Tesla has driver-assist, it's like comparing a bicycle to a motorcycle. Tesla autopilot is doing a lot of harm to its brand and self-driving industry by using half-baked driver-assist on a production car and implying that it's self-driving. Your Tesla won't stop in front of a brick wall if your life depended on it.
That's a semantic distinction without real meaning. Tesla autopilot may have been designed/deployed to different standards of autonomy, but it's fundamentally solving exactly the same technical problem.
And it's actually doing quite well, though given human psychology any crash is going to be turned into "doing a lot of harm to its brand". Alas, we can't automate away knee-jerk conservatism.
> Your Tesla won't stop in front of a brick wall if your life depended on it.
You folks are going to come after Waymo with your hatchets too, once a few juicy wrecks show up on Twitter.
Are you kidding me? It is far more than a semantic distinction - Tesla is actively telling people that their car is "on autopilot". People, perhaps out of the same level of boosterism you're demonstrating here, believe them. Some growing percentage of them die.
It isn't going well - they're actively harming the future of self driving in the world.
> It is far more than a semantic distinction - Tesla is actively telling people that their car is "on autopilot".
Uh... what Tesla is "telling people" (or what they are "calling" their product) is the very definition of a semantic distinction. The point was that the technical problem being solved is the same one Waymo is working on.
> Some growing percentage of them die.
Are you kidding me? (See what I did there?) That statement isn't remotely true. There have been two fatal accidents involving Tesla autopilot. Sales are growing rapidly quarter after quarter. How exactly do you support that?
That kind of fake analysis (driven by your personal feelings about the company or its owner, presumably, and not numbers) is what's harming the future of self driving in the world.
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.
Pilots don't let planes on autopilot take off and land. They don't just turn it on and go for a beer. Same with boats - there's still a human despite the fanciest autopilot on a ship, especially when navigating a harbour.
However the term is overloaded - consumers think it means something different. In the consumer understanding, autopilot means "go to sleep I've got this."
But it's never been that. Tesla's mistake is using a term people might misunderstand, not misusing rhe term themselves.
I took commercial as ‘operating a service’ like Uber wants to do. I’m not aware of Tesla having similar plans. Either way they’re not fully autonomous yet.
No, but they do have an active fleet and an early rider program, with some cars having no safety driver. In those instances, it's effectively the 'commercial program' without actually charging riders -- but all of the rest of the infrastructure is in place.
The most important evidence for imminent commercialization is Waymo's recent order for 62,000 minivans. If we're assuming a conservative $100k a pop once they're loaded up with all the technology required, that's a $6B investment in depreciating assets. They wouldn't do that unless they were extremely confident that the technology is ready, and given their safety record, I'm inclined to believe them.
The frequently-repeated caveats certainly apply -- they'll only be operating in limited geographies with ideal weather; there will probably be roads, conditions and maneuvers that are restricted. But having actual autonomous vehicles on the road performing commercially useful tasks is by far most important step. The rest will come incrementally.
I agree with you about it being exciting, but Waymo haven't actually ordered the minivans - the press release says up to 62,000 minivans, so the actual size of the order is not clear.
These deals are often structured with outs and progressive volume so you’re probably correct. Realistically they probably ordered 500 minivans with a deal to get more based on some targets.
You're spot-on here. The 62k number is almost certainly the negotiated volume pricing amount they negotiated as a maximum under the current contract to get favorable pricing. I would be highly surprised if the actual order volumes are more than a few hundred at a time for the next couple years.
Or they just want you (and more importantly their potential fleet/commercial customers) to think this. Whether they are ready or not, that kind of announcement further marginalizes the credibility of competing efforts. It could also be interpreted as a move to counter the growing public perception that self-driving cars are "not ready."
No matter how much money Alphabet has lying around, 6B is a ridiculous amount of money to spend on credibility. They are confident in their product and confident in return of their investment. Tesla is doing a great job at screwing up public perception on self-driving cars, when it itself is not a self-driving car (it's a glorified driver assist).
60k FCAs plus 20k Jaguars x 50 trips per day (Waymo's estimate) = 4 million trips a day.
While there are all sorts of funky technological and regulatory factors to consider, unimpaired this is enough to displace the entire rideshare/taxi industry across the southwestern United States.
Mentioned on page 2 multiple times as the closest competitor. But they get something like 1/20th the monthly miles Waymo does so they’re only close in relative terms.
From the article:
> That leaves Cruise, which got a $2.25 billion cash infusion from Softbank on Thursday, as seemingly Waymo's top rival.
> Cruise has begun testing its cars at significant scale...
Perhaps one of the most interesting parts of this deal are the fact that the order seems to be on the scale where things appear to be ready off-the-shelf. Does an order of 62,000 imply that Waymo is close to having hardware that can be quickly adapted to multiple cars (e.g., the Jaguar and now the Chrysler) that can be produced at scale? Or have they managed to convince manufacturers to heavily customize their existing vehicles (seems unlikely)?
That's something I'm not sure is even on the horizon for any self-driving car effort other than Tesla.
I imagine it might be easier than you are thinking. I think it’s mostly a matter of mounting all the sensors (camera, lidar, laser), securing the main computer(s) somewhere, routing between everything. For the electronics I think most(all) modern cars are electronically steered so inserting your computer into the steering is probably straightforward. To get and control most of the other auxillary features it might be enough to insert yourself on the CAN bus?
I’m sure there are a lot of details in the above, and a lot of stuff I’m missing, but I’m not convinced it’s something a group of good engineers couldn’t do in a few months or a year. Certainly not something that is not even on the horizon for most companies.
The Jaguar I-PACE deal envisioned designing a custom version of the I-PACE so that the sensors and computing hardware could be installed at the factory. I haven't seen specifics on the Pacifica deal but it seems likely they'd take the same approach.
Unless I have missed something they are getting the cars over several years. A modified car (presumably during the manufacturing process), over several years, is very much reasonable. Radar and camera is part of the production of the modified car, LiDAR and compute system snaps into modified body. Waymo keeps control on their most important hardware. Seems very doable over several years.
On that particular aspect, I'd bet money on Cruise, having been bought by GM, who has more expertise in manufacturing automobiles, far beyond what Tesla has accomplished to date, simply by virtue of being an incumbent car maker.
waymo started collecting lidar data so much earlier, it would be shocking if they weren't ahead! they're ahead in so many ways, too. some of those google vehicles I used to see driving around mountain view were so friendly and round, making constantly awkward if perfectly safe stops like a student driver, you felt empathy for the cars like you were watching "brave little toaster." they're taking their time on a complex problem. meanwhile uber and tesla were racing to be the first to earn the inevitable "robo-car kills human" headline. there's a tenacity to that, but it's not winning any fans. by staying mostly in the background so far, waymo will likely be perceived by the public as more of a lovable underdog, even though it's an alphabet brand and not at all an underdog.
Are Waymo somehow much better at handling corner cases, identifying obstacles and their trajectories correctly and not having to disable emergency breaking to stop the car from randomly breaking at speed? Or are they just more careful/skilled with their marketing and publicity? Are they only planning to deploy the vehicles in specific, fairly controlled conditions, rather than claim general self-driving capabilities?
The main point is that Waymo had a 4-6 year head start over virtually everyone else in this industry. So it's not surprising that their technology seems significantly better! Obviously we don't yet know how much better, or if it's good enough to improve on the average human driver. But they seem to be pretty close.
Not only the head start but they weren't a business, they did deep research for the sake of it. Not to sell it right away. Many said that mobileye and others were very crude systems. I think the idea was that having real mileage in Teslas and similar experience would accelerate improvement .. so far it didn't help AFAIK<
The downside to their head start is that they had to build their system from scratch, and there was a great deal of trial and error.
The upside is that they had the very best, most passionate and talented roboticists in the field working on it, as nobody else was hiring at the time. For every company entering late in the game, securing top tier talent is the biggest hurdle.
Given they have the weight of Google behind them it wouldn’t surprise me if they are better at figuring this stuff out.
But they’ve certainly been much more cautious and mature. I imagine they recorded street view driving for years and used that as additional training data. It seems pretty clear they’re not pushing the envelope of their own abilities nearly as far as Uber, sticking to what they’re confident they can do as they keep trying to learn more.
If waymo launches a self driving car either for sale or as part of some ride sharing network, will it be trouble for Tesla? Will their stock tank? Could be trouble for the model 3 production line.
I've been pondering the economics of this and wondering what they will end up charging.
Assumptions:
* $100k / car
* 250k miles / car before replacement (I'm assuming these aren't fleet vehicles which could go for longer)
* $3 / gallon for gas
* 30mpg
* 10k car / year of opex
* 250k miles / car / year
Per mile costs are:
* $0.10 for gas
* $0.40 for capex
* $0.04 for opex
For a total of about $0.54 / mile. The biggest win is likely getting vehicles that are fleet capable to reduce the overall cost / mile, followed by switching to the all electric vehicles like the i-Pace.
You basically just derived the IRS deduction rate for personal vehicles used for business. While not a perfect proxy for costs (which is a complicated question), it's a reasonable back-of-the-envelope estimate for operating a vehicle--including depreciation--without a driver.
Heh yeah. I assume that the cost structure will be something like $0.70 / mile for single person rides, and then dramatically dropping it for carpooled rides.
If the cost structure is accurate, then this could really take a way the need for a second car for most people, and maybe even a primary car for folks that are ok with renting for their occasional weekend jaunt.
If they can get significantly higher mileage out of each car and/or reduce the cost of LIDAR too, then that will make a large improvement in profitability / car.
I'm fairly skeptical about how soon door-to-door autonomy arrives but figuring out the cost basis seems fairly straightforward. You also have to account that there will be some deadheading to get to you, to the next fare, and to park/get cleaned/fueled.
I suspect Uber/Lyft/Zipcar/etc.have already significantly cut into the second car in a lot of greater urban area households--or will when the next refresh cycle comes around. Anything that further reduces on-demand costs just increases the effect. (Though it's probably worth at least noting that current rideshare (where available) pricing is <= 2-3x base operational costs. That's not nothing obviously but it's not a difference between a luxury and too cheap to meter.)
I did similar calculations but using a different methodology.
Cheapest ZipCar is $8/hr and has the same cost structure (i.e. ZipCar also pays for car, insurance, gas, maintenance etc.).
So let's generously assume it's $10/hr.
That's 3x cheaper than taxi or Uber so they instantly kill taxi/Uber where they operate.
Assuming an average in-city trip of 15 minutes, that's comparable to $2.5 ticket for a bus in SF.
Obviously, an autonomous car can take up to 4 passengers, so if you agree to share, the price can be 2-3x cheaper. That kills buses where they operate.
And long term (5-10 years out) when they operate at scale, they should be cheaper than ZipCar (say $5-$6 per hr).
Why?
Because their cars will have higher utilization.
Because they'll all be electric and electricity is 3x cheaper per mile than gas even if you buy for utilities (as opposed to building your own solar farm nearby to supply most of the electricity)
Because insurance cost will be dramatically lower (near zero).
Because they'll buy them in millions so cost savings due to manufacturing scale.
Because they'll be constantly re-designing cars to make them cheaper to make and more reliable as opposed to trying to guess what color or form factor will be popular tomorrow with humans or making high-performance version.
Yesterday afternoon I had two encounters with driverless cars in SF within a few minutes. Both showed how hilariously far from ready these things are.
First, I came to a 4-way stop here where a Waymo car had arrived and stopped maybe 1-2 seconds prior. Waymo was on 15th heading East, I was on San Bruno heading north. Nobody else was at the intersection: https://goo.gl/maps/S2GmNTuWNDq
I stopped fully and gave it about 2 seconds to start moving since it had right-of-way. It made a one or two false starts, then I just gave up and went through the intersection because it was clearly having issues. Looking back in the rear view I saw it did the same thing with the next car. Maybe it didn't think it was a 4-way stop or something?
A few minutes later, I was on Bryant just past 7th outside the Hall of Justice in dense traffic that was moving quickly (as people do on that street). A Cruise Automation car (I think?) was in one of the right lanes next to me and, for whatever reason, decided it wanted to get over to the left quickly. When it didn't get a good opening after a couple seconds, it more or less stopped in the middle of the road to wait for an opening. Unsafe and almost caused an accident with the cars behind it.
Of course human drivers do stuff like this or worse all the time ¯\_(ツ)_/¯ It's going to be interesting to adjust to all the ways these things behave unpredictably (relative to what we'd expect from human drivers).
> I stopped fully and gave it about 2 seconds to start moving since it had right-of-way. It made a one or two false starts, then I just gave up and went through the intersection because it was clearly having issues. Looking back in the rear view I saw it did the same thing with the next car. Maybe it didn't think it was a 4-way stop or something?
> A few minutes later, I was on Bryant just past 7th outside the Hall of Justice in dense traffic that was moving quickly (as people do on that street). A Cruise Automation car (I think?) was in one of the right lanes next to me and, for whatever reason, decided it wanted to get over to the left quickly. When it didn't get a good opening after a couple seconds, it more or less stopped in the middle of the road to wait for an opening. Unsafe and almost caused an accident with the cars behind it.
If you asked me ten years ago what a future with self-driving cars would look like, it would be exactly this. I wouldn't have believed that we'd have the hubris to put them in situations in which someone could be hurt or injured, however.
I personally don't need around-town self driving. I really wish highway and congested highway automation would be worked on first, it is way way way way easier to do.
Automated taco bell drivethru? I can do that for now.
49 comments
[ 2.1 ms ] story [ 108 ms ] threadThey’ve bought 82k vehicles recently. They’re clearly up to SOMETHING soon.
That was a large chunk of the article.
And it's actually doing quite well, though given human psychology any crash is going to be turned into "doing a lot of harm to its brand". Alas, we can't automate away knee-jerk conservatism.
> Your Tesla won't stop in front of a brick wall if your life depended on it.
You folks are going to come after Waymo with your hatchets too, once a few juicy wrecks show up on Twitter.
It isn't going well - they're actively harming the future of self driving in the world.
Uh... what Tesla is "telling people" (or what they are "calling" their product) is the very definition of a semantic distinction. The point was that the technical problem being solved is the same one Waymo is working on.
> Some growing percentage of them die.
Are you kidding me? (See what I did there?) That statement isn't remotely true. There have been two fatal accidents involving Tesla autopilot. Sales are growing rapidly quarter after quarter. How exactly do you support that?
That kind of fake analysis (driven by your personal feelings about the company or its owner, presumably, and not numbers) is what's harming the future of self driving in the world.
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.
Pilots don't let planes on autopilot take off and land. They don't just turn it on and go for a beer. Same with boats - there's still a human despite the fanciest autopilot on a ship, especially when navigating a harbour.
However the term is overloaded - consumers think it means something different. In the consumer understanding, autopilot means "go to sleep I've got this."
But it's never been that. Tesla's mistake is using a term people might misunderstand, not misusing rhe term themselves.
If you're in Phoenix, you can even apply: https://waymo.com/apply/
The frequently-repeated caveats certainly apply -- they'll only be operating in limited geographies with ideal weather; there will probably be roads, conditions and maneuvers that are restricted. But having actual autonomous vehicles on the road performing commercially useful tasks is by far most important step. The rest will come incrementally.
Exciting times.
https://venturebeat.com/2018/05/31/waymo-orders-62000-fiat-c...
While there are all sorts of funky technological and regulatory factors to consider, unimpaired this is enough to displace the entire rideshare/taxi industry across the southwestern United States.
I'm sure softbanks $2.2 billion investment recently is worth a confidence boost in their technology.
That's something I'm not sure is even on the horizon for any self-driving car effort other than Tesla.
I’m sure there are a lot of details in the above, and a lot of stuff I’m missing, but I’m not convinced it’s something a group of good engineers couldn’t do in a few months or a year. Certainly not something that is not even on the horizon for most companies.
Are Waymo somehow much better at handling corner cases, identifying obstacles and their trajectories correctly and not having to disable emergency breaking to stop the car from randomly breaking at speed? Or are they just more careful/skilled with their marketing and publicity? Are they only planning to deploy the vehicles in specific, fairly controlled conditions, rather than claim general self-driving capabilities?
The main point is that Waymo had a 4-6 year head start over virtually everyone else in this industry. So it's not surprising that their technology seems significantly better! Obviously we don't yet know how much better, or if it's good enough to improve on the average human driver. But they seem to be pretty close.
The upside is that they had the very best, most passionate and talented roboticists in the field working on it, as nobody else was hiring at the time. For every company entering late in the game, securing top tier talent is the biggest hurdle.
But they’ve certainly been much more cautious and mature. I imagine they recorded street view driving for years and used that as additional training data. It seems pretty clear they’re not pushing the envelope of their own abilities nearly as far as Uber, sticking to what they’re confident they can do as they keep trying to learn more.
I would say it’s the opposite. If anything I would say their autopilot tech is weighing the stock down due to its constant liabilities.
Assumptions:
Per mile costs are: For a total of about $0.54 / mile. The biggest win is likely getting vehicles that are fleet capable to reduce the overall cost / mile, followed by switching to the all electric vehicles like the i-Pace.If the cost structure is accurate, then this could really take a way the need for a second car for most people, and maybe even a primary car for folks that are ok with renting for their occasional weekend jaunt.
If they can get significantly higher mileage out of each car and/or reduce the cost of LIDAR too, then that will make a large improvement in profitability / car.
I suspect Uber/Lyft/Zipcar/etc.have already significantly cut into the second car in a lot of greater urban area households--or will when the next refresh cycle comes around. Anything that further reduces on-demand costs just increases the effect. (Though it's probably worth at least noting that current rideshare (where available) pricing is <= 2-3x base operational costs. That's not nothing obviously but it's not a difference between a luxury and too cheap to meter.)
Cheapest ZipCar is $8/hr and has the same cost structure (i.e. ZipCar also pays for car, insurance, gas, maintenance etc.).
So let's generously assume it's $10/hr.
That's 3x cheaper than taxi or Uber so they instantly kill taxi/Uber where they operate.
Assuming an average in-city trip of 15 minutes, that's comparable to $2.5 ticket for a bus in SF.
Obviously, an autonomous car can take up to 4 passengers, so if you agree to share, the price can be 2-3x cheaper. That kills buses where they operate.
And long term (5-10 years out) when they operate at scale, they should be cheaper than ZipCar (say $5-$6 per hr).
Why?
Because their cars will have higher utilization.
Because they'll all be electric and electricity is 3x cheaper per mile than gas even if you buy for utilities (as opposed to building your own solar farm nearby to supply most of the electricity)
Because insurance cost will be dramatically lower (near zero).
Because they'll buy them in millions so cost savings due to manufacturing scale.
Because they'll be constantly re-designing cars to make them cheaper to make and more reliable as opposed to trying to guess what color or form factor will be popular tomorrow with humans or making high-performance version.
First, I came to a 4-way stop here where a Waymo car had arrived and stopped maybe 1-2 seconds prior. Waymo was on 15th heading East, I was on San Bruno heading north. Nobody else was at the intersection: https://goo.gl/maps/S2GmNTuWNDq
I stopped fully and gave it about 2 seconds to start moving since it had right-of-way. It made a one or two false starts, then I just gave up and went through the intersection because it was clearly having issues. Looking back in the rear view I saw it did the same thing with the next car. Maybe it didn't think it was a 4-way stop or something?
A few minutes later, I was on Bryant just past 7th outside the Hall of Justice in dense traffic that was moving quickly (as people do on that street). A Cruise Automation car (I think?) was in one of the right lanes next to me and, for whatever reason, decided it wanted to get over to the left quickly. When it didn't get a good opening after a couple seconds, it more or less stopped in the middle of the road to wait for an opening. Unsafe and almost caused an accident with the cars behind it.
Of course human drivers do stuff like this or worse all the time ¯\_(ツ)_/¯ It's going to be interesting to adjust to all the ways these things behave unpredictably (relative to what we'd expect from human drivers).
> A few minutes later, I was on Bryant just past 7th outside the Hall of Justice in dense traffic that was moving quickly (as people do on that street). A Cruise Automation car (I think?) was in one of the right lanes next to me and, for whatever reason, decided it wanted to get over to the left quickly. When it didn't get a good opening after a couple seconds, it more or less stopped in the middle of the road to wait for an opening. Unsafe and almost caused an accident with the cars behind it.
If you asked me ten years ago what a future with self-driving cars would look like, it would be exactly this. I wouldn't have believed that we'd have the hubris to put them in situations in which someone could be hurt or injured, however.
Automated taco bell drivethru? I can do that for now.