I don't know if this is a fair point of view but from an outsider's perspective it seems like any company that isn't Waymo is simply burning money. What companies actually have a reasonable chance at developing a usable autonomous driving system that won't kill me while I cross a street and won't kill the driver while on the highway? Anyone have a list of real contenders free of the marketing bs?
Consider the smartphone revolution. Apple came out with a revolutionary design in the iphone. The Android team was already up and working in that area and were able to pivot quickly to mimic the iphone. Both Apple and Android became huge businesses.
Being prepared as a second mover once a trailblazer demonstrates a good path can be good business. It even has some advantages over trying to lead and possibly wasting lots of R&D money on dead ends.
Several important differences between self-driving and smartphones:
1. Driving automation is orders of magnitude more expensive
2. Improperly-tested vehicles are much more likely to have fatal consequences, meaning a 90% perfect smartphone is OK, but 90% perfect driving software would kill people
3. It's much, much easier to clone an iPhone than Waymo hardware + software
You're both right. It is much harder to clone Waymo. Doesn't mean that there isn't a second mover advantage.
Plus the amount of patents involved in this mean even if you're burning money down a dead end you might stumble upon an insight / patent that would help regardless of the approach taken. Often time with engineering a good design comes out of a known shortcoming of some other subsystem.
> Driving automation is orders of magnitude more expensive
Only if you pretend it didn't take decades of gradual improvement to make the iPhone possible, and ignore the epic manufacturing cost of the components. Basically only if you ignore everything that has / does go into smartphones. It took hundreds of billions of dollars of investment into the underlying iPhone components and their near ancestors, to make the iPhone possible.
What's the total cost of the specific driving automation segment so far? A fraction of all the money plowed just into semiconductor fabrication development and construction over the last ~25 years so we could have smartphones at all. Then move on to the investment into storage, software, glass, communication chips, security, glass, etc.
It'd be equivalent to pretending that we're going to leap from Tesla's hands-free to full automation without anything inbetween. The iPhone didn't materialize out of nowhere.
Agreed, the cost of the original iPhone vs existing phones was likely “orders of magnitude more” but took a while to proliferate across the competitive space.
Couldn't you just mount your sensors next to an existing car's sensors, measure the car's actions then tell your network "do that" based on the mounted sensors input?
If by "do that" you mean mimic what a real driver would do for a specific set of sensor inputs, that is precisely ML tries to do.
To understand what the difficulty is, it's important to consider that the size of the sensor input is very large. Don't think of it like twenty range finders around the car, rather a 360 degree medium resolution color + depth image (about 0.5 million data points coming at 30 fps).
It's difficult because you will never encounter the same set of sensor inputs twice, so you can't treat it like a search space problem. Once you've accepted that, you're in AI/ML territory where you might try to reason about what the closest set of known sensor inputs and action would be (classical AI, expert system), but that is impractically difficult with as 0.5 million dimensional search space, or train an ML model to 'reason' about the sensor space to make a decision about the appropriate action.
Approaches using a small number of sensors can do automatic breaking and smarter cruise control, but haven't been seen to be successful about navigating and making strategic decisions. The current belief is that more can be done by using denser sensors and more data and seems to be the case. There are people working on reducing the sensor density requirement, but the main focus right now is building a successful and safe self driving car, regardless of sensor and compute costs.
According to Waymo most of the miles driven are simulated ( 2.7 billion miles in 2017). That's an order of magnitude more than actual miles (25K per day)[1]. And even the actual miles mostly don't have any user input.
Because of this I'm leaning towards thinking waymo isn't trying to mimic actual human input.
Movies are filmed at 27fps so the reasoning is humans have high confidence that they aren't missing any significant information between the frames, it should possible to make a 'mental model' of a road scene at the same fps to human skill level.
In the future we'll likely have super-human spatial and temporal resolution, right now more improvements have been gained from highest possible spatial resolution with minimal plausible temporal resolution.
>> Movies are filmed at 27fps so the reasoning is humans have high confidence that they aren't missing any significant information between the frames, it should possible to make a 'mental model' of a road scene at the same fps to human skill level.
I hope there is a better, more technical explanation that ML researchers are using, because as someone who is somewhat of an expert on human vision and building products around it, this foundation is godawful if it is to be taken at face value. Which again, I am sure this is a simplification. Or at least, that's what I am telling myself.
This whole driverless car thing reeks of more vaporware, more public-yet-profitless unicorn companies promising to promise to change the world and aside from surface level consideration of edge case accidents with pedestrians (hasn't one of them already killed or seriously injured a pedestrian??) there isn't much deep talk about less straightforward issues such as (former) members of the trucking industry sabotaging these new fleets or how liability and insurance is REALLY going to work out.
File the promises and the problems under fiction because it appears to be more important to keep the world order, its financial system and these ridiculous media darling fluff piece corporations alive while they bleed money.
And no, im not closed to the idea of successful work being done on automated driving but 30 fps, WTF? too much going on in the larger context of the world, this shit isn't happening in 2020 or 2024 or whatever else many might say.
Look at it this way: human ability to act is (for elite gamers) 300 actions per minute. That's 5fps. So with 30fps the AI could theoretically already have 1/6th the latency of the most-responsive human drivers
They are not actually responding at 300 actions per minute to changing input, a large percentage of those clicks are constant selections of team shortcuts.
This is even worse. Humans do not process 300 APM. They merely are limited physically to outputting 300 APM. You have no idea what the brain's capacity to process and analyze information that led to the output of 300 APM. If you think 5 FPS is the capacity of the brain's ability to process vision... well, don't make a driverless car, please.
We move and react slowly but we respond to info which comes at a much higher rate. I can notice the individual frames at 5 FPS. I know I'm not getting enough info.
I was going to post that the impact of higher FPS is likely too low to justify needing 4x higher processing power. But then the difference in reaction time between 30fps and 120fps is O(30ms). At 60mph that translates to almost 1 meter in stoppping distance. Tough call.
It isn't just stopping power. It is fidelity, confidence, and completeness of data. A camera quickly moving across a panorama opening/closing the lens every 1/30th of a second loses a ton of data compared to one opening its lens every 1/120th of a second.
Yea, but then you have to have 4x the processing power It's pretty easy to scale frame rates once you have stuff figured out, but I can see very good reasons to not have to try to get up to 120+fps right away. In addition, I'd imagine the ML they're using probably has a lot harder time distinguishing valid motion when the motion is 4x as small.
Yup, I didn't learn much from my image analysis class but the professor made sure that over fitting and under fitting concepts were firmly lodged in our heads.
Isn't trust a massive problem? If I'm getting into a car without a driver, I have to trust that the system is smart enough to get me to my destination safely.
The problem with Uber is that they already operate on a trust deficit. They've had a long history of ignoring local rules and had a string of high-profile exits - including the CEOs - after workplace harassment issues.
That's not exactly a company I want to trust with my life.
Waymo, otoh, is operated by a company known for its "braininess". Self-driving is a computationally challenging problem. I trust Google to solve it much more than Uber.
There's also the fact that Waymo doesn't have to be a profit center for Google at all. They can afford to incur losses for years until they get the technology perfect.
For Uber, self-driving cars HAVE to be a profit center. If the company's existence depends on it, can I trust them to not cut corners?
> Isn't trust a massive problem? If I'm getting into a car without a driver, I have to trust that the system is smart enough to get me to my destination safely.
Car manufacturers have decades of experience in shifting danger from inside to outside. Passengers are not the ones we need to worry about.
For me? the kicker would be insurance. Is the company wiling to insure it? I mean, like enough to cover my lifetime expected earnings? then sure, I'll get in the car.
Driving is very dangerous, and most drivers don't even carry enough insurance to pay for a couple months time off if they hit me. This, essentially, make my injuries an externality that isn't priced into the cost of transportation.
For that matter, if I'm injured in an accident and it's my fault, I don't have a lot of coverage for myself; I mean, through work, I have a long-term disability package, but that's still not great compared do what I'd make were I not disabled, and disability, as I understand it, is a huge pain in the ass when it comes to you being, you know, 'partially disabled' - still able to work, but not able to make nearly what you made before.
Carrying adequate insurance insures that the total cost of the injuries caused is rolled into the operating cost of the vehicle, and helps to price it rationally.
> 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.
I dont think anyone is misunderstanding how two or more players can make it wrt developing new tech. I think the concern is about what that process looks like when the tech being developed can kill people if it goes wrong.
> any company that isn't Waymo is simply burning money
Waymo/Google is winning and has self-driving vehicles all over Chandler, Tempe and Gilbert Arizona at any given time.
For the most part it is going well, sometimes they act too safe but that isn't a bad idea for now, there have been a few complaints of Waymos going when fire engines or safety vehicles are on the road not slowing and in normal traffic slowing more than needed, or cars partially in a lane and slowing or stopping for them too much. I think the more grandma driving any self-driving is right now is good.
The biggest win of Waymo which is a competitive advantage and huge bonus is that by being Google/Alphabet, they also have access to tons of data and the best maps / map data i.e. speed limits, traffic, routes, dark/light changes to behavior, weather detection, data outside the CV/LIDAR that may be hugely beneficial. Other companies are going to have to pay for that or use Google, which they will collect immensely on in addition to use for free.
Think of that Tesla crash where other Teslas commonly had the same issue by drifting towards the ramp, Waymo/Google could correct that quicker it seems with their access. They need to be autonomous but also a neural network that can learn from other vehicles. They also need to tie into other systems like weather, day/night, emergency vehicles, traffic, routes etc etc. That Uber crash in Tempe, with day/night data and known pedestrian trails nearby + across the road, it may have been more cautious through that area and Waymo has that data with Google. Waymo is leading in multiple ways not just the self-driving part.
This. Google's economic moat is their data. Proprietary data they've accumulated for the past two decades for various products and strategies. We should see a serious google competitor but we don't. Bing tried it and failed because they just cannot match Google's superior data.
So whoever is trying to emulate Tesla is in for a nasty surprise, Tesla is automation 2.89. The last miles is the hardest and requires an ever increasing amount of data. while safer than humans driving Tesla's autopilot failures have done far more damage to the brand than accounted in the stock price.
Google/Alphabet/Waymo figured out their end game was to build the data that their competitors will not be able to emulate easily or costly.
So far Uber is spinning its wheels and Tesla has lost focus of the killer feature of their car, coming dead last in terms of automation ranking when compared to other major luxury brand manufacturers. But this is HN, and they love Don Papi Musk
Toyota plays closer to their vests, so I'm not sure we even know enough about what they have.
I suppose I should clarify since I was downvoted, Forbes wrote a story "what Ubers crash tells us about Japan's silent strategy for driverless cars"
"it illustrates a very important distinction between how U.S. and Japanese companies are approaching the research and deployment of autonomous vehicles.
Japanese firms have maintained a much lower profile in autonomous vehicles than their American counterparts, and many outside the industry assume this lack of publicity is a sign of lagging technology. Speaking with several industry leaders, however, makes it clear that this is not the case.
"While American firms have been grabbing the headlines, Japanese firms have been making steady progress behind the scenes. The fact that they have been less anxious to promote that progress outside of Japan is more a reflection of go-to-market strategy than the state of the underlying technology.""
This is an interesting perspective but knowing Japanese culture, it's more to do with risk aversion and losing face by announcing something and then having it blow up in public. Very different environment.
In the US you can commit white collar crimes and still have people invest and give you the benefit of the doubt. There's no sense of shame rather glee and glorification, the i-get-money-fuck-you attitude.
In Japan, people will commit suicide for way less or even perceived wrong because there is a great deal of shame imposed on the individual for fucking up. Tough to innovate when failure is so taboo, people are willing to remove themselves from earth.
I think that is a bit of a trope if somewhat true one. Thing is that Toyota and all of the other Japanese car companies are invested in self driving and have been pumping money and buying/investing/contracting every ML related shop they can get their hands on in Tokyo to work on it. I recently talked to a guy who just finished his PhD focusing in NLP, started a company trying to do some clever autonomous agent app, got given bunch of cash to work on self driving instead(not by Toyota)
Japan has some unique advantage for go to market self driving cars. Japanese highway system is MUCH MUCH simpler to deal with then US one. If Toyota or Nissan or whoever can come up with a system that can drive you from Tokyo IC to wherever on Honshu following highways(mostly 2 lane, clearly marked, meticulously maintained..) this is going to be a huge deal. And few of them do have the whole thing mapped out.. we do know that.
Is Waymo might win.. RIM also seemed unassailable as was Nokia.
Driving on highways is quite expensive in Japan - it's fact generally more expensive than trading the shinkansen. Driving within cities is a different thing.
It depends, but yes it is not free. It is however quite cheap if you have a full car. So for example anyone going for weekend with family, highway is cheap and is MUCH faster then normal roads.
Japan also has a developed highway rest areas, you will often see people sleep in the car there, often whole families in vans that been outfitted for it. There is washroom, restaurants, sometimes shower and even hot spring, they are also clean, might have local produce being sold, alcohol etc
Here is a priceless use case for a salaryman from Tokyo. Get off work, get drunk with coworkers, get in the car with family at 10pm on friday. Car delivers you to the closest michi-no-eki to a ski resort and parks itself. Wake up at 6-7am, get to the resort by opening lift, grab breakfast at conbini drive to resort. There are lots fo places less then an hour off highway. Being able to sleep and being driven on highway would be incredible.
Company that delivers that feature in the family van will destroy competition until they can duplicate it.
If you're solo yes, once you're two people it tends to be on par, 3 or more car is cheaper. Every family I know owns a car. The highways are full of cars and trucks.
Purely from a financial perspective, I think Uber can't possibly admit that. They are gunning for an IPO at a valuation where they are most definitely selling the dream. That dream has to include autonomous vehicles.
I think it's not nearly as exciting a dream if they're just buying the tech like everybody else. Aside from a strong brand and a lot of customer relationships, Uber doesn't have much in the way of assets. Google's got a stronger brand [1] and a lot more customer relationships, as do many others. If Uber is just one of many companies slapping labels on other people's tech, I'm not sure what makes it worth $75-100bn.
Waymo will be first to market and will not likely sell their self-driving software to other would-be rideshare players (or their suppliers) for several years. All other OEMs, Tier 1s and startups in this space will be quick to follow with a comparable self-driving system by 2021. With all the new regulatory hurdles in mobility (as seen with scooter mania), I doubt that Waymo will just be able to throw money to scale fast enough to push out future competitors.
Once those self-driving systems hit around ~99.99% reliability, they will be indistinguishable from each other. Therefore, self-driving systems will be commoditized and sold to any vehicle manufacturer. Generally speaking, existing Tier 1/2 suppliers already plan to do this.
From there, anyone with $25M can probably just lease a bunch of vehicles, software, parking space and routing software to launch a regional TNC. Given a single region focus, these new regional TNCs should be able to outcompete Uber/Lyft. Remember that Uber/Lyft's greatest advantage is their supply of drivers; riders will happily download an app that delivers cheaper rides.
Today, it doesn't make sense to build core self-driving technology unless you have a strong resume to do so. Just wait 4-5 years and you'll be able to buy it off the rack (more or less). I am very happy that Waymo isn't the only one building because otherwise we wouldn't have strong enough market forces to drive down the cost of mobility.
>From there, anyone with $25M can probably just lease a bunch of vehicles, software, parking space and routing software to launch a regional TNC. Given a single region focus, these new regional TNCs should be able to outcompete Uber/Lyft. Remember that Uber/Lyft's greatest advantage is their supply of drivers; riders will happily download an app that delivers cheaper rides.
If you are willing to lose money like uber was, why wouldn't you be able to launch a competitor with drivers? I bet drivers would download a new app to make more money even faster than passengers would download such an app to save money.
I mean, if anything, a system where you own (or lease in an inflexible way) the cars makes your up-front costs higher; I think new driver incentives right now are on the order of a few grand; a lot less than the cost of a new car with top-end tech. I mean, double the signup bonuses of uber, and you'll quickly get drivers to open your app, and you've still spent a lot less up-front.
Of course, your long-term profitability looks a lot better if you can just own the cars, but make no mistake about it; uber's moat is that you have to lose money at first to build up a critical mass of vehicles (and that uber is willing to continue to operate at a loss)
Directly owning/leasing the vehicles rather than paying owner-operators per mile might be cheaper long term, but it doesn't lower the up-front cost of getting a critical mass of vehicles on the road.
The upside to solving driverless is so huge that companies feel they can't miss out on it. It's still very early in the game. Uber is a great example, everyone assumed with all their VC money that they were serious contenders to Waymo...then the crash happened and we get a (literal) glimpse under the hood and realize things aint so pretty. These operations are opaque by necessity and from an investor perspective, it's not a guarantee that Waymo won't have their own setback which becomes Uber's saving grace (however unlikely that seems at the moment).
I doubt it. It seems like a top down strategy outside of Waymo. Operate in silence training on vast amounts of data while your competitors are splashing in VC money thinking if they throw money at it they can solve it.
I don't see how the market is winner take all and it's certainly big enough for more than one player. Everyone who comes after whoever's first to market will have the advantage of a clear regulatory pathway and an example of what works, simplifying many (though clearly not all) decisions.
A Chinese competitor doesn’t even have to be truly competitive with Waymo to stand a chance.
Also, I don’t think Waymo is guaranteed to be the first to reach fully autonomous driving. What I know of their approach (which is fairly little, but wat does anybody outside Waymo know for sure there?), to me, seems they are building not cars that are autonomous, but networked. That has advantages (if one car notices a bump in the road, one riding the same road 10 minutes later can slow down), but also has the disadvantages that their software may depend too much on always having that network knowledge available. That may leave space for a competitor to grow in regions where roads barely exist.
Waymo also spends a lot on things that, in a few years, autonomous cars may not need not do at all.
For example, countries that want to speed up adoption of autonomous cars may make detecting traffic signs and traffic lights easier (e.g. by superimposing a high-frequency signal on LED traffic lights or by giving cars read access to a city’s traffic light infrastructure), make la e detection easier (e.g. by embedding cat eyes in the center of lanes) or require cars to better signal their intentions. If that happens, a lot of the work Waymo does on detecting all kinds of ‘static’ infrastructure becomes worth less.
How long would changing infrastructure and cars to better support self-driving take? Many cars will continue to be on the road for many years. What are the costs of adapting all of them?
IF that really happens, it would take at least a decade and Waymo's lead from network effects, more mature tech, more self-driving data, etc would be enormous by then.
A few weeks or so, for the first stretches of road. Self-driving cars can be useful even if limited to limited stretches of road (for example, many people would pay for the ability to read a book or play a video game in the 30 minutes they spend on the highway to work, even if that initially were the only stretch of road where their car were autonomous), and such roads could be made off limits to older cars.
And cost? Not that large, compared to the cost of building roads.
> a usable autonomous driving system that won't kill me while I cross a street and won't kill the driver while on the highway?
There are no such systems, even organic ones. I despair that the conventional wisdom here among the "hacker" intelligencia has pushed the burden so far. Autonomous vehicles are worth it if they are safeER than human drivers, not 100% free of accidents.
And it's noisy data anyway. Waymo is notable for the fact that their measured fatal accident rate is still zero, where the others are hovering in the small integers. That's not the same thing as a true safety advantage (though I'll buy that they're way ahead technologically).
One piece of evidence is disengagement rates. In 2017, Waymo led with ~6,000 miles per disengagement. Cruise was in second with ~1,000 miles per disengagement. Also, if you look only at the last few months of 2017, Cruise had improved to around ~3,000 or so as I recall. The other players testing in CA are far behind Waymo and Cruise.
Of course, there are many caveats to comparing disengagement rates. These are self-reported by companies, and each company may have different policies about which disengagements are serious enough to report. Also, not all miles are equal. Cruise likes to emphasize that its SF miles are more difficult than Waymo's more suburban Mountain View miles.
To "get confused" is not the same thing as giving up completely and needing another individual to rescue you, in traffic. I think people are overly minimizing what disengagement signifies.
It might be interesting to know in what percent of disengagements the car was actually doing the right thing, and only the human driver thought it was wrong and intervened anyway.
Those are actually already removed from the disengagement count if the operator can show (with simulation) that the car would have done the right thing without intervention.
If it didn't work this way you would be incentivizing operators to train their drivers to take risks in order to hit disengagement metrics which would be bad.
I've seen a few where the car was right and a few where the driver was right, but my assumption is that the car just got lucky because, in my experience, the Cruise cars are very hesitant to make left turns.
>in my experience, the Cruise cars are very hesitant to make left turns
I'm sure unprotected (i.e. no arrow) left hand turns remain one of the more difficult routine scenarios to fine tune. What works in (I assume) Chandler Arizona and San Francisco are probably two very different things. A large busy city like San Francisco simply requires a degree of aggression--Pittsburgh lefts, cutting through fairly small gaps, etc.--that would be inappropriate in a calmer and less busy environment.
Well one failure was at 19th and Mission where left turns are illegal, another was at 13th and Folsom which I think is a protected turn, another was on 11th street where the drivers decided to make an illegal u-turn.
The only times I've come across Cruise vehicles they've been under manual control — and their drivers certainly aren't making much of an attempt to follow the traffic laws. Illegal left turn off of Mission? Why not? Illegal u-turn on 11th St? Sure. Run a red light because the car wouldn't make a legal turn on its own? Of course!
Note: I've already got a negative opinion of Cruise regardless of their purported abilities. I followed up on some headhunter spam from Cruise a year or so ago. I bailed before even getting to a phone screen because as persistent as they were, they bailed or were late every time I agreed to one of their proposed times.
Waymo is killing it in the press releases, but what's to say other players aren't doing just as well while not telling the world all about it?
Also, is this really going to be just one "Self Driving Vehicle Systems" market? I can imagine Waymo being great in one segment, and other companies doing well in others.
>> What companies actually have a reasonable chance at developing a usable autonomous driving system that won't kill me while I cross a street and won't kill the driver while on the highway?
What chances does a regular human driver have in killing someone else while they cross the street? Its pretty high, given the number of casualties per day. All the people on their phones while driving and being otherwise distracted it's an enormous risk... Chances will never be 0, but autonomous cars have the potential to be safer then majority of human drivers.
Way too many people die in traffic, but per mile an unimpaired human is actually not that dangerous. In the US there are about seven deaths per billion(!) miles and that includes drunk drivers. It's not trivial to beat human drivers.
At least you're honest. Low number statistics so far but AV's have been worse per car per year compared to human drivers. And you certainly can't trust Uber's engineering department given they literally raised the detection threshold because their computer vision sucked.
There's still plenty of execution risk ahead for Waymo and a narrow opportunity for a very focused competitor to catch up to their technology. I don't think Uber is that competitor and I don't think that competitor exists, but it's possible. Cruise would be the obvious candidate and they have a big war chest from Softbank.
Tesla is selling autonomous-but-don't-quote-us but that doesn't qualify them as a strong contender. The autobahn system isn't large enough to be difficult to map out, and highway driving is the easy case.
You're right. But Audi (with BMW and Daimler) acquired the only provider that can compete against Google Maps in completeness and accuracy. Traditional car makers work in their own ways, which might look outdated from the typical point of view on HN (and justly so). It's pretty hard to estimate how far they've reached.
AFAIK Audi's driver is legally allowed to keep hands off the wheel in specific circumstances. Meanwhile Tesla's driver is theoretically driving even when "autopilot" is engaged.
Highway driving is indeed the easy case. But even if autonomous vehicles only ever worked on the highway, that would still be a world changing technology.
I am surprised Toyota would want to work with Uber at this point - it seems Uber's ills are piling up faster than their successes. Could this be a desperation play where if Toyota doesn't partner up with somebody, they won't have a part in a driverless future that looks like it will belong to Waymo and friends?
I wish they had invested 500 million into buying a self-driving startup or something instead. Or just opening up a new AI branch for Toyota / Lexus. It's not like people en masse are driving around in self driving cars while asleep at the wheel yet so the race for self-driving is still a turtle race. They have the resources imho to do great things, maybe they're afraid after the software issues where cars would keep accelerating, but that shouldn't stop them from getting in on the game.
Ah well, I'm nobody anyway, just a guy hungry to see tech companies compete and see less of tech being monopolized between a few companies.
"Driver error or pedal misapplication was found responsible for most of the incidents.[28] The report ended stating, "Our conclusion is Toyota's problems were mechanical, not electrical." This included sticking accelerator pedals, and pedals caught under floor mats."
I love the concept of electrical cars but that has to come with a leap forward in software quality and electrical engineering quality, which I am not really seeing happen.
The idea of Quality is still, basically, alien in the software field. Some of that can be blamed on 'the marketplace' but then again, I believe that if people are going to call themselves "Engineers" they should take up that old mantle of the public responsibility inherent in that title, as it was developed after much blood and horror was spilled over the failed bridges and railroads of the 19th century.
Yeah I remember reading about the bug here on HN sometime back. Maybe it was a mix of both mechanical issues and software issues to some extent? Not totally impossible. Heck wouldn't surprise me if one end said "don't worry about that, the software will handle it" and the other end said "don't worry about that, the hardware will handle that" kinda thing. But that's just me speculating, which does nobody any good.
This video is an excellent analysis. I love the detail these investigators went into to uncover the software development practices during the early 2000's at Toyota. I really hope the automobile industry has taken this to heart and is creating more process around the software that our lives depend on.
They already have Toyota Research Institute, which they've invested 1 billion and got the Gill Pratt to head so I not sure why they want to team up with Uber now.
Unfortunately, it's not unusual for large corporations to invest millions or even billions into new R&D projects that yield no results.
This is especially true for high technology projects, and only becoming more of a risk as technology becomes more complex.
Can totally see it happening with something like self-driving cars, which is currently still a black art (remember how even Uber apparently needed to filch some IP from Waymo to even get its own project going?)
To digress a bit, if development and research is becoming more and more expensive, that would seem to imply increasing concentration of wealth is necessary to keep tech progress chugging along. To build pyramids, you have to have pharaohs...
The skills necessary to "win" driverless cars (i.e. first/second place) are scarce as is, and ridiculously competitive. There will not be a long tail of driverless car companies, so it makes more sense to invest in an existing one that already has momentum, than to take up the challenge from scratch yourself.
I agree wholeheartedly with this sentiment, but I doubt they are letting the sticking accelerator software bug affect their positions on mobility in the future.
It appears that autonomous vehicle progress has an interesting moat where being good at it (the ability to drive a lot of miles) allows you to get better faster (drive more miles, more quickly). It is similar to a network effect, but with competency rather than number of connections.
Does anyone have a name for this type of advantage, or know of examples in other areas?
> It appears that autonomous vehicle progress has an interesting moat where being good at it (the ability to drive a lot of miles) allows you to get better faster (drive more miles, more quickly). It is similar to a network effect, but with competency rather than number of connections.
This doesn't seem to be the case. Look at the California disengagement statistics over cumulative miles - it looks like it's logarithmic or something, certainly not accelerating. Or look at Waymo in Arizona: they are driving, what, 70,000 miles per month and have racked up cumulative millions there already; if there really were some sort of compound interest or exponential improvement, then after they solved it well enough to go live and started racking up all those miles, self-driving cars should be on the verge of being solved & going global unless you postulate an enormous distance in between 'good enough to drive customers around Arizona' and 'good enough to drive customers around everywhere'.
What ills do you speak of? The negative media narrative is not a full reflection of reality, rather what will attract the most eyeballs to monetize in this current media cycle.
To summarize what happened - Uber's autopilot makes decisions about when emergency braking is necessary. But the software was braking so often on public roads that overall it's driving was erratic.
So Uber knowingly disabled emergency braking in all cases.
This essentially turned the car into a missile on our public roads.
at the time of the accident, the autopilot detected the pedestrian. Calculated an imminent collision, and initiated emergency braking. Instead of actually breaking, the message to break simply went to the logs.
Combine that with a safety driver that was looking down at the console and not up at the road - and the fatality was the result.
--
It is completely beyond me why there isn't a battery of government saftey tests that vehicles with autopilot must pass before they get approved for use on our public roads.
How many times have you seen footage of a crash test dummy safely landing in an air bag? I want to see footage of crash test dummies getting pushed out in front of autopilot cars - and hopefully, never getting run down.
> It is completely beyond me why there isn't a battery of government saftey tests that vehicles with autopilot must pass before they get approved for use on our public roads.
Because the inefficiencies of government are catching up with them as they'd rather right each other (different parties) rather than be productive and move the country forward.
Technology always outpaces legislation. It's been that way since the telegraph and earlier. Legislators can't rule for stuff that isn't there, and tech changes spec all the time. Once there actually are fully-autonomous vehicles on public roads, the rules will come together. For now it's humans, just like every other vehicle.
Everything that came out after that last fatality points to Uber trailing everyone else in terms of functionality and sacrificing safety to appear further along.
I hope this is the strategy instead of Toyota wanting to use Uber on their cars as the main source of AI. To this day Uber hasn't done anything to impress me and probably most people. Their corporate culture doesn't help either.
Their reckless (compared to Waymo) behavior with self-driving car testing is certainly more than negative press. It seems to reflect their entire culture.
I think this isn't just about self-driving tech (I believe Toyota already partnered with Waymo on that given the old Google self driving cars were Lexus RX and Prius) - but this is also about car sharing. I think car sharing might be a bigger and sooner threat to car manufacturers than self driving cars. By partnering with Uber Toyota gets to hedge.
Do you have other information about the self driving fatality? While its a tall order, the self driving cars should perform better than humans when it comes to reaction times in incidents like this.
The car didn't do anything wrong, emergency braking was disabled and relied on human intervention.
Now why it was disabled is another question. Most likely because their system isn't anywhere near ready.
The only value I see from the whole of Uber is know how and practical examples of how not do solve this. The fatality didn't change anything, it is (and was) painfully apparent that it was only a matter of time.
The Volvo collision system had been disabled, the Uber system didn't detect the woman crossing the road, and the driver was watching Hulu or similar on her phone. Uber had also gone from teams of two doing the autonomous testing to teams of one.
Oh yeah, and Uber released a super contrasty picture to justify their car's behavior. Less biased sources showed that there was plenty of visibility on that stretch of road.
If you're looking for sources, Ars covered everything in pretty good detail.
This isnt true, Uber system detected obstacle, categorized it as bicycle and send emergency braking signal .. which went straight to /dev/null and logs.
A huge fraction of cars driven by Uber drivers are made by Toyota. More often than not, you'll get picked up by a Prius, Corolla, or Camry. So to some extent, this is about maintaining market share.
Toyota has an existing partnership with Uber for leases to drivers.
If Uber's thesis - that app-hailing will supplant car ownership - is correct, then car manufacturers have an incentive to either create their own hailing networks or partner with existing ones.
> As filed with the U.S. Securities and Exchange Commission on June 25, 2018
> Toyota and Uber Technologies, Inc. have entered into a memorandum of understanding to explore collaboration with respect to ridesharing. As part of the partnership, the companies created new leasing options in which car purchasers can lease vehicles with connected terminals from Toyota Financial Services and cover their payments through earnings generated as Uber drivers
So not their first rodeo, Toyota sells cars after all
Also Toyota Connected is making quite a big splash in Europe at the moment with a London office having been established March 2018 to soak up local talent.
I thought the investment was to open a channel for selling more Toyotas. They're going to offer preferential lease deals that can be paid through Uber earnings.
I think Tesla is highly underrated in this space. All Tesla cars now have sonar and radar capability built in, in addition to the extensive camera system.
As someone who frequently commutes in Teslas, I would say Tesla is closest to actually deploying something close to self-driving to the general population out of all the current self-driving competitors. Furthermore, I imagine they're collecting data on all the Tesla cars currently being driven.
I have seen the model updates substantially improve the auto-pilot capabilities with every update. And as someone who works with lots of quantitative data on a daily basis, I think there is a fundamental difference between training on billions of miles of simulated data vs. actual data. Simulated data may reduce overfitting, but there is still scope to overfit compared to raw data gathered from the field. Simply put, you won't generalize as well to data that's been simulated vs. field-generated data.
> you'll still be wasting your time if you need LIDAR
This is not for certain. I'm quoting Tesla's CEO, but it's suggested that LIDAR leads to optimizing for local minima. It means that LIDAR can get you 90% of the way there, but there's still going to be a lot of situations where LIDAR might not be capable.
The problem that needs to be solved is vision. Humans only have vision, and they can drive with it. Radar and LIDAR are nice-to-haves, but until you solve vision, you haven't solved self driving.
"Solving vision" is a 10x more harder problem than building a practical self-driving car with LIDAR. We might get it working with LIDAR in 2-5 years, but without LIDAR in 20-50 years.
Arguably Alphabet (Google + Waymo + Deepmind + co) is the largest group of AI and ML experts in the world. If someone is going to figure out general computer vision, then it's likely to be them - and if they are working on LIDAR first then there must be a good reason why.
I believe Tesla is promoting LIDAR-less systems just because it's expensive bulky hardware, and they can't possibly sell cars with it at a reasonable cost. Calling cameras and radars "full self-driving hardware" is a good marketing tactic, but without any guarantees that they can build the software for it in the next 20 years.
The issue from my (admittedly little) experience is that it's all about edge cases and reliability.
It's easy to make a demo with cameras only but to deal with the countless unimaginable edge cases, you need LIDAR.
The other factor is that you need damn near six sigma accuracy for ML algos for robust SDCs, and while it's theoretically possible to match LIDAR with cameras and current DL, it's impossible to do so with the needed 99.99 percent accuracy for autonomy.
No they aren't. Teslas only self drive on the highway, which is great, but wouldn't last a minute in a more urban scenario with crosswalks, lights, tight merges, etc. And radar is not good for these scenarios either, because of the huge amount of false positives with the system.
It doesn't matter if you're not profitable, don't have the technology, or have huge risks - if people with money believe you, you can make the future you want (and get extraordinarily rich in the process). Seems to be the theme of the tech industry lately. At least the end "goal" is a bit more appealing than the Magic Leap.
I must be in the minority here, but the AR future is really compelling to me. The amount of money that Magic Leap was able to raise without a real product is astounding. But I do see wearable computing as the future. Of course we all hope it doesn't become a steady stream of marketing directly into our retinas, but I joined tech because I wanted to help build the future, not just watch it happen.
I am actually stunned how there was no displacement of the current input devices.
We fitted keyboard onto our screens, AR devices normally come with a joystick or something.
I have high hopes for similar projects to Tap Keyboard or the Myo armband, that continue to evolve on something that IMO has large implications on how we can evolve as a society the same way smartphones did. A wearable input device that was _good_ (defined by not being cumbersome, probably even fashionable and functional, as should match current input) would be for me the extra step .
I think there's some interesting work done around natural-user-interfaces (using a kinect to control software, etc). I think this is probably a more promising direction than new custom hardware - because hardware is expensive, and quality hardware is far more so.
I mean, if you could use a pen and paper to interface with a computer, you're standing on the shoulders of centuries of experience about how to produce a nice pen-paper experience, at a really cheap price. A newcomer in the world of interface hardware has to overcome the fact that everybody finds a new interface horrible for years, aside from the crazy technical and industrial challenges of producing something at a price/performance ratio that's better than entrenched technologies.
Computer/console input has been really resistant to change. Even the Wii controller, which was something of a fad at one point, pretty much faded away. Six axis controllers is another one that never went mainstream though they were somewhat popular for a time.
There have also been various alternate input styles for phones but nothing's ever taken off.
If anything the amount of money they were able to raise and resources they were able to throw at the problem proves that its at minimum 10 years away, barring some ground-shaking advances in displays and sensors. With billions of dollars they ended up with a clip-on computer, a 50 degree FOV "lens", no compelling demo and the same judder the hololens had. It seems much closer to being at the VirualBoy stage than the Oculus/Vive.
i hope you will actively help to ensure that the future that is built is the one you want to be built, instead of just hoping that everything will magically work out. A lot of old-school nerds (including yours truly) had the same ideological blind spot as your comment, and look where that's gotten us. a god damned cyberpunk dystopia.
I actually can imagine a future AR being a lot more compelling than VR. But it would take a lot of significant advances both from a hardware and software/data perspective to be useful. There's also a pretty significant question of whether it caan overcome the glasshole factor for mainstream use.
I'm not saying that Kalanick et all aren't going to be very well off in any event, but all this cash they raise at eye popping valuations is going to have liquidation preference...and that's scary for a company with serious cash flow issues like Uber.
people with lower liquidity preference (from earlier rounds or common share holders) usually have some opportunities to exit in secondary markets at each of these funding events.
It's only logical that management raises money when money is cheap. That way the company has extra breathing room to survive and pivot if necessary. And if it turns out they don't need the cash, they can always just return it to investors.
I think there is more to it. For example, when Uber was talking about getting into self driving cars there was really no reason to believe they could take a lead besides having a boatload of money to buy the tech or people.
Driving a vehicle is the most dangerous thing most people do regularly. Not just to themselves but in terms of potential of hurting or killing others.
Software is the most unreliable thing that most people use regularly.
Combinging the two is kind of crazy. Especially since no one has even begun to demonstrate a self driving car that can handle inclimate weather (like, say, the 5 months of snow covered pavement for a large part of the USA).
I think people underrate human drivers. When you take into account the massive amount of miles driven each day, the amount of crashes isnt that bad, and currently self driving cars have much more fatalities per mile than humans.
The leading cause of transportation deaths are, by a long mile, automobiles. Automobile fatalities also have the highest rate of people killed per miles traveled. I think human driving ability is overrated, but I agree with you that self driving cars aren't doing too great either
This makes no sense to me. They already have TRI which has billions in funding and poached the former head of the DARPA challenge. Why do they need Uber?
This is quite strange. Toyota has a research center in SC and Boston (TRI) and has invested about 100M in PFN.
This seems to indicate the complete lack of progress on this front. This'd not be surprising. I think Renault Nissan is the only one with a working system.
There are reasons to think that Toyota's automated driving strategy is a good one. Wrote about it here https://www.economist.com/business/2018/05/19/toyota-takes-a.... I'd assume that the Uber partnership is about mobility and ridesharing. One of Toyota's main focuses with autonomy is on making the driving safer and more comfortable. Gil Pratt talks explicitly about improving the experience for ride-hailing drivers
It seems reasonable enough. If driverless happens, Uber would be rushing to acquire a very large number of cars. Regardless of autonomousness, there are paths that Uber could take which would make them a huge consumer and operator of cars.
It's a hard to predict future. There is at least some possibility that driverless will eat a market fast, and the market changes for car makers. They need a hedge.
Overall though... how much money is going in to autonomous r&d? It's also possible that driverless doesn't get to the making money part for a long time. Is the funding sustainable?
Toyota likes their best selling car in America moniker. This is why they sell so many cars to rental car agencies. Honda and others don't aggressively pursue this and have lower "sales" as a result.
So I don't find this "investment" very surprising at all.
Right now there is too much money chasing too few startups. Some startups literally operate with a "we have a bottomless pit of money and it doesn't matter if we never show a dollar of profit, we just need to keep on raising more from investors who think they'll miss the train".
Uber and Cruise kind of feel like that. I'm sure Waymo has a bottomless pit from Google too. I guess that's the reality of moonshot projects.
A lot of them are going to burn billions of dollars throwing lavish office parties and off sites, but one of them will be huge. That's essentially what SoftBank's 100B vision fund is kind of doing.
A lot of investors operate with a "fear of missing out" syndrome and that's probably why Uber gets more funding than 99% of all US startups combined.
They're not doing a full pivot - rather they're shifting from "car-ride-hailing," to the all-encompassing "mobility" and diversifying their revenue across a number of transportation means.
My read: Once Waymo has released re-usable tech, Toyota will install that on its cars, and Uber will facilitate the dispatch - And that Uber won't do this with another car vendor for X years (in the deal)
From what I have heard Toyota’s approach is to prefer taking over the driver in difficult situations instead of the other way around. Uber could benefit a lot from this.
