package is the easy thing, collecting the data doesn't seem like it'd be easier than starting from scratch.
The invisible code is a very important part of a self-driving car. If you wanted to enter this market you'd need a company with the ability to devote billions to multi-year R&D activities and hire great engineers & scientists with the right skills. Uber ATC has those things, and while they may be significantly behind Waymo it's hard to see a more likely candidate for second place. (Maybe Cruise/GM?)
Other aspects of a self-driving car business are more easily observed, including the significant regulatory and public perception risks. Here's one scenario: Waymo creates a technically excellent self-driving car, but it kills a handful of people in particularly gruesome way that causes the public to lose trust in a fully autonomous vehicle. As a result, "partially autonomous" vehicles are perceived as "just as good", and a company that has a massive ride-hailing business through which to monetize the technology has an advantage over another that has only a theoretical technology advantage.
Where would you put the odds of a regulatory or PR catastrophe causing an existential threat to Waymo?
The other self-driving efforts seem to believe that additional sensors are required, whereas Tesla is sticking to camera-only solutions. If Tesla is wrong in this, they are not well-positioned to follow as a second mover.
And Comma.ai is cameras only, so it’s not fair to say all the others believe LIDAR is necessary.
Also, just because someone has LIDAR on their test rig, that doesn’t mean they think it’s necessary for production vehicles. They could just be using it for validation. They also might start off using it, and wean themselves off as they progress through development.
Waymo I believe had explicitly said they think LIDAR is needed but I haven’t heard anyone else say that explicitly.
The last major Tesla crash in California was a clear situation where LIDAR would've prevented a crash and yet the Tesla team continues to insist that cameras and radar (sonar really doesn't help much...) are all that is needed.
Engineers will move around. You can't lock up knowledge that's in people's heads, and the basic concepts will get implemented everywhere even if they have to find different specific approaches to avoid encroaching on patents.
I'm not so sure. How strong of a moat is the data, really?
Waymo leads right now with 7 million miles driven.
If you wanted to drive that many miles in the next year, how much would it cost?
I estimate it would cost maybe $60M or so:
-$30M on cars (assuming a fleet of 200 cars driving 100 miles a day, costing $150K each)
-$5M on safety drivers (assuming $20/hr, 30 mph)
-$1M on fuel (assuming 30 mpg, $4/gal)
-$5M on insurance (no idea)
-$5M for a garage
-$5M for a couple dozen techs/engineers working in the garage
-$10M overhead, supplies, other costs?
And this total of ~$60M is with a bunch of upfront fixed costs (mostly cars, but also the garage). Year two the cost would drop in half to ~$30M.
By my math, if you had a working system and were bottlenecked only by data, you could catch up to Waymo's 7M miles for just $60M. Maybe my estimates are way off and actually it's $100M. Or even $200M. That's expensive, but I'm not so sure it's a moat when we're talking about companies with 10s of billions of cash on hand chasing a market that could eventually be a trillion dollars.
At this stage, I think if you have a working system, then by my math the cost per training mile is well under $10 per mile.
I agree with you: data is not so much of a moat as a fixed expense for anyone wanting to join in. And for the big automakers or Uber, it is not such a big cost to pay.
You did miss in your cost estimate scaling the compute costs of ML training to ingest that data. Also test courses and simulation to amplify and elaborate on tricky cases found in the data. Adding those things in adds 10's of millions to the estimate but doesn't change the fundamental analysis.
I think the real moat is in fleet networked data. If you have a lot of cars on the road and they are sharing info about strange situations to expect, it could be a big advantage for how "smart" the driving seems. I am thinking things like broken stoplights or badly placed traffic cones. A networked solution can alert other cars of the puzzling condition and the interpretation. Then later cars passing that location can proceed more confidently than if each vehicle has to work out an interpretation itself.
Arguably it's not a pure ML problem where you can just learn what to do from human drivers. You need to know what the alternatives are, which actions are potentially dangerous etc. So you run your "beta" system and record when the safety driver takes over. Then you improve the system from that and drive some more. These iterations will take a lot of time, and you can't just scale it with more engineers (see The Mythical Man Month etc).
Maybe. It took them a long time to build up the current one so I am not sure it was only 60M. Also, what about the machine learning model they built from it?
A lot of people are critiquing the iPhone example specifically, but consider that Google itself was way beyond the 2nd search engine on the internet... more like the 5th at least, just counting ones with significant traction like (from memory) Webcrawler, Excite, Yahoo, Alta Vista, etc. And search engines are black boxes.
I think in general, it's easy to overestimate the long-term value of a first-mover advantage.
Does anyone think this space will be regulated across all companies to create an autonomous standard for safety reasons under the department of transportation etc or simply that the autonomous tech gets into all cars across manufacturers via some partnership?
Just like how we see all newer cars coming out with the same types of tech e.g. lane change alerting or backup camera monitoring or Android auto/Apple carplay etc, as standard features? Or is automation tech more of a first mover monopoly in and of itself since it's so unique and patentable?
> Just like how we see all newer cars coming out with the same types of tech
A lot of the current driving assist features are made by the same companies (e.g. mobileye, bosch) and then installed/licensed to car manufacturers. The same thing could happen with self-driving depending on who develops it first.
Offtopic: a grid of square icons, fancy animations and double touch is a lot easier than copying Waymo
You just described Palm, Blackberry, and Windows Mobile...all of which Apple was copying. The iPhone was evolutionary, but not revolutionary by any means.
Back on topic: It's more about letting the first mover spend the money on figuring out the business plan. The second mover can then jump in with a more efficient business model based on the lessons learned from the first mover.
While the scale of effort required in autonomous mobility is indeed much larger, there are still parallels to the iPhone: Consider the years it took Google to build a digitizer stack and mobile graphics pipeline that could match the scrolling responsiveness of iOS.
It's not just about the compiled code, which shouldn't be impossible to reverse engineer, although very difficult. It is more about the data, which will be the uncopyable advantage.
We aren't talking about AI or how the data is collected.
The point is that this data would have been collected by the first-movers the entire time they are on the road. Such data would not be made available to newcomers.
If a new self driving car company would join the scene, they would need to gather all this data again themselves, somehow. This is the "uncopyable advantage" mentioned by OP.
The first iPhone was revolutionary and ahead of its time, but copying a user experience, a grid of square icons, fancy animations and double touch is a lot easier than copying Waymo. You can't just copy Waymo by looking at it like you could by looking at an iPhone. Waymo's code is all in a black box of compiled code and how it works really is anyone's guess besides neural network and all the captors and millions of hours of simulations and drives etc.