NVidia has a self-driving software stack that they can drop into a 3D sim. Same stack works in the sim (ie real world has a real camera, sim world has a sim camera, real world has a real lidar, sim world has a sim lidar...) They will drive billions of sim miles with their stack.
GM bought cruise for 1 billion. Softbanks stake would value it at 11 billion. Even if GM put in some money into cruise their return on investment seems pretty good.
I think softbank isn't just a regular VC. My speculation is that it may provide other services like dealing with govt(s) that makes this offer even more compelling. Otherwise I don't see why GM would need an investment from softbank.
It’s not always just about the money. One of the most effective ways to ensure that a long-term project stays alive at a big company as people come and go is to create an external tie-up or partnership.
Since being bailed out, GM's debts have exploded once again, rapidly hitting the $100B mark with an average interest around 5%. A $2.5B cash infusion, and the added confidence, is very much something they would need.
How much do you think a robotaxi costs to manufacture?
I think it's a lowball estimate, but for argument sake, let's say they're each worth $150k. At that price $2.25 billion will get you 15,000 robotaxis, and that's far from the only expense.
Waymo has announced a plan to deploy 20,000 vehicles in 2020, and that is what GM is up against.
You just can't have enough money in the bank if you're comitted to a scaled commercial robotaxi deployment and are planning for world domination.
The investment today is only $0.9 billion with a further $1.35 billion at an unknown time in the future. This makes a present valuation of only $4.5 billion. And GM must commit $1.1 billion now. So GM will have invested a total of $2.1 billion for a share of Cruise which is valued at 80% of $4.5 = $3.6 billion.
>This makes a present valuation of only $4.5 billion.
in 2 years from 1B->4.5B (with GM original stake of 1B growing 2.5x during those 2 years(4.5 - 1.1 - 0.9 = 2.5)). This is why i love tech. We call it "only" here.
Subtext here is that Cruise's driverless product is done. And the cash wont be spent on further R&D. But on go-to-market strategy. Ultimately culminating in the public rollout of a fleet for major markets. Perhaps centered around the Bolt EV 5-door subcompact.
The seismic shift in marketing will be fascinating. From the freedom evoked by taking hairpin curves with the wind tossing your hair about. To a focus on the hardships and hidden costs of car ownership.
As passenger attention shifts from watching the road and scenery. To focusing on windowpane display screens. Expect to see branded partnerships with Disney / Netflix. As well as content specifically formulated to entertain or shop during the 20 minute car trip ;)
> As passenger attention shifts from watching the road and scenery.
Why would it shift from scenery? Not having to watch the road means you're free to take in the views. That same classic PCH car shot can be done while looking to the ocean instead of straight ahead.
I’m sorry but I disagree. Driverless cars are a pipe dream at this point. A lot of people are going to die and legislation will catch up eventually. The tech isn’t even standardized with everyone taking different approaches. The article about TI millimeter wave radar that’s was on the front page shed some light.
I’m sure this won’t be popular on HN. The Uber crash gave some insight on the real state of things. Tesla is in full PR mode even getting banned from the recent crash’s investigation.
I could not agree more. Softbank and the investment community may just be in greater fool mode. Or, to be more generous, perhaps they are taking a long term approach. But anything close to Level 4/5 is a _long_ way away, as even senior technical folks have been admitting recently.
From the abstract: "Until these problems are solved,
human beings will remain an integral part of the driving task, monitoring the AI system as it performs anywhere from just over 0% to just under 100% of the driving".
Now there's room to be optimistic or pessimistic about how much an AI can help, but the academic consensus is that level five will need many more research breakthroughs. There are some good video lectures that give an overview of the field (but not that much depth) here:
Lex Fridman's team has drawn their highly disingenuous conclusions based on studies of automated (not autonomous) systems from Volvo, Tesla and Range Rover. They didn't study a single product from a company working on an actual autonomous driving system, and of course they can't because those products are all currently in development and not available to study.
It's fair to criticize a study but when you start calling an academic researcher 'highly disingenuous', you just sound like you have a horse in the race. Did you found an autonomous driving company or something?
The cars that have been driving around Mountain View have been level 3, since there was still a driver. Waymo does have level 4 cars in Phoenix right now, which is incredibly impressive. Whether or not they are capable of fully driverless functionality in Mountain View right now is unknown, but I sincerely doubt it given it's their hometown and if they could they would.
Waymo is clearly the leader and making great strides, but they're not there yet. By the end of this year when they officially launch to the public in Phoenix we'll have a much better idea of how far along they really are.
There are two kinds of cars, the "golf cart" class ones that are level 4 and the SUV ones that are level 3. Both have been active around Mountain View for years (though the level 3 ones longer).
What makes you think the "golf carts" are level 4? Level 4 implies you don't need a driver, but I don't believe they were ever tested without drivers on public roads.
A lot of people already die everyday in car accidents. The only difference will be that we have no one we can blame if the vehicle in question does not have a driver.
I do not need an algorithm or even to think when I’m in a dangerous situation. My mind will react and I will assess the situation afterwards. Millions of years of experience has trained life. I can see the same traits in my dogs.
I believe accidents and fatalities per vehicle mile traveled are the most relevant statistics. If driverless cars can outperform humans on those metrics then they will be a great boon even if they roll over the odd cyclist or get in weird crashes.
Although I agree that the comment to which you're replying deserves a bit of dismissiveness, I can't help but think that your response is overly colored by the "speed kills" rhetoric we often hear.
I'm pretty sure, though, that the autonomous vehicle collisions that have made the news lately haven't been at "highway" speed and that that's true of most fatalities (possibly because most vehicle miles traveled are in dense cities?).
Now, that isn't to say that we've ever had 45mph or even 25mph heavy objects constantly moving around us in our evolutonary past, either, but the occasional fast predator wasn't unheard-of and a particularly fast one survives today.
Moreover, the typical difference in speed between (unpredictable) objects in traffic is much closer to human walking or running speeds.
We get into trouble when there's simply too many of them or the difference in speed gets too high.
It's an interesting fact that dogs can't learn to drive cars. Neither can non-human primates (low-speed circus acts don't count: I mean safe city and highway driving). Driving requires several seconds of look-ahead, which animals can't learn to do, and some skill analogous to tool use.
I would argue that just means we've particularly customized the driving UX [1] to humans, not that other species lack some fundamental capacity for it.
It has saddened me that we've traditionally started from a position of human exceptionalism when studying other animals and, to a large extent, still do so today, despite discoveries such as tool manufacture by the New Caledonian crow and dolphin whistles following Zipf's law. I can only imagine how shocked (shocked!) we would be by cephalopod capabilities compared to humans if we were to, say, genetically modify them to live as long as a macaw.
All that said, my point is that I don't think there's anything particularly special in human evolution to suggest that we would have more skill at lookahead. One might even speculate that we would have less than an animal that has to chase at relatively high speed in order to survive, such as a cheetah or wolf [2].
As for tool use, I do admit that, though we're not unique in that regard, it's probably still safe to say it's rare enough to be remarkable. Perhaps if we hadn't been running around for so long assuming our uniqueness, someone might have tried teaching a corvid or octopus to drive by now.
[1] including such things as engineering trade-offs in areas like suspension where one might sacrifice pure performance (and therefore, arguably, control and subsequently safety) for comfort
[2] which are, of course, not domestic dogs. Thousands of years of human breeding may well have removed any evolutionary advantage related to driving.
There should be some sensible kind of road test, but it'd be a terrible outcome for regulators to specify exactly how the technology should work for a given model year.
IIRC the robotaxi service will only be offered in Arizona for the time being, which is basically the easiest place to perform at level 4. This might be a good start, but don't be fooled into thinking the improvements required to bring level 4 into markets with reasonable amounts of rain and snow are linear; if the rest of the self driving car industry is any indication, we're nearing a plateau.
Cruise strikes me as the ideal acquisition, and a formula others should copy. Kyle Vogt and the rest of the Cruise founding team got personal liquidity; a huge influx of capital to continue R&D; access to GM's marketing, manufacturing, and talent infrastructure; but retain general independence in the pursuit of the ultimate mission of self-driving cars.
Pretty impressed with GM for having the insight to understand that arrangement was best and the flexibility to pull it off. And incredibly impressed with what Kyle and his team have achieved post-acquisition. They clearly remain as hungry to make this happen as on day 1.
really? have a friend who interviewed there pre GM acquisition. said it was a colossal joke of a place, engineering wise. and hes worked at some sh*thole startups, so i take his word for it.
honestly, it was 100% luck GM took notice of them. weren't many other alternatives around at that time. GM buying them spurred the explosion of self-proclaimed software/full stack self driving startups we see today.
So Im not as impressed with the achievements so far as I am with the absolute dumb luck these guys have had. Kudos, i guess.
It's all the same person. In addition to having a 'friend' who 'interviewed' at Cruise, they've been an 'engineer' at Cruise, a 'founder' of Cruise, and also the 'cofounder' who left.
We've banned all the trolling accounts we know about and their main account as well.
Large investments in all the major ride sharing companies + large investment in Cruise, Softbank is becoming, in the background, the future of transportation. By networking all these companies together, preventing cross-market competition and supplying self-driving tech to their players, their market power can be seen to be pretty massive.
Sometimes I wish Deutsche Telekom would go head to head with Softbank in doing these kinds of Investments and Aquisitions.
Not because I think they'd be good at it, but more for the fact that this would give Germany a real chance for a seat at the table of future tech-giants.
55 comments
[ 3.0 ms ] story [ 105 ms ] threadNVidia has a self-driving software stack that they can drop into a 3D sim. Same stack works in the sim (ie real world has a real camera, sim world has a sim camera, real world has a real lidar, sim world has a sim lidar...) They will drive billions of sim miles with their stack.
Hour 1 Minute 58 https://www.youtube.com/watch?v=95nphvtVf34
Would be curious on accidents? Wish Waymo would share?
I think it's a lowball estimate, but for argument sake, let's say they're each worth $150k. At that price $2.25 billion will get you 15,000 robotaxis, and that's far from the only expense.
Waymo has announced a plan to deploy 20,000 vehicles in 2020, and that is what GM is up against.
You just can't have enough money in the bank if you're comitted to a scaled commercial robotaxi deployment and are planning for world domination.
in 2 years from 1B->4.5B (with GM original stake of 1B growing 2.5x during those 2 years(4.5 - 1.1 - 0.9 = 2.5)). This is why i love tech. We call it "only" here.
It sounds like they're serious business, but I don't want to dilute a bet on self-driving cars with a bet on consumer trucks.
The seismic shift in marketing will be fascinating. From the freedom evoked by taking hairpin curves with the wind tossing your hair about. To a focus on the hardships and hidden costs of car ownership.
As passenger attention shifts from watching the road and scenery. To focusing on windowpane display screens. Expect to see branded partnerships with Disney / Netflix. As well as content specifically formulated to entertain or shop during the 20 minute car trip ;)
Why would it shift from scenery? Not having to watch the road means you're free to take in the views. That same classic PCH car shot can be done while looking to the ocean instead of straight ahead.
I’m sure this won’t be popular on HN. The Uber crash gave some insight on the real state of things. Tesla is in full PR mode even getting banned from the recent crash’s investigation.
https://youtu.be/c6twRCBtd-g
https://arxiv.org/abs/1711.06976
From the abstract: "Until these problems are solved, human beings will remain an integral part of the driving task, monitoring the AI system as it performs anywhere from just over 0% to just under 100% of the driving".
Now there's room to be optimistic or pessimistic about how much an AI can help, but the academic consensus is that level five will need many more research breakthroughs. There are some good video lectures that give an overview of the field (but not that much depth) here:
https://selfdrivingcars.mit.edu/
And not just any humans, for that matter: humans over a certain age who are trained and licensed.
A seven-year-old human is smarter than any AI and has better knowledge about the world, yet we don't let them drive.
Waymo is clearly the leader and making great strides, but they're not there yet. By the end of this year when they officially launch to the public in Phoenix we'll have a much better idea of how far along they really are.
Edit:
also articles like this: https://www.mercurynews.com/2015/11/13/driverless-google-car...
I think it's not considered "level 4" because of specific state exemption and the speed limit.
A lot of people already die everyday in car accidents. The only difference will be that we have no one we can blame if the vehicle in question does not have a driver.
Yeah, it's like dogs digging on paved roads -- your millions of years of experience don't apply to when the technology changes.
I believe accidents and fatalities per vehicle mile traveled are the most relevant statistics. If driverless cars can outperform humans on those metrics then they will be a great boon even if they roll over the odd cyclist or get in weird crashes.
Yeah, we've had cars and fast moving transportation for about that long.
I'm pretty sure, though, that the autonomous vehicle collisions that have made the news lately haven't been at "highway" speed and that that's true of most fatalities (possibly because most vehicle miles traveled are in dense cities?).
Now, that isn't to say that we've ever had 45mph or even 25mph heavy objects constantly moving around us in our evolutonary past, either, but the occasional fast predator wasn't unheard-of and a particularly fast one survives today.
Moreover, the typical difference in speed between (unpredictable) objects in traffic is much closer to human walking or running speeds.
We get into trouble when there's simply too many of them or the difference in speed gets too high.
It has saddened me that we've traditionally started from a position of human exceptionalism when studying other animals and, to a large extent, still do so today, despite discoveries such as tool manufacture by the New Caledonian crow and dolphin whistles following Zipf's law. I can only imagine how shocked (shocked!) we would be by cephalopod capabilities compared to humans if we were to, say, genetically modify them to live as long as a macaw.
All that said, my point is that I don't think there's anything particularly special in human evolution to suggest that we would have more skill at lookahead. One might even speculate that we would have less than an animal that has to chase at relatively high speed in order to survive, such as a cheetah or wolf [2].
As for tool use, I do admit that, though we're not unique in that regard, it's probably still safe to say it's rare enough to be remarkable. Perhaps if we hadn't been running around for so long assuming our uniqueness, someone might have tried teaching a corvid or octopus to drive by now.
[1] including such things as engineering trade-offs in areas like suspension where one might sacrifice pure performance (and therefore, arguably, control and subsequently safety) for comfort
[2] which are, of course, not domestic dogs. Thousands of years of human breeding may well have removed any evolutionary advantage related to driving.
There should be some sensible kind of road test, but it'd be a terrible outcome for regulators to specify exactly how the technology should work for a given model year.
https://www.theverge.com/2018/5/31/17412908/waymo-chrysler-p...
Pretty impressed with GM for having the insight to understand that arrangement was best and the flexibility to pull it off. And incredibly impressed with what Kyle and his team have achieved post-acquisition. They clearly remain as hungry to make this happen as on day 1.
honestly, it was 100% luck GM took notice of them. weren't many other alternatives around at that time. GM buying them spurred the explosion of self-proclaimed software/full stack self driving startups we see today.
So Im not as impressed with the achievements so far as I am with the absolute dumb luck these guys have had. Kudos, i guess.
Previous examples are this account: https://news.ycombinator.com/threads?id=tmpnam1234567
and this user _Fricken: https://news.ycombinator.com/threads?id=_Fricken, who based on choice of username seems to actually be impersonating the submitter of this thread.
As to the validity of the criticism I have no idea, but it strikes me as very odd.
edit: Fricken (no underscore) didn't actually submit this, but submitted the same article and is here in the comments.
We've banned all the trolling accounts we know about and their main account as well.
Not because I think they'd be good at it, but more for the fact that this would give Germany a real chance for a seat at the table of future tech-giants.