The argument is that self-driving won't work because Uber and Tesla had well-publicized crashes. But I don't see how this tells us anything about other, apparently more cautious companies like Waymo. There seem to be significant differences in technology.
More generally, machine learning is a broad area and there's no reason to believe that different applications of it will all succeed or all fail for similar reasons. It seems more likely there will be more winners along with many failed attempts.
The argument is that self-driving won't work because Uber and Tesla had well-publicized crashes. But I don't see how this tells us anything about other, apparently more cautious companies like Waymo. There seem to be significant differences in technology.
Yes. I've been saying this for a while. Waymo's approach is about 80% geometry, 20% AI. Profile the terrain, and only drive where it's flat. The AI part is for trying to identify other road users and guess what they will do. When in doubt, assume worst case and stay far away from them.
I was amazed that anyone would try self-driving without profiling the road. Everybody in the DARPA Grand Challenge had to do that, including us, because it was off-road driving and you were not guaranteed a flat road. The Google/Waymo people understood this. Some of the others just tried dumping the raw sensor data into a deep learning system and getting out a steering wheel angle. Not good.
And Waymo is not close to what Tesla does. Different strategy coming at it from different sides.
The question is not, who has the first taxi on any public road anywhere, but who first can launch 10000 of autonomous taxis and make money.
Waymo so far is just a gigantic money fire, and even if you assume their autonomy works, many people have made arguments that the businesses case is really not very solid at all. Waymo has incredibly few ours driven on public roads outside of a limited area.
Self driving is mostly a software problem, and to solve general driving, a gigantic learning problem. How is gone solve that first and scale it, is open.
I would put my money on Tesla, and I wouldn't invest in Waymo if I could.
Tesla's approach thinks white trailers are open sky. The river comparison is not unreasonable at all.
Almost all of the startups in the space are following the Tesla approach of just throwing machine learning at the problem which means almost all work in the space suffers from novel input producing unpredictable results. This is one of the things that has killed trained and evolved systems in the past and the fact that few of the companies in this space are even trying to manage it (either by building interpretable models or by using models and using ML to do parameter fitting) is a good indicator that the whole business is either a fraud or is built on the premise that with enough data or, for the un-cautious and unaware, enough simulation, the problem solves itself.
I think what we are actually seeing is that Waymo (where, in ten years, they might have a solution) and Tesla (which is mostly worst-in-class but as a company is the master of hype) drove hype around self-driving. Then Uber and Lyft latched on (because they needed a story to paper over their terrible economics) and pushed it even higher. The Otto thing and Cruise's acquisition made VCs pay attention.
So these companies are acquisition bait for the assumed-clueless big auto companies. They will not deliver self-driving cars. At best they will deliver next-generation enhancements to emergency braking, etc.
I’m pretty sure no-one at Waymo believes they have solved self-driving, and likewise to say they aren’t the current leader in demonstrated autonomous capability is fanciful.
Maybe I haven't seen enough large-scale corporate failures (and the associated damage control, which is how it seems you are characterizing this post), but it feels like you're making this prediction on the basis of not enough data.
I can imagine there are other scenarios where Waymo PR makes this post, even though Waymo has not totally failed at producing self-driving cars, but perhaps I'm being insufficiently cynical.
Waymo has self driving cars with a tiny amount of paying customers in a tiny are losing 10s of millions of dollars every year.
Tesla on the other-hand is MAKING money with self driving technology already.
What something is called doesn't really matter, I bet on the company that has 100ks to millions of cars with advanced self-driving computers and hardware on the road in daily use over a company that has a couple 100 cars in Arizona any day.
The complexity of self driving is in the complexity of the road network in the real world. The only way to solve it is to have millions of cars learning all the complexity and finding the corner cases.
The real money in self driving is not offering ride-share in small geo-locked areas.
The honest argument is that we have seen Tesla’s technology exposed to unsupervised and uncontrolled adversarial conditions across hundreds of wildly diverse cities in the USA, whereas we haven’t seen Waymo vehicles doing anything outside of curated geofenced areas or curated marketing videos. Right now if you dropped a Waymo and a Tesla on an unsealed road in Michigan, one of them will drive at least as well as a human learner driver and the other will probably refuse to drive.
I agree that Waymo could well be far ahead of Tesla, but there isn't enough information in the public domain to say this with confidence. We don't have the ability to make a proper comparative assessment.
Did you work at waymo or is this speculation? Some seems a bit like a Markov chain built from random self-driving objections ("never cover enough cases", "schoolchildren"), and some seems wrong (e.g. [1] mentions an expanding use of deep learning at waymo starting in 2013, well before Urmson left).
The usual big corp objections -- unsatisfied with agility, level of control, level of payoff and recognition if it succeeds -- seems a lot more likely.
Just because Tesla is profitable and making money by selling vehicles does not mean they are on a better path to engineering a self driving system than Waymo.
The opposite is also true, just because Waymo does not make money does not reflect the capability of their self driving systems. Saying "Waymo has not proven N, because they don't make money on any of these things." doesn't make any sense, and is not even true.
I can go to downtown Phoenix right now and request (and pay for) a fully self-driving ride from point A to point B. Teslas can not reliably complete any self driving route without any disengagements.
We are discussing who is closer to realizing a fully self-driving system, not who runs a better business.
Waymo hasn't been in any serious accidents yet because they have human safety drivers and because they don't put their vehicles in situations where they can cause a serious accident even with a safety driver.
In that sense, expecting that Waymo is doing better than the rest and it's going to get self-driving cars into production is a bit like expecting that a person who learned how to walk on a line drawn on the floor is going to perform tight-rope walking because they're more careful than all those other idiots who actually tried to walk on a rope.
The thing to keep in mind is that self-driving is hard, much harder than tight-rope walking. It's so hard that it remains unsolved, currently, and no amount of careful application of non-solutions will result in a solution.
Hi, why in your analysis you spoke only about the companies that are not doing so well in self driving leaving out waymo success story?
They are already have been hauling passengers without a safety pilot since last October. I guess without the minimum problem otherwise we would have heard plenty in the news like it happened for Tesla and Uber accidents.
Is it not too convenient to leave out the facts that contradict your hypothesis?
I've said it before, and I'll say it again: Waymo's approach to self driving is a fool's errand. The only reason it works is that they have mapped out cities with straightforward roads and sunny weather to millimeter precision, and have hardcoded vehicles to operate in those conditions.
Compare this to Tesla's AI-first approach, aggregating training data from millions of drivers across the US. There's no overhead to introducing Autopilot in a new city it's never been in before--it just "works," with no expensive 3D mapping required, by extrapolating cheaply acquired vision data from other cities.
Elon really got this right from the beginning, but Waymo's aggressive PR and marketing have pitted public opinion against Tesla, as is very clearly evident even in these HN comments.
I think the primary fallacy in the theory is confusing companies like Waymo with companies like Tesla, and mingling their (in)capabilities.
Both have a very different approach in trying to get to self-driving, and when combining what both can't do, it would seem nobody can do anything. And whilst both throw out their fair share of "demo videos" (=advertising), they have different track records for what they can (and actually do), and what they can't (but claim to do).
Except Tesla doesn't have the sensors and hasn't demonstrated the chops.
How many fatal accidents involving Waymo vehicles are there? Sales aren't even remotely associated. I support it based on "autopilot" versus true self driving systems. Which is safer?
"Fake analysis" won't matter when the NTSB bans all self driving systems because a few companies keep screwing it up. My hope is that they only target the egregious offenders - Tesla and perhaps Uber.
Driverless cars are coming along just fine. Waymo is making steady progress. The sensors are getting better. The problem comes from all those "fake it til you make it" startups, Uber and Tesla being the worst. Both have killed people.
This is mostly about sensors and geometry. Machine learning has a role, but only in target identification. That's how Waymo does it. The fake it til you make it crowd had the fantasy that you just hook up some cameras to a machine learning system, train it, and you have self driving. Doesn't work. Machine learning is way too dumb.
You can maybe identify "traffic light", "car", "pedestrian", and "deer" with machine learning. That's just used to guess what they're going to do next. It's not used to decide if they're an obstacle. Obstacle detection and avoidance is all sensors and geometry.
Also, "self driving car", "electric car", and "transportation as a service" are all independent. All will be available, but not from the same companies.
I don't think this proves anything about the state of self-driving tech generally. It just proves that Tesla doesn't have any, and that camera vision alone can't do much. Waymo's car doesn't drive like this.
Tesla aren't the only people trying to make self-driving cars, famously Uber tried and Waymo looks like they're slowly succeeding. Competition can be useful, but it's not a panacea.
More generally, machine learning is a broad area and there's no reason to believe that different applications of it will all succeed or all fail for similar reasons. It seems more likely there will be more winners along with many failed attempts.
reply