It's pretty much the other way around (1% done, 99% to go) except for contrived locations that have perfect climates. It's like wifi devices claiming they can work over miles but that's only true if you have line of sight.
A long term snow covering of lane markings typical for the winter months of much of north america and the emergent driving lanes people flock to cannot be handled by any existing driving AI. The places people drive are wrong according to absolute road positions and the relative markers are obscured to both human and machine. Unless car AIs can do the wrong thing like all the humans will in those situations it won't work. And that's a hard problem.
You're saying that for humans it requires perhaps 30% additional learning effort but for computers 9900%, and I'm supposed to accept the 30/9900 discrepancy because… what?
When I learned to drive a car, I lived with an uphill driveway in a town whose winter climate varies by the hour. It's quite normal there to get temperatures around zero, some rain, then the temperature drops and 20cm snow fall. So on a January morning I often had to start the car uphill on steel ice covered with water and snow. With a sharp curve as a bonus.
That kind of start isn't the simplst, and you could persuade me that that kind of thing is as difficult as navigating an intersection with ~10 other cars, cyclists and pedestrians. But not that it's tens of times as difficult.
Humans generalize what they learn extremely well. Humans also spent their whole life learning how to navigate their environment, and this is transferable to a range of activities.
Yes, that's what I think too. I just don't see a reason why driving skills should not be largely transferable.
Being able to drive in California doesn't include all the skills needed for e.g. winter driving in Norway, or for driving in an Indian city, or (list goes on), but I fail to see that it's just 1% (as an upstream poster claimed) or even close to 1%. I fail to see that driving skills are 99% intransferable.
I also fail to see that while the skills are transferable for humans, they are intransferable for computers. Various people in the thread assert so, but argument by repeated assertion is meh.
> I also fail to see that while the skills are transferable for humans, they are intransferable for computers. Various people in the thread assert so, but argument by repeated assertion is meh.
That they are nontransferable is the null hypothesis. I mean, we need evidence for this transfer. It's up for self-driving companies to produce this evidence. But they didn't, unfortunately.
That they are nontransferable and 99 times as diffcult is not any kind of null hypothesis. That's just a nasty rhetorical trick: Take something reasonable, piggyback something ridiculous onto it, and argue that the reasonable thing is reasonable.
> You're saying that for humans it requires perhaps 30% additional learning effort but for computers 9900%, and I'm supposed to accept the 30/9900 discrepancy because… what?
Because AI is not even remotely close to “solved”, and driving is a combination of a million edge conditions, social knowledge and communication, common sense knowledge, vision understanding, planning, and force feedback adjustment.
What’s unclear to me is why you’d expect these things to be so easy. To even say “for computers” doesn’t seem to recognize this isn’t just about executing basic math, but is rather trying to represent intelligence. There is no single computer here, these are implementations of ML models, which are each unique with their own architecture, parameter space, and capabilities.
> You're saying that for humans it requires perhaps 30% additional learning effort but for computers 9900%, and I'm supposed to accept the 30/9900 discrepancy because… what?
Have you noticed how a small child will learn to identify things like cats and birds after only seeing a few examples? But computers need orders of magnitude more data and instruction.
Anyone who doesn’t understand how modern day deep learning works really ought to do an intro course.
Throwing random percentages is silly. In reality, these vehicles have only been tested within a very tight scope, in near-controlled conditions in Arizona, California, etc. with very specific conditions for their operations.
I would consider it 99% complete when I can pull up a map, point to any stretch of drivable road in the United States and ask it to autonomously operate there.
Specifically, there's an especially tricky road in I-35N in Texas where the lane markings come ago, yellow shoulder markings _merge_ into barriers, and the road condition is so bad that the steering rack will _turn_ by itself(!). We're not even close to calling this thing autonomous. If and when it gets to that point, I would consider it "99%" complete.
In reality though, I really see this tech being viable in continued controlled conditions. Maybe there will be a point when certain lanes will become dedicated "autonomous lanes", where they're completely isolated from the rest of traffic, come with special markings and sensor suites to assist driveler-less cars operate efficiently.
As long as the chaotic human element is ever present, regardless of how well they model these systems, it will fail to cope with the massive amount of discontinuous information humans inject into a situation (swerving into a lane, road rage, cargo suddenly coming loose from a truck, etc.)
Oh, and have we tested these things during winter? :p
> Specifically, there's an especially tricky road in I-35N in Texas
Seems like you could just get some data on that before declaring that "we're not even close". How do shipping Teslas do?
FWIW: my impression is that the things people think are hard aren't the problems that are actually hard. People here and elsewhere on the internet have spent years screaming about lidar vs. radar vs. vision. And... it turns out vision basically won. Something like half the Tesla fleet in ths US are vision only in the US now, and... the cars see stuff just fine. We're not having detection failures. It works.
If you watch all the FSD beta video being spread around youtube, much of it by drivers specifically looking for edge cases like this, most of the remaining problems are in pathing and decisionmaking, not detection. The cars see a busy intersection and then get paralyzed (there was an amusing clip from one guy of a car trying to make a left turn, giving up, and repathing a right turn that took it back in a loop only to fail at the same left turn repeatedly), or creep too slowly and annoy other drivers. Or they choose the wrong lane, or misread a multiple traffic light environment, or misread the need for detection (the one really dangerous video I remember had the car deciding to take a left turn across a broad street with a complicated blind corner instead of creeping out farther).
Those are bugs to fix, for sure. But they're not the bugs we thought we had to fix. Those bugs... got fixed.
So you can look at this as a pessimist and say "we'll never get there", or you can judge from previous experience that the hard problems won't turn out to be that hard in the end.
Personally, having been watching the progress in this space, my guess is that Tesla and Waymo (and maybe Mobileye -- no one else is close) get this basically fixed within a year or two.
At which point the debate will shift from "we'll never get there" to "we'll never be able to regulate this appropriately" or somesuch.
> Something like half the Tesla fleet in ths US are vision only in the US now, and... the cars see stuff just fine. We're not having detection failures. It works.
You and I have extremely different experiences with Teslas. About the only place I could conceive arguing it works is highways and even then I wouldn't trust riding one down the highway without a steering wheel. Hell, my coworker can't even call his out of the driveway because it has been thinking his mailbox is a car for the last 6 months.
I think we'll get there and I don't think it'll matter if it's with LIDAR or pure vision but we're certainly not close to being able to remove the steering wheel (regulation aside).
Mine drove me to Yellowstone and back with like two disengagements which were my paranoia about other traffic.
But yes: the example was an interstate in Texas. And I happen to know it actually works quite well on interstates. So a very reasonable experiment about whether or not automation "will never get there" on edge cases like the referenced interstate is to take a commonly deployed, already available automation technology there and test it.
But also recognize that shipping autopilot is now about a year old. All the FSD work hasn't made it to released cars yet. It's very much evolved, but visible only on youtube right now.
A lie. Or at least a giant [citation needed]. What evidence exists seems to show (not really surprisingly) that the cars are safer with AP engaged than when not.
> What evidence exists seems to show (not really surprisingly) that the cars are safer with AP engaged than when not.
Good way to change goal posts. There are many, many instances of Teslas hitting large stationary objects on roads, and who knows more instances of near misses due to driver intervention.
When the AP can be engaged, which is usually 'better' conditions. Unfortunately humans don't always have the option of driving in 'better' conditions, so this is an entirely cherry picked subset of data.
Cmon. There's a video of a Tesla FSD driving on the wrong side of the road, cutting off a car it did not see at an intersection, then accelerating straight into a fence, all in the space of five minutes in Oakland. Tesla has by no means solved the vision problem. Their software is laughably bad.
Teslas are not self driving cars regardless of what Musk claims. They aren't even close. Even Tesla admits that their "full self driving" mode is only level 2 automation.
What do you call the video of a Tesla mistaking the moon for a yellow traffic light[1], if not a detection failure? That wouldn't happen with a lidar-only or lidar-assisted system.
I'm not sure how lidar would help in detecting yellow lights correctly. Seems to me like that has to be a vision system, at least until we get all traffic lights to report their state via radio.
I was thinking distance—if the object emitting the yellow light is 240K miles away from your car, it can be assumed that it isn’t a traffic signal, right?
This is a strange bug. As if they weren't making predictions about where the yellow light would have to be one second later, by using at least speed and accelerometer data, and compare the expected result with the new reality.
This is not necessarily a job for lidar, they are just omitting some checks.
At second 18 [1] a street light comes into play which looks similar to the moon. That relative movement is what the system should have been expecting from the moon, but they apparently just re-computed the assumption that the moon is a traffic light which is moving along with the car.
It's somewhat revealing that they are not doing these kind of checks. I actually can't believe that they aren't doing it, but I can't see a case where the traffic light would always be at the same spot, if you have velocity and acceleration data at hand.
The traffic light detection runs on an older NN which works over individual image frames instead of the new video-based system. They are migrating towards the latter. In any case, they can ask the fleet for examples of this and train the NN against it.
Yes, it’s kind of a silly headline. A percentage compares two numbers, in the numerator and a denominator. If you don’t know what’s being counted or measured, it’s not a real statistic.
However, people do uses percentages metaphorically and we should try not to be so distracted by silly headlines.
That's exactly what this is. It's the classic "The first 90% is 90% of the work. The last 10% is the other 90%." Or whatever the exact quote is. It's not about the number. It's the fact that it's a lot of work to get close. It's as much or more work to get over the finish line.
I look at it this way: I haven't had a car accident in over 25 years of driving. Now if there was a 1% chance of having a car accident every time drove, I'd be having multiple accidents a year. This seems bad.
except 99% doesn't necessarily mean "every time you drive"; it's obvious to me that if you drive for 100 miles you are more likely to get into an accident than 10 miles. It's probably closer to "in 99% of situations".
The reality is that long term trips may, in general, be safer since you're on highways. Some statistics show that short term trips are more likely to have accidents.
Humans are particularly bad at highway driving long distances because they get highway hypnosis or fall alseep. On local roads on the other hand humans tend to do a good job reasoning with situations.
Because manoeuvers are hard. And whatever the task a human takes, they need up to a few minutes to get "in the zone" and fully focused on it. This includes driving.
Highways without intersections are very safe. Only on/off ramps. Mixing traffic directions and intersections are risk factors which mean that non-highways are more risky per kilometer.
A steelman interpretation would for example be "works on 99% of north American roads" in which case preventing it from driving anywhere near known bad patches would get it complete enough.
It's clearly based on the HYPER-idealistic driving conditions of primarily high-angle sun with dreary-yet-predictable Phoenix "stroads".
They picked a good first experiments location but it's NOT representative of more than 1% of roads on the planet! It's a bizarro-world of simplified challenge and problem space. Drop such a system into any real world driving and it will fall flat on its face very quickly and endanger both passengers, other drivers and pedestrians.
That stretch of road you describe as difficult is actually not difficult at all for Waymo's current technology. Their navigation does not depend on vision.
Context. The faint wear marks that tires make on the road over years, the seam between lanes, the car in front of them, the distance away from the guardrail or ditch, or there are no marked lanes at all and you just have to share the road with oncoming traffic. Furthermore, it's not uncommon for three lane roads to turn into two lanes during heavy snow, and in those case, the lanes are wherever everyone decided they were, collectively.
The case I was thinking of is a highway I was in last week, where temporary work has one of the directions going through the other direction, so the 2+emergency lanes in one direction become 2 lanes without emergency in one direction and one lane in the opposite one.
In this case, there are temporary lane markings on the ground indicating you should not follow the old ones, but rather align yourself further to the right then you would normally do.
OTOH, I am familiar with the situation in Rome where often the lane markings are deleted and a road may be 2-3-4 lanes depending on how people queue up, in which case people would generally rely on social/experience cues, which I'm unable to formalize.
I'd be really curious to see how a map/gps based system handles situations where a divided highway is under construction and has temporary lines painted to direct the right side of the road to drive on the left side while left side traffic is held up by a man in a vest holding a sign. This isn't a particularly uncommon scenario, and if the road crews are hard at work, the zone where this is occurring can move by miles a day. To navigate this, a driver needs to rely on road markings, temporary signs, and humans directing traffic. Maps won't cut the mustard in a situation like this.
If a driver assistance system can't handle this sort of situation, that's fine. I'm sure it's still useful most of the time. But it's clearly not "99% of the way to FSD."
I assume that, once you get vehicles driving autonomously on sections of highways, you have some sort of beacon/alerting system that cause the vehicle to tell the driver that it's going to be dropping out of autonomous mode.
The "beacon/alerting system" are the signs, and the man in a vest. If the car can't respond to those, then the words "full self driving" should never be used to describe it.
On a highway, you often don't have a man in a vest and the signs I see are often vague alerts about construction and lane shifts ahead that may or may not be actually happening at the moment.
But I don't really disagree. Some mechanism (perhaps a redundant mechanism) when approaching a possible construction area to give a human a minute or so to take over.
I'd see self driving as not needing a person at the wheel. I don't care if there are changes to the road, it's being driven without a person nat the wheel
Autonomous driving systems do not need to be door-to-door. They also can't require taking over control on a moment's notice. But, systems that can fully autonomous for long stretches of highway driving seem useful.
They just don't address use cases where people aren't able to drive at all. Which are interesting to a lot of people but they may just not happen for a long time.
There are people who just want cheaper Ubers but there are others who are fine with handing off long boring sections of highway driving.
Waymo's system doesn't depend on GPS for navigation either. They use their "realtime sensor suite" (lidar) to localize the vehicle position within the map.
As part of that localization, there will be a metric of how good the local solution is vs. the map. If a divider moves, for example, the data will be inconsistent with the map, and the software will know that something is amiss, without having to intelligently classify anything. The software can drive confidently when in a consistent environment, and drive conservatively when in an inconsistent environment.
Generally, any object that isn't in the map will stand out obviously. That's where having stuff like cameras becomes useful. Asking a computer, "is there a construction worker in this 20MP image of a street?" is hard, but asking a computer, "is this 0.1MP human sized blob a construction worker?" is a lot easier. Waymo has been responding intelligently to bicyclist hand signals for years. "Is this sign shaped blob a construction sign?" is even easier.
The lidar and mapping is a key difference between Waymo and Tesla. Waymo works because they have the big data infrastructure, and the budget to deck out every car with orders of magnitude more sensors and compute.
> Specifically, there's an especially tricky road in I-35N in Texas
Why are we ok with streets in such slapdash condition in the first place?
Yes I think self driving cars need to handle it, but I also think we aren't handling it.
Self driving cars have the potential to greatly reduce deaths due to accidents and improve accessibility for often ignored segments of our population. It is sad to see it so close, but held back by reasons like this.
There was always where this was going to end -- demands that we spend (from the public) hundreds of billions to upgrade roads to make self-driving tech "work".
Also,
> and right now, no driverless car from any company can gracefully handle rain, sleet, or snow
Good thing the entire US has the weather patterns of the bay area, and never experiences rain, sleet, or snow!
Lots of people aren't okay with road conditions. But lots more people aren't okay with a myraid of other pet peeves and legitimate grievances they have with just about every single aspect of our society. The process of prioritizing which matters to address is called "politics", a process so notorious I assume I don't need to explain this further.
IH-35 has been under continuous construction since .. the 90's?
Hindsight being what it is .. they should have built a new 8-lane highway a few miles to the West rather than attempting to widen it multiple times (with all the bridge replacements needed) through towns like Temple and Waco. But as the Texas Central Railroad has found, you can't just eminent-domain land in Texas for large infrastructure projects.
I live about an hour west of Boston. There's a sort of half-ring six-lane highway called I-495. It mostly handles the traffic load fairly well. But it's a source of sometimes genuine surprise to me that a highway of this capacity for the area was built in the late 1950s given that must have seemed almost outrageously overbuilt for at least most of its route at the time.
I almost feel that relatively short range (meaning they can make the hops between charging stations) self-flying passenger drones/planes would be better than cars. Leet air-traffic control systems with decades of anti collision experience interact with and track all flights and inform all flying systems of the telemtry data for all slights so they have spatial awareness that way. Set flight paths that are low, but avoid densly populated areas/neighborhoods/infrastructure etc.
Create small emergency landing areas all over...
The only ground based vehicles I care about being 99% auto would be long-haul trucking. With a station to pickup a local last-mile-human-driver for actual drop-offf etc.
Lex interviewed
Dmitri Dolgov CTO Waymo
https://www.youtube.com/watch?v=P6prRXkI5HM
I think you are conflating operating a public service within a very tight scope and "only been tested within a very tight scope"
I have elderly relatives who won't drive at night, or on highways. I have friends who won't drive in various degrees of inclement weather - and I know a LOT of Californians who can't drive in the snow.
Human skill is extremely variable. A car that can drive to any road under any condition will be vastly superior to the average driver.
Reducing such a complex system to a single percentage may be silly, but the idea that a car might have partial coverage under certain conditions is quite reasonable and is reflective of how real human drivers work as well.
I think the GP post also points out that some roads are, frankly, terrible. I've had more than a few times where I wasn't clear if the cones were set in such a way that I could continue down the road or not and I've been wrong at least once (though the cop was very understanding after I explained it).
Some randomly placed cones are enough to confuse humans in many circumstances, as well. I doubt we'll ever have an all-encompassing solution for that, either. Even for things like autonomous-only lane, well it's not like everyone obeys the HOV lanes currently...
There probably isn't a perfect solution, especially when I've seen plenty of ambiguous markings.
Just yesterday I ran into some confusing cones and signage where I made the right merge onto the exit, but the car behind me made the wrong call and drove a kilometre or so down a newly built, unopened highway in parallel to me on my left until they got to some concrete barriers and realized their error.
And there's zero possible way self-driving cars are going to handle country roads that are sometime dirt in the snow during a blizzard.
Humans absolutely can do that. But the very architectural basis via ML that's being used never will. Different architecture? Maybe. Maybe in 50-100 years.
The average quality isn't the only thing. What about the correlation of errors?
There are situations humans can't handle, and notorious "hot spots" where crashes repeatedly happen, but if AI is more consistent, it might lead to more concentrated failures.
Have you heard about the problems caused by navigation routing, where large numbers of people are directed through roads that aren't suited for heavy traffic?
That awful stretch of I95 can be special cased. It's a great example of what a computer can be better at than humans who each have to individually figure it out.
An unprotected left turn is probably a harder problem. They'll be expected to do it aggressively like humans, and also expected to do it safely. Those are incompatible directives.
The (IMO) "right" way to do an unprotected left is also somewhat situational. If there isn't a lot of traffic generally on a country road but a somewhat bunched up group of vehicles is coming, I'll be patient until the bunch has cleared (unless someone flashes their lights at me). If on the other hand, I'm in a busier urban area and I can see there aren't a lot of openings, I'll be more aggressive including at traffic light changes.
In general, my expectation--which may be completely wrong--is that we'll see autonomous driving on sections of highways in good weather long before we have door to door almost everywhere. Which is useful by itself for long highway drives.
I think responding to the human element is huge for safe driving. You would not want to safely drive the same way in LA as Boston, or Seattle. Other drivers make very different assumptions on your response to their (and others) actions, and that can create danger irrespective of navigation.
Google Maps navigation can't even reliably call out which lane to use for turning in major metro areas, and that's kinda the basic best case scenario: clear images/footage, unlimited compute time, static road markers, common use case.
I'm not sure how they can do better with far more stringent resources and requirements.
This has always had me wondering about the viability of self driving: what's it going to be like in different countries?
Not only are there different signs, different line marking & colours, & different widths. Cars are different: not all models are sold equally throughout the world. People are different: through their clothing, height and build. But the really big difference is that not everyone drives on the RHS...
The article lists left turns as a problem, whereas on a LHS road system, right hand turns will be the problem.
In NZ (a LHS system), I've heard that turning right across traffic has priority over oncoming traffic.
In Melbourne, VIC Australia (also LHS), right turns are made from the left most lane because of trams lines.
I'm sure there are loads of other issues specific to other geographies.
To me, this never ending self driving experiment screams out that the wrong solution is being pursued. (That doesn't imply that I know what they should be using as a solution)
> In NZ (a LHS system), I've heard that turning right across traffic has priority over oncoming traffic.
This is incorrect and would be utterly chaotic. You may be thinking of an outdated rule in NZ where right-turning traffic takes precedence over left-turning coming in the opposite direction to the same side road. This was the case for a long time but was reversed[0] in 2012, overnight, almost without a hitch.
My current work involves ground-truthing sensor and classifier hits from various automotive OEMs. Take, for example, reading old-fashioned speed limit signs. In principle, doing this correctly and reliably in context is a relatively simple machine learning problem, particularly when restricted to a single country.
I have yet to work with an automotive OEM that doesn't consistently mis-classify some speed limit signs in some common cases that are completely unambiguous. It leads one to wonder about the quality of classifier design and testing such that these end up in production vehicles. The cases that cause a classifier to fail are different across automotive OEMs too, it isn't intrinsic; other automotive OEMs will handle that specific case just fine.
On the other hand, when it works it often works surprisingly well. The good/bad news is that most of the problems I see are ultimately caused by poor model design and testing. These are all surmountable with better quality AI engineering.
There is more AI to self-driving cars than the classifiers that pull features out of their environmental sensors. However, the driving AIs are making decisions based on the output of those classifiers -- garbage in, garbage out.
These are the kinds of situations that will produce relentless incremental improvements over time, and there is still a lot of low-hanging fruit to improve on.
Right now, our roads are only designed to be human-readable.
But what if machines become the dominant drivers? We need to make the roads machine readable: Road signs redesigned 'QR Code' style, maybe even some kind of wireless broadcast system to communicate traffic light changes.
From anywhere to anywhere is also an unnecessarily lofty goal. Bringing me to the front of a store in a strip mall is nice, but dropping me off at the street is good enough. Even doing enough streets in a city to pickup/dropoff within 1/4 mile would be revolutionary.
We can... but it's not needed at the rate things are going. AI will be able to understand almost everything humans can and it's way cheaper to not have to rebuild things for machines.
I work in the area. I would say the static stuff that’s hard is stuff that’s also hard for human drivers, like poor lane markers or ambiguous signage. So I’m not sure that QR codes will help—-just fix the signage.
That makes a lot of sense, on a highway with three lanes, I'm always making sure to not merge to the middle lane, if there is a car on the opposite lane, to name another situation.
The hard question which really has nothing to do with the tech, is what we do if a self driving car kills a person. Made worse if the accident could be avoided by the majority of human drivers.
>The hard question which really has nothing to do with the tech, is what we do if a self driving car kills a person. Made worse if the accident could be avoided by the majority of human drivers.
we have already had exactly such a case - Uber. Even though Uber was totally negligent, nothing happened to them, beside probably some settlement behind the scene. In part they were able to wiggle it by showing a bad dynamic range video (which is completely different from how human eyes see in such situation) there the victim appears as if out of nowhere.
I think that case just sets the pattern that the autonomous vehicles will not be judged according to human driver standard.
So what happens with the human driver (you) if the same accident happens ?
If the Uber case sets the standard, then negligent homicide is a likely charge.
The sentence for that in the US is "A minimum of 4 years in prison and up to a maximum of 8 years in prison" according to this link:
That’s really expensive on top of the inherent inefficiency of cars: you’re paying a ton of money for something which uses a lot of energy and pollution (yes, even BEVs) to carry slightly over one person on average. Rebuilding the road system won’t change that or make climate change go away, especially since you’d need a lengthy transition period.
What might make sense is limited deployment in areas where the problem can be constrained: dedicated bus routes, truck convoy lanes on an interstate, etc.
For the rest of it we should be focusing on how to get people out of cars since even BEVs pollute far more than buses, rail, bicycles, or walking.
I've always wondered whether it would be possible to have some kind of compartmentalized bus.
Some kind of compromise between asking people to cram themselves like sardines against a bunch of strangers, while still allowing people to sit next to their family members and friends.
It's convenient to say a desire to not jam up close to strangers implicates its holder in some terrible character flaw of not caring for ones fellow human. But if you really want to get more people on buses it's a desire we'll have to accommodate and contend with.
You're describing a system that is operating beyond capacity. There's no special fix required other than increasing capacity, by adding more buses to existing routes or new routes. There may be political issues with allocating funding or getting authorities to acknowledge/address lack of capacity. But it's not something that requires redesigning the bus.
Energy consumption is a factor of distance. "Get people out of cars" in inherently implying removing the fundamental right to freedom of movement and living a better life. Cars have done more to equalize rural populations and give them a quality of life closer to city dwellers than any other modern invention maybe short of publicly funded water and power infrastructure.
I'd highly encourage you to move to the middle of the country and live a 30 minute drive outside of town (because that is all you can afford). You'll quickly realize that super bad evil cars are the means by which a good percentage of the population has access to fresh food, medical care, and other basic needs.
Energy consumption is a factor of both distance and efficiency: this is why trains are so much more efficient than semi-trucks which are more efficient than personal vehicles even if they’re traveling between the same points: steel on steel rails have less friction and the first two have better engine to cargo ratios. My comment was specifically focused on the latter since a huge fraction of vehicle pollution comes from affluent people driving in urban areas, not farmers.
Also note that I’m not disagreeing that mobility is important but that we literally cannot afford to continue polluting the way we have. You tried to make this emotional with the “super bad evil cars” phrasing but it’s a simple engineering question: right now, people have built lifestyles based on low subsidized fossil fuel prices and being allowed to ignore externalities. I’m aware of what that lifestyle is like – and how often it’s not “all you can afford” but “where you can afford to buy the house as big you think you deserve”, too - the latter being far less sympathetic when asking everyone else to subsidize it.
When energy is cheap and you can ignore pollution, you can drive an overpowered vehicle on frequent trips with minimal use of the total cargo capacity. If the cost goes up, those calculations all change: people pay attention to fuel efficiency when buying vehicles and combine / reduce trips, invest in household efficiency, etc. Cost-constrained American rural dwellers and those I’ve met in other countries with higher fuel costs don’t drive a vehicle designed to haul livestock to pick up groceries because it’s overkill.
That extends to things like zoning: the majority of people living in exurbs for financial reasons are doing so because closer in development was low density, often required by code, and significant amounts of land were required to be used for car storage.
Part of climate change mitigation will be reversing those problems, and that kind of thing seems like a more fruitful area for us to be spending time than trying to make high-pollution commuting more appealing.
Machine readable roads already exist. They are called train tracks and are already used by autonomous trains. They also allow for far more energy and space efficient transport with better throughput than cars.
Reading signs is the really, really easy part. If your vision system can't read and recognise the signs and signals reliably, it's not even beginning to solve the other problems. Trying to adapt the roads to make this part easier is a terrible return on investment. 99% of what you could do to make roads easier to use for self-driving cars is basic maintenance: making sure the signs, markings, and signals remain clearly visible, which also helps improve the performance of human drivers.
I remember watching a documentary on Discovery in the early 90s where they showed a motorcade of 10 Audis closely following each other using radar and some magnetic markers embedded in the road surface. They could maneuver within inches of one another, stop on a dime etc. I wonder how much embedding such passive markers into the new roads would have cost compared to the amount of money pumped into autonomous driving companies so far.
Maybe? You'd likely need to be full time to get equity, though. I'm personally not familiar with spring boot, but it sounds like a framework for implementing REST services? I'm mainly an embedded developer.
Mind expanding? Keeping my eye out for new things to work on: if there’s scaled-down autonomous navigation problems that might be both immediately useful and more generally useful when solved, that sounds interesting.
I'm an embedded software engineer at Zipline. We're delivering high value medical supplies via UAV as a service. We've been operating in Rwanda and Ghana for several years, and we're expanding.
Yeah, use air and all nasty problems go away. I am shocked that tech isn't mainstream already given multitude of benefits compared to terrestrial transports.
Well, if I code something that just move the car forward if nothing is in front, I am probably 80% there. Not sure if I trust their 99%. It seems an excuse for academics to not release an actual technology that works.
Waymo has raised over $5.5 billion and has thousands of employees without so much as a product.
If this were any other startup it would be considered insane and never gotten to this point. Don't know why people keep dumping money into this project thinking they're just around the corner.
Even Tesla sold some cars when it raised a fraction of that amount, and $70 million came from Musk
> By January 2009, Tesla had raised US$187 million and delivered 147 cars. [0]
Why do you need to jump to "Full Self Driving"? Presumably there are intermediaries.
They're whole plan lacks a feedback mechanism.
You don't even know the return if they succeed. Drug companies have a long history so they can anticipate the cost of development, odds of success and payoff if successful. They do all these things prior to any investments being made.
The closest Waymo has is # of truck drivers * salary = profit.
> Why do you need to jump to "Full Self Driving"? Presumably there are intermediaries.
They've explained this very often. Unlike Tesla, they don't think you can go through the intermediaries without the dangerous situation of drivers not paying attention because it usually works. Based on their experiences a decade ago of employees using it as drivers assist, people check out and it becomes more and more dangerous even if the software is improving.
So their plan is to go into cities one by one, map out the terrain to the centimeter and have a few thousand engineers support each region in which they operate?
I don't even know what their plan is. Do they want to sell cars, operate trucks, license out the software, have robo-taxis?
Not sure if you are honestly asking or just expressing skepticism but right now it's clearly robotaxis then trucks. They've talked about the rest but much less.
As far as expansion plans there's no reason to think each place they go will require thousands of engineers.
If the last 1% is as hard as the previous 99% then they are not at 99% but at 50% ;=)
It's just that a part much larger then 1% of the way is perceived to be just 1%. I.e. there is a insane lot of hidden complexity generally/often overlooked.
Also, often we'll change such titles (as I just did above), and then the shallow-title-objections become a sort of uncollected garbage, referencing something that no longer exists.
I'm going to make a stub reply and bundle those comments underneath it so as to collapse them. There are more interesting things to discuss here.
Stub for shallow title dismissals. I'm omitting a few that led to interesting subthreads and/or contained an interesting observation. Admittedly a judgment call.
This article is hilarious. It's all about the march of 9's. Saying they're 99% done means they've only just begun.
Also it doesn't account for their model which requires paying engineers and installing tens of thousands of dollars in every vehicle to make it work. And the fact that they need pre established HD maps to do anything.
Am I allowed to opt-in to the timeout box if I have a knee-jerk reaction to the new title? (In which I Clone the current title so that my reference to it doesn't become invalid when someone overwrites it on a different thread: "Waymo has lost its CEO and is still getting stymied by traffic cones".)
At first glance I thought the title was saying that a Waymo car carrying the CEO has been lost for days because it got trapped by traffic cones.
I don't think "At first glance I thought the title" comments are very meaningful. You can always wait for the second glance before posting, which takes care of 100% of these.
Thanks for posting it in the right place though :)
It's the article's own subtitle; I wouldn't say that's "nowhere near". In fact the subtitle is often much closer to the article's actual content, especially when the main titles are written by headline writers optimizing for linkbait.
Able to pick me up at my apartment in Boston and take me to a location in Cambridge. I would not say the current "self driving cars" in Phoenix are anything close to ready (even for Phoenix!)
I think if I knew you I'd happily bet this, but I don't make it a habit of randomly betting people online. I think if the bet is a robotaxis ride across Boston area by 2031, should be no problem. Frankly a Tesla could probably do it now.
There is a website https://longbets.org/ Long Bets that's let people do this. Essentially it's to publicly state to the world you are betting, and the winner donates to charity. I would do this here with my real name.
My requirements:
- The time of day or night of my choosing
- Anytime type of weather - except:
- Will not be during snow or with excessive ice coverage (although this should be fine, IMO)
- Must not have a driver physically there who takes over, nor be driven primarily by remote drivers (remote support could be an option, but it's not a "remote control car" robo taxi)
If you accept, I'll arrange it and wager the $1000
> In reality, skilled disassembly is required. Engineers must take apart the cars and put them back together by hand. One misplaced wire can leave engineers puzzling for days over where the problem is, according to a person familiar with the operations who describes the system as cumbersome and prone to quality problems.
I think this is one of the more interesting pieces in the article for me. We all kinda know by now that there are issues around AI handling all driving situations. I think it's interesting to hear Waymo having difficulty with the manufacturing aspect.
It does make a certain amount of sense. Waymo is a company that is mainly trying to solve the AI problem around self-driving cars, but other things can stymie those efforts like assembly. This isn't meant as a dig at Waymo, just more an observation that just because a company is excellent in one area doesn't mean that they will be excellent in all areas and that even excellent companies like Alphabet/Google/Waymo can stumble in new areas.
I think it explains why the established automakers don't take any of the aspiring automakers from tech seriously. They know how hard it is to supply, assemble, and mass produce a car, and they know how little investment to desire there is from big tech to be better at that aspect.
It doesn't matter how good your product is if you can't actually ship it.
You basically are describing big auto's attitude towards Tesla 10 years ago. Look at where big auto is now.
Terrified, and behind by 10 years. And by terrified I mean their CEOs. Do you think the BMW CEO would still have his job if he didn't squander their EV efforts that started about 15 years ago?
The BMW CEO getting bounced was the warning shot that made every automaker executive sit up and realize: my company either has an EV platform program underway, or the investors will bounce me.
If they are still not taking Tesla (to a lesser degree Rivian and Lucid) seriously, they're in big denial at this point.
GM and VW have basically woken up. Ford and Mercedes are toe-dipping. BMW seems paralyzed by internal management dissent,
Look where Tesla is now. At every step that have faced production consistency and reliability issues. They flat out suck at making cars and have 20 years of quality engineering learnings to catch up on.
Consumer Reports 2021 report puts Tesla in 16th place among US car brands. In JD Powers 2021 Dependability Study Tesla came in 30th, followed only by Jaguar, Alfa Romero, and Land Rover.
Underrated truth - have owned a Tesla for 5 years, and the number of stupid things that go wrong with it is incredibly high. It works like a phone...after a number of years, it takes forever to boot and has plethora of bugs. Random stuff breaks/falls off. We're going to see even more of this in 10 years, as the cars that were produced during the scaling period start falling apart. What sets automakers apart is that you can buy a used car after 10 years and expect it to still work.
i think one of the main principal errors on Waymo (and other large players) part is not working with military. Given that for military the task is simpler while safety requirements is significantly laxer, i'd have expected that we'll see self-driving tanks/transports/etc. well before cars.
One of the reasons tanks exist is to protect the operators. If you’re not going to have boots on the ground in the first place (autonomous tank), you can rethink the vehicle completely.
there are still no autonomous drones either. I posted long time ago that before autonomous ground vehicles we should see autonomous airborne vehicles because of that task being much simpler. And of course the military would be the starting point here too.
>tanks exist is to protect the operators
operators of a large gun quite close to the target. If one can do without such a gun in such a role on the battlefield, then the need for tanks become less. We do see that in the recent wars - like Azerbajan/Armenia - the high-precision stand-off weapons fired from drones did that job perfectly. There were no countermeasures though, like say electronic communication and positioning blocking which would have necessitated autonomous capabilities for the drones.
I used to believe in the self driving hype. Then I learned ML and started working in related areas. Now I know full self driving is not happening, not with the technology/algorithms we have. So what ever anyone does is just an approximation and we will never be able to trust the car will handle any situation thrown at it. So its always going to be a souped up cruise control, the likes of which Tesla and others are selling. We need some kind of breakthrough to get fully autonomous self driving where we can sleep in the car while it takes us to the destination.
I would, of course, love to hear if folks here think I am wrong.
Anyone committed to self driving cars needs to be a road building company and own the road network that will manage these mostly-automated efficiency lanes, with car provision (more likely tech/compatability licensing) as a secondary income stream.
Waymo is currently successfully handling extreme edge cases on city streets via a call center. Which is a viable option as their cars average 30,000 miles completely autonomously. Worst case come to a complete stop and then ask a human to take over is perfectly reasonable option.
Google running this as a pure Taxi substitute is seemingly a completely viable option today. They don’t need to be 100% autonomous as long as one person can handle a dozen or more cars that’s easily good enough to out compete traditional taxis or ride sharing services. Granted this is limited to a 50 square mile area, but collecting more data isn’t the difficult part.
Including all the expense of r&d, call center training, liability insurance etc.
Is this a viable business in a 50 square mile area? Is it scalable nationwide? Is it price competitive with traditional taxi and ride share?
The overnight use case is compelling as it opens entirely new transit options. Cars could be made competitive with air travel for the overnight use case. And traffic wouldn’t be nearly as large of a concern for most if they weren’t driving.
> Is it scaleable nationwide? Is it price competitive with traditional taxi and ride share?
Yes to both, critically you can use a hybrid approach for outlying areas and cities lacking regulatory approval. By comparison DC is only 61 square miles and San Francisco is 47 square miles so this is plenty of area for local service and a reasonable benchmark for expansion.
What’s not mentioned is the cars do need someone to clean them and fill up the tank etc, but that’s hardly an additional expense.
How is what Waymo offering anything like a souped up cruise control?
Sure it's EXTREMELY limited but that doesn't seem like an accurate description since they trust it enough to be liable for it to drive by itself or stop safely when it can't. Given the number of their cars going around SF lately I think they'll expand some soon too.
> How is what Waymo offering anything like a souped up cruise control?
It is actually less than that, since there is no Waymo car you can go out and buy. Driverless Uber rides are absolutely useless. When I order an Uber, how does it matter to me if it is driven by a human or a computer?
Modern cars depend on modern road infrastructure. Like traffic lights, lane markers, traffic engineering, etc, enabled the efficiency and convenience we enjoy today.
And this infrastructure was built for human driver. And they do not work well for software driver.
>And this infrastructure was built for human driver. And they do not work well for software driver.
How much would it cost though for parallel infrastructure that is suitable for a software driver? (better signage, rfid tags in the roads, 'smart' traffic cones, etc.) For say the Bay Area. 10 billion to a few hundred billion? Speaking for myself, I'd be happy to pay say an additional 1% sales tax if it meant facilitating an area wide 24/7 autonomous 'public transit' taxi service, presumably with at-cost per mile fares.
I don't know. There has not been any serious effort in theoretizing the changes needed for software driver. The so-called vechle and road codesign has been mostly gimmick as far as I know.
I've considered this before, but never ironed out details. What would it entail? Magnets/IR/Beacons of some type marking roads and lanes? And each car would need them too I guess. It doesn't sound insurmountable, especially if the auto manufacturers worked together on speccing it out.
Recognizing the roads is not the biggest problem, it seems. It's the myriad of purely human situations. Anything and anyone can suddenly appear on the road in front of the car, not just traffic cones. Which means the AI needs a pretty much full understanding of the world it operates in.
On the other hand, isolating the roads for AV's from us may become a major inconvenience. I really don't see how this tech can evolve today, unless cars learn to fly already.
You don't need a long train, you just need enough to cause most people to not drive, then the grocery store will come to you.
Seriously, though, check out the density of grocery stores in areas with good public transportation. Your grocery store is far away from you, and huge, with a giant parking lot, because of cars. I have 3 grocery stores within a 5 minute walk. I also have 3 bakeries, a general butcher shop, two chicken butcher shops, 4 convenience stores, 3 dry cleaners, numerous clinics, and too many restaurants to count.
I think the real innovation would be having my car be equipped with a bunch of sensors and cameras, and then have a well paid dedicated remote driver in a third world country doing the driving for me. I'm only half joking.
It's pretty crazy how tunnels can sometimes just pop out of nowhere such that an automated system with extensive mapping can't predict them more than 5 seconds early.
That’s my point. The moment that a car says the safest thing to do isn’t to veer off the road (maybe to avoid a large compressed gas carrier in front) even when the human says it is, then we’ve achieved L5
Perhaps we should just bring back horses. BI as in biologic intelligence. Where I'm from it was common merely 100 years ago to exit the tavern, plonk yourself into the wagon, let the horse free and fall asleep in pleasant stupor. Usually the horse had no issues finding its way home. Usually. Sometimes it supercharged itself on a nearby pasture. Crashes were uncommon.
If you think this is sarcasm, it is not. That horse was clearly more reliable, adaptable and performant, not to mention safer, than anything Tesla or Wayomo is offering today.
Horse manure is mostly dry and inoffensive, plus solutions to that problem have existed for hundreds of years in the form of bags hung from the harness, to collect the manure, which makes for excellent fertilizer.
I see what you are saying. We don’t need to build something as smart as a human. It just needs to be as smart as a horse. ABI. We can even train ABIs using horses. Where are you from? East europe or central asia? South america? I imagine itd be easier to train ABIs in certain countries. Heck even in western europe they can start with the beer deliveries. In america, the ABIs wont be so smart, they are limited to safe tourist areas.
This viewpoint seems to be a strange version of Schneier's law. Just because you learned ML and can't think of a way to make a car drive itself doesn't mean it is impossible.
I believe within the next few years we will see driverless long haul freight operations. Freeways are orders of magnitude easier than dynamic city streets, longer routes can be taken at little cost to avoid adverse conditions (a human rider isn't going to be ok with adding 10 hours to avoid a snow storm, a truck load of coffee beans doesn't care), and the cost of humans in this case is stupid high.
Once it is demonstrated in the real world that corporations stand to profit from eliminating drivers, you'll start to see pressure on the public sector to annotate the world. Stop signs will by law include an RF beacon, lanes will no longer be defined by paint but by transmitters in the pavement, federal law will mandate that personal vehicles transmit their location and dimensions to other nearby cars, etc.
Eventually once we have the infrastructure built, every new car will piggyback on it. Remember the Interstate Highway System was started in 1955 and completed in 1992 - progress just takes time.
Annotating every object on the road with RFID beacons seems like it actually would be an impossible project. Yes, you can attach beacons to stoplights and signs (although they already have “beacons” that display in the visible light spectrum).
But you can’t attach them to people like the homeless pedestrian killed by Uber’s test self driving car, animals, trash thrown on the roadway or falling off a truck, or a boulder falling from the side of a mountain.
And looking at the decades-long process it took to install PTC on train lines, it would take decades to install radio beacons on everything we can and by then technology should have advanced to the point where we don’t need them.
I saw a trash bag on the highway yesterday. It was full but looked light, as if it were stuffed with pillows or plastic bags. I could have safely hit it with my car (I didn't) whereas other objects that size I wouldn't. This decision making happens often like with pieces of paper or plastic bags. I thought about how much contextual knowledge it takes for us to make these everyday decisions.
There s however one thing you cant do but an ML solution probably could: share your contextual experience at the instinct level with every driver.
Say you're right and an ML model cannot know intrinsically the trash bag is safer to hit than to avoid, however, if a few car hit or avoid those for a little while they can learn of the consequence of both actions and then share them back to every other car.
I suppose as long as it can properly learn, it'll get there (and, more valuable: STAY there) eventually. The brain is an efficient machine that can work in highly complex network, but I dont know if the fact every brain is born mostly blank of experience is a good or bad thing to build much more complex specialized behaviour like driving with 0 death in an entire nation.
The leg up that I gave over a super-networked AI: I’ve seen a trash bag before. If the cars have to hit every item they see to find out what happens then we’re going to be here a while!
> share your contextual experience at the instinct level with every driver.
> Say you're right and an ML model cannot know intrinsically the trash bag is safer to hit than to avoid, however, if a few car hit or avoid those for a little while they can learn of the consequence of both actions and then share them back to every other car.
He is definitely, 100% right.
All this talk of networked AI/ML is a particularly insidious form of evangelism. Because it doesnt answer the questions that need asking.
There are people who will go out of their way to mess with driverless cars and I think the car needs to handle that somehow.
Heck, I’ve seen kids mess with real cars. Like the bag you mentioned, some idiots were taking cardboard boxes, filling them with rocks and putting them in the street waiting for the drivers to think “oh just it’s an empty cardboard box, not gonna slow down for it”.
Somehow I can see an AI driven car, especially without a person inside, becoming a target of bullying like that.
It varies from state to state, but on average Type 1 (the most important) signs are replaced every 7 to 10 years.
If you tag each one with some sort of RF beacon, not only do you tell the car "this is the spot to stop" you also have an absolute fixed reference point and can correct GPS or other navigational systems.
At the point you know where everything is, and cars begin to share their location with each other, you have an absolute map of everything in the environment. If LIDAR detects anything that isn't part of the gold standard of what the world looks like, _that_ is an exception and handled as a danger (the person Uber should have detected had they had all the sensors turned on).
It would be sufficient just to have a) consistency, and b) 2D bar codes.
In most countries, every council does their own road markings and signage. Sure, there are laws, but they're implemented inconsistently. Worse, those laws are often ambiguous, intended for humans, and defined at the state level.
Ideally, all of the critical markings should have a single, consistent, internationally agreed standard. This doesn't have to be every road sign, but all of the really important lane markings and basic speed limits and stop signs should be consistent.
Then, these signs could have infrared 2D barcodes layered on top, which would guarantee machine readability.
Humans would benefit from the consistency, but wouldn't be able to even see the bar codes.
Cones are already, within a specific jurisdiction, a rather specific shade of orange. Except for sunlight, and shade variations, and fog, and dirt, and mud, and snow. An orange cone is easy to detect, visually. The problem is all of the real-world details and complications. If cones are the issue, the first city in the world to be brave enough to issue a requirement that all their contractors use cones that are unavoidably bright to self-driving cars (whatever the specific technology for them to be unavoidable. A discussion of how the shade of orange is/isn't enough, vs a radio marker isn't interesting to me), and will license companies for self-driving cars in that specific jurisdiction, will ride to prominence. Self-driving car companies will move there, and so will tech companies and then.... So Arizona realized this and did that a few years ago. The technology's still not there but it has become some sort of hub. If the Federal government had the bandwidth to do something about it and set standards and work with manufacturers to ensure it was safe along federally recognized guidelines, then.... Okay I'm sure the big players have lobbiests and that they're working with government to do so. The problem is that'll take time and, oh hey, have you seen the news lately? I get the sense the lunch to talk about self-driving cars got cancelled.
Small nit - drivers are only about a third of the cost of shipping. Adding 10 hours to remove the driver probably wouldn't make sense in a lot of cases.
I don't think stop signs with RF beacons is needed.
The cameras on the car can provide just as good an image as what a human eye can. The problem to be figured out is generating a mental model of the environment and reasoning about it.
> Freeways are orders of magnitude easier than dynamic city streets
Yes they are. Although it's still not an easy problem. Occasionally I get sucked in to watching highway accident and road rage videos on youtube and try to consider which ones a ML could realistically handle. It's still a very complex random environment, even on interstates far away from cities.
I don't believe full self-driving has any chance of being a reality this century. It's far too complex a problem in an environment where all the edge cases do matter.
> annotate the world
This I do believe may start to happen in coming decades. But this is effectively an admission that actual self-driving isn't really possible, so the problem gets redefined to a smaller and easier scope.
Although even that will only work in limited access roads where nothing else can be present. Like barrier-separated HOV lanes but for these trucks.
He learnt the field and know what can be done and what cannot be done. Go lookup the history of Ai with lisp and then deep blue vs Karpersky. Back then people believe humanoid like Star Trek Data is possible within our lifetime. If you have studied the AI of its day you would know it is near impossible. The OP above you is right, based on current ML knowledge self-driving car is impossible. Assisted self-driving car is more likely. The problem of self-driving car isn't like chess or go. There isnt simplified rules. Humans and humans needs are messy. What you're thinking of is some new technology say super LIDAR++ and new method say doubly deep organic learning might solve the problem. But as of now, they are not yet invented or extremely infant stage. But one thing for sure those aren't the current knowledge ML. Perhaps Waymo will get there...perhaps no. Many decades ago, there are tesla-like cars but it was successful because of battery tech and cheap oil price. They basically went out of business. So Waymo maybe be like them, full of possibility but the isn't ripe yet. Just like Friendster and MySpace before FB.
Remember Taylor Series or polynomial expansion from calc I. You can fit something until it doesn't. With a really large number of terms you can fit a lot.
I feel like the problem involves a liability shift.
If you stand or park on a railway track and get creamed, the pedestrian is going to be the target of blame.
We've come to the conclusion that the track belongs to the trains; at best the engineer can perform a limited, good-faith effort to stop and everyone knows that's not going to happen when you've got a kilometre-long, 10,000-tonne wad of physics rushing down slick metal rails at 60kph.
But if you walk into the middle of I-5 at rush hour, we still throw the legal and financial liability at drivers. Everyone is supposed to expect a rando suddenly stepping into the road and brake/steer around them. That's hard for humans and potentially harder for machines.
I could imagine a network of "autonomous only" roads which were much regulatorily much more like the rail infrastructure-- designed for minimum surprise (limited on/off ramps, limited turning points, predictable advance signing and marking, perhaps a block-control system to mitigate traffic flow) and limited recourse for pedestrians or human operators who enter the territory.
I don't think I-5 is the problem. City streets are the problem. Cars, bicycles, and pedestrians will always mix there, and there is no reasonable street design where that won't happen.
So, genuinely curious, how can it happen that a system that's been developed over 12 years, can not deal with a common road obstruction? If something truly rare/dangerous happens I can imagine it not being able to deal. But it seems this sort of situation is not only easy to deal with, it's actually easy to predict.
Maybe not something you could predict in a product with just a couple years of time, with just a couple employees, but we're talking over a decade, and thousands of employees. Oh and in Phoenix Arizona, one of those ultra uniform grid cities the US is famous for.
What's stopping them from having their AI operate in simulations in literally every possible traffic situation? Every location a pot hole can be in, every type of traffic cone arrangement, every kind of jaywalking, every kind of vehicle approaching any kind of street from any kind of location, possible speed, reaction time, everything? How could a government approve any autonomous agent that is not tested in such a way?
My world view gets rocked a little everytime I see a headline like this. I want to believe these companies are ran by the smartest engineers on the planet, but then I read about a Tesla running into an emergency services vehicle, or a Waymo being confused about traffic cones and I can't help but wonder what the heck are they doing?
It's because they're trying to code the problem instead of solving it end to end. As you can imagine if you're trying to code driving scenarios you're never going to finish. Only Tesla and Comma.ai are on the right track. Think Alpha Zero vs Stockfish.
This problem has to be solved end to end. It can be done with just vision as humans can drive with just vision. It's literally just a software, machine learning and data problem. Just takes time
Except didn't Leela Zero (the closest thing we have to Alpha Zero anyone can actually use) beat StockFish, then StockFish was upgraded to use neural networks in a few edge cases and otherwise keep doing what it was doing and now StockFish is back on top? Seems like a Waymo-style approach is actually on top there.
Except Tesla is doing the exact same thing as it's the only practical approach at this time. The only place Tesla is extensively using general approaches is in building a model of the world from camera inputs. Other companies aren't doing that because they get much, much better results than Tesla simply by using lidar.
> What's stopping them from having their AI operate in simulations in literally every possible traffic situation? Every location a pot hole can be in, every type of traffic cone arrangement, every kind of jaywalking, every kind of vehicle approaching any kind of street from any kind of location, possible speed, reaction time, everything? How could a government approve any autonomous agent that is not tested in such a way?
Because this approach has huge limitations, principally that you’re now optimizing against a cost function that isn’t the real world. RL has a place for sure, but limitations.
It’s like saying that you’d expect someone who’s played 10k hours of Call of Duty to be equivalent to a Navy Seal. They’re just really good at playing the simulation, not the real thing.
I'm not saying they should optimize for these simulations, just saying they should test for them.
Also simulations are definitely a part of military training, and if they perform badly in the simulation they'll have to fix that before going on a mission I'm sure. I don't know much about Navy Seals specifically, but I've got friends who built full body VR suits for special forces training.
Agreed mate. It's like that grumpy bloke from England who can only swing the ball when the weather is perfect - clouds hovering, moisture in the air, slight breeze and yes, some fish and chips during the breakfast.
Yes full self driving is not happening any time soon unless the roads and cars are purpose built for it.
Everyone here has worked in some sort of tech company where some procedures are just wrong, some training material doesn't work, or the content is full of rot and unmaintainable, sometimes the code of the product has a couple bugs, never to be fixed, or the back office.
Or better how often people make human mistakes and general imperfect things are wiped under the carpet.
Code testing? Happens on all new features and to what degree?
Think of all these things and then reflect on the possibility to fsd implementation within, say 10 years.
Not going to happen.
I don't know anything about AI, but I've heard that asking "how do humans do it?" is not the right question.
Still, though, how _do_ humans drive? One answer that comes to mind is "humans are smart." True. But I suspect a severely "dumb" person can drive safely in a variety of novel scenarios.
It makes me think of child development, all that time making sense of depth perception, distinguishing objects, fine motor controls, expectations of motion, etc. It takes years. Could a non-human animal trapped in a "car suit" drive effectively, if it could somehow overcome the panic? Maybe it could maneuver safely, but what about a four-way stop?
I have no idea what it takes to drive a car in general, and specifically I don't know how humans do it. Where do you even start when trying to model the task in a computer?
224 comments
[ 98.6 ms ] story [ 616 ms ] threadA long term snow covering of lane markings typical for the winter months of much of north america and the emergent driving lanes people flock to cannot be handled by any existing driving AI. The places people drive are wrong according to absolute road positions and the relative markers are obscured to both human and machine. Unless car AIs can do the wrong thing like all the humans will in those situations it won't work. And that's a hard problem.
I don't know anyone from Arizona, but the Californians I've known adapted quickly.
When I learned to drive a car, I lived with an uphill driveway in a town whose winter climate varies by the hour. It's quite normal there to get temperatures around zero, some rain, then the temperature drops and 20cm snow fall. So on a January morning I often had to start the car uphill on steel ice covered with water and snow. With a sharp curve as a bonus.
That kind of start isn't the simplst, and you could persuade me that that kind of thing is as difficult as navigating an intersection with ~10 other cars, cyclists and pedestrians. But not that it's tens of times as difficult.
Being able to drive in California doesn't include all the skills needed for e.g. winter driving in Norway, or for driving in an Indian city, or (list goes on), but I fail to see that it's just 1% (as an upstream poster claimed) or even close to 1%. I fail to see that driving skills are 99% intransferable.
I also fail to see that while the skills are transferable for humans, they are intransferable for computers. Various people in the thread assert so, but argument by repeated assertion is meh.
That they are nontransferable is the null hypothesis. I mean, we need evidence for this transfer. It's up for self-driving companies to produce this evidence. But they didn't, unfortunately.
Because AI is not even remotely close to “solved”, and driving is a combination of a million edge conditions, social knowledge and communication, common sense knowledge, vision understanding, planning, and force feedback adjustment.
What’s unclear to me is why you’d expect these things to be so easy. To even say “for computers” doesn’t seem to recognize this isn’t just about executing basic math, but is rather trying to represent intelligence. There is no single computer here, these are implementations of ML models, which are each unique with their own architecture, parameter space, and capabilities.
Have you noticed how a small child will learn to identify things like cats and birds after only seeing a few examples? But computers need orders of magnitude more data and instruction.
Anyone who doesn’t understand how modern day deep learning works really ought to do an intro course.
I would consider it 99% complete when I can pull up a map, point to any stretch of drivable road in the United States and ask it to autonomously operate there.
Specifically, there's an especially tricky road in I-35N in Texas where the lane markings come ago, yellow shoulder markings _merge_ into barriers, and the road condition is so bad that the steering rack will _turn_ by itself(!). We're not even close to calling this thing autonomous. If and when it gets to that point, I would consider it "99%" complete.
In reality though, I really see this tech being viable in continued controlled conditions. Maybe there will be a point when certain lanes will become dedicated "autonomous lanes", where they're completely isolated from the rest of traffic, come with special markings and sensor suites to assist driveler-less cars operate efficiently.
As long as the chaotic human element is ever present, regardless of how well they model these systems, it will fail to cope with the massive amount of discontinuous information humans inject into a situation (swerving into a lane, road rage, cargo suddenly coming loose from a truck, etc.)
Oh, and have we tested these things during winter? :p
Seems like you could just get some data on that before declaring that "we're not even close". How do shipping Teslas do?
FWIW: my impression is that the things people think are hard aren't the problems that are actually hard. People here and elsewhere on the internet have spent years screaming about lidar vs. radar vs. vision. And... it turns out vision basically won. Something like half the Tesla fleet in ths US are vision only in the US now, and... the cars see stuff just fine. We're not having detection failures. It works.
If you watch all the FSD beta video being spread around youtube, much of it by drivers specifically looking for edge cases like this, most of the remaining problems are in pathing and decisionmaking, not detection. The cars see a busy intersection and then get paralyzed (there was an amusing clip from one guy of a car trying to make a left turn, giving up, and repathing a right turn that took it back in a loop only to fail at the same left turn repeatedly), or creep too slowly and annoy other drivers. Or they choose the wrong lane, or misread a multiple traffic light environment, or misread the need for detection (the one really dangerous video I remember had the car deciding to take a left turn across a broad street with a complicated blind corner instead of creeping out farther).
Those are bugs to fix, for sure. But they're not the bugs we thought we had to fix. Those bugs... got fixed.
So you can look at this as a pessimist and say "we'll never get there", or you can judge from previous experience that the hard problems won't turn out to be that hard in the end.
Personally, having been watching the progress in this space, my guess is that Tesla and Waymo (and maybe Mobileye -- no one else is close) get this basically fixed within a year or two.
At which point the debate will shift from "we'll never get there" to "we'll never be able to regulate this appropriately" or somesuch.
You and I have extremely different experiences with Teslas. About the only place I could conceive arguing it works is highways and even then I wouldn't trust riding one down the highway without a steering wheel. Hell, my coworker can't even call his out of the driveway because it has been thinking his mailbox is a car for the last 6 months.
I think we'll get there and I don't think it'll matter if it's with LIDAR or pure vision but we're certainly not close to being able to remove the steering wheel (regulation aside).
But yes: the example was an interstate in Texas. And I happen to know it actually works quite well on interstates. So a very reasonable experiment about whether or not automation "will never get there" on edge cases like the referenced interstate is to take a commonly deployed, already available automation technology there and test it.
But also recognize that shipping autopilot is now about a year old. All the FSD work hasn't made it to released cars yet. It's very much evolved, but visible only on youtube right now.
https://www.foxnews.com/auto/tesla-smashes-overturned-truck-...
How often do people hit a giant stationary object in the middle of a nearly-empty freeway?
Good way to change goal posts. There are many, many instances of Teslas hitting large stationary objects on roads, and who knows more instances of near misses due to driver intervention.
When the AP can be engaged, which is usually 'better' conditions. Unfortunately humans don't always have the option of driving in 'better' conditions, so this is an entirely cherry picked subset of data.
https://youtu.be/antLneVlxcs?t=633
Edit: FSD 9 accelerates across double lines toward bridge support in San Francisco. https://youtu.be/GlIdu7prsAw?t=155
https://techcrunch.com/2021/05/07/tesla-refutes-elon-musks-t...
[1] https://www.youtube.com/watch?v=7UF-S2czdCk
This is not necessarily a job for lidar, they are just omitting some checks.
At second 18 [1] a street light comes into play which looks similar to the moon. That relative movement is what the system should have been expecting from the moon, but they apparently just re-computed the assumption that the moon is a traffic light which is moving along with the car.
It's somewhat revealing that they are not doing these kind of checks. I actually can't believe that they aren't doing it, but I can't see a case where the traffic light would always be at the same spot, if you have velocity and acceleration data at hand.
A traffic light is a very important thing.
[1] https://youtu.be/7UF-S2czdCk?t=18
However, people do uses percentages metaphorically and we should try not to be so distracted by silly headlines.
See https://wtop.com/dc-transit/2020/09/study-25-percent-of-car-...
Not until you added the geofencing I wrote about :-)
> with "the last 1% [being] the hardest."
I might have misunderstood the context here, but my point was to make an example were the last percentage wasn't that hard.
Waymo Is 99% of the Way to Self-Driving Cars. The Last 1% Is the Hardest
They picked a good first experiments location but it's NOT representative of more than 1% of roads on the planet! It's a bizarro-world of simplified challenge and problem space. Drop such a system into any real world driving and it will fall flat on its face very quickly and endanger both passengers, other drivers and pedestrians.
"stick to the right as far as possible" is not a reliable approach when the emergency lane may or may not be there.
The case I was thinking of is a highway I was in last week, where temporary work has one of the directions going through the other direction, so the 2+emergency lanes in one direction become 2 lanes without emergency in one direction and one lane in the opposite one.
In this case, there are temporary lane markings on the ground indicating you should not follow the old ones, but rather align yourself further to the right then you would normally do.
OTOH, I am familiar with the situation in Rome where often the lane markings are deleted and a road may be 2-3-4 lanes depending on how people queue up, in which case people would generally rely on social/experience cues, which I'm unable to formalize.
If a driver assistance system can't handle this sort of situation, that's fine. I'm sure it's still useful most of the time. But it's clearly not "99% of the way to FSD."
But I don't really disagree. Some mechanism (perhaps a redundant mechanism) when approaching a possible construction area to give a human a minute or so to take over.
They just don't address use cases where people aren't able to drive at all. Which are interesting to a lot of people but they may just not happen for a long time.
There are people who just want cheaper Ubers but there are others who are fine with handing off long boring sections of highway driving.
As part of that localization, there will be a metric of how good the local solution is vs. the map. If a divider moves, for example, the data will be inconsistent with the map, and the software will know that something is amiss, without having to intelligently classify anything. The software can drive confidently when in a consistent environment, and drive conservatively when in an inconsistent environment.
Generally, any object that isn't in the map will stand out obviously. That's where having stuff like cameras becomes useful. Asking a computer, "is there a construction worker in this 20MP image of a street?" is hard, but asking a computer, "is this 0.1MP human sized blob a construction worker?" is a lot easier. Waymo has been responding intelligently to bicyclist hand signals for years. "Is this sign shaped blob a construction sign?" is even easier.
The lidar and mapping is a key difference between Waymo and Tesla. Waymo works because they have the big data infrastructure, and the budget to deck out every car with orders of magnitude more sensors and compute.
Why are we ok with streets in such slapdash condition in the first place?
Yes I think self driving cars need to handle it, but I also think we aren't handling it.
Self driving cars have the potential to greatly reduce deaths due to accidents and improve accessibility for often ignored segments of our population. It is sad to see it so close, but held back by reasons like this.
Also,
> and right now, no driverless car from any company can gracefully handle rain, sleet, or snow
Good thing the entire US has the weather patterns of the bay area, and never experiences rain, sleet, or snow!
Hindsight being what it is .. they should have built a new 8-lane highway a few miles to the West rather than attempting to widen it multiple times (with all the bridge replacements needed) through towns like Temple and Waco. But as the Texas Central Railroad has found, you can't just eminent-domain land in Texas for large infrastructure projects.
Create small emergency landing areas all over...
The only ground based vehicles I care about being 99% auto would be long-haul trucking. With a station to pickup a local last-mile-human-driver for actual drop-offf etc.
But only IFF there is no external accident and no construction site.
If both would be handles truckers could sleep/nap while driving on the highways, which could have interesting effects tbh.
Handling construction sites could be made feasible.
But I have no idea how to handle accidents happening around/in-front of the truck in a good way.
Similar I would be worried about improvised marking of fresh accident sites before emergency responders arrive...
Human skill is extremely variable. A car that can drive to any road under any condition will be vastly superior to the average driver.
Reducing such a complex system to a single percentage may be silly, but the idea that a car might have partial coverage under certain conditions is quite reasonable and is reflective of how real human drivers work as well.
Some randomly placed cones are enough to confuse humans in many circumstances, as well. I doubt we'll ever have an all-encompassing solution for that, either. Even for things like autonomous-only lane, well it's not like everyone obeys the HOV lanes currently...
There probably isn't a perfect solution, especially when I've seen plenty of ambiguous markings.
Humans absolutely can do that. But the very architectural basis via ML that's being used never will. Different architecture? Maybe. Maybe in 50-100 years.
Why's that? Can't ML be trained you handle these conditions?
There are situations humans can't handle, and notorious "hot spots" where crashes repeatedly happen, but if AI is more consistent, it might lead to more concentrated failures.
Have you heard about the problems caused by navigation routing, where large numbers of people are directed through roads that aren't suited for heavy traffic?
An unprotected left turn is probably a harder problem. They'll be expected to do it aggressively like humans, and also expected to do it safely. Those are incompatible directives.
In general, my expectation--which may be completely wrong--is that we'll see autonomous driving on sections of highways in good weather long before we have door to door almost everywhere. Which is useful by itself for long highway drives.
I'm not sure how they can do better with far more stringent resources and requirements.
Not only are there different signs, different line marking & colours, & different widths. Cars are different: not all models are sold equally throughout the world. People are different: through their clothing, height and build. But the really big difference is that not everyone drives on the RHS...
The article lists left turns as a problem, whereas on a LHS road system, right hand turns will be the problem.
In NZ (a LHS system), I've heard that turning right across traffic has priority over oncoming traffic.
In Melbourne, VIC Australia (also LHS), right turns are made from the left most lane because of trams lines.
I'm sure there are loads of other issues specific to other geographies.
To me, this never ending self driving experiment screams out that the wrong solution is being pursued. (That doesn't imply that I know what they should be using as a solution)
This is incorrect and would be utterly chaotic. You may be thinking of an outdated rule in NZ where right-turning traffic takes precedence over left-turning coming in the opposite direction to the same side road. This was the case for a long time but was reversed[0] in 2012, overnight, almost without a hitch.
[0] https://www.nzta.govt.nz/driver-licences/getting-a-licence/r...
I have yet to work with an automotive OEM that doesn't consistently mis-classify some speed limit signs in some common cases that are completely unambiguous. It leads one to wonder about the quality of classifier design and testing such that these end up in production vehicles. The cases that cause a classifier to fail are different across automotive OEMs too, it isn't intrinsic; other automotive OEMs will handle that specific case just fine.
On the other hand, when it works it often works surprisingly well. The good/bad news is that most of the problems I see are ultimately caused by poor model design and testing. These are all surmountable with better quality AI engineering.
There is more AI to self-driving cars than the classifiers that pull features out of their environmental sensors. However, the driving AIs are making decisions based on the output of those classifiers -- garbage in, garbage out.
These are the kinds of situations that will produce relentless incremental improvements over time, and there is still a lot of low-hanging fruit to improve on.
But what if machines become the dominant drivers? We need to make the roads machine readable: Road signs redesigned 'QR Code' style, maybe even some kind of wireless broadcast system to communicate traffic light changes.
The hard question which really has nothing to do with the tech, is what we do if a self driving car kills a person. Made worse if the accident could be avoided by the majority of human drivers.
we have already had exactly such a case - Uber. Even though Uber was totally negligent, nothing happened to them, beside probably some settlement behind the scene. In part they were able to wiggle it by showing a bad dynamic range video (which is completely different from how human eyes see in such situation) there the victim appears as if out of nowhere.
I think that case just sets the pattern that the autonomous vehicles will not be judged according to human driver standard.
Buying a governor really payed off for them it turned out.
https://www.bbc.com/news/technology-54175359
So what happens with the human driver (you) if the same accident happens ?
If the Uber case sets the standard, then negligent homicide is a likely charge. The sentence for that in the US is "A minimum of 4 years in prison and up to a maximum of 8 years in prison" according to this link:
https://www.feldmanroyle.com/homicide/negligent-homicide/
You better hope you can explain it was a bug in the software in this situation.
What might make sense is limited deployment in areas where the problem can be constrained: dedicated bus routes, truck convoy lanes on an interstate, etc.
For the rest of it we should be focusing on how to get people out of cars since even BEVs pollute far more than buses, rail, bicycles, or walking.
Some kind of compromise between asking people to cram themselves like sardines against a bunch of strangers, while still allowing people to sit next to their family members and friends.
It's convenient to say a desire to not jam up close to strangers implicates its holder in some terrible character flaw of not caring for ones fellow human. But if you really want to get more people on buses it's a desire we'll have to accommodate and contend with.
I'd highly encourage you to move to the middle of the country and live a 30 minute drive outside of town (because that is all you can afford). You'll quickly realize that super bad evil cars are the means by which a good percentage of the population has access to fresh food, medical care, and other basic needs.
Also note that I’m not disagreeing that mobility is important but that we literally cannot afford to continue polluting the way we have. You tried to make this emotional with the “super bad evil cars” phrasing but it’s a simple engineering question: right now, people have built lifestyles based on low subsidized fossil fuel prices and being allowed to ignore externalities. I’m aware of what that lifestyle is like – and how often it’s not “all you can afford” but “where you can afford to buy the house as big you think you deserve”, too - the latter being far less sympathetic when asking everyone else to subsidize it.
When energy is cheap and you can ignore pollution, you can drive an overpowered vehicle on frequent trips with minimal use of the total cargo capacity. If the cost goes up, those calculations all change: people pay attention to fuel efficiency when buying vehicles and combine / reduce trips, invest in household efficiency, etc. Cost-constrained American rural dwellers and those I’ve met in other countries with higher fuel costs don’t drive a vehicle designed to haul livestock to pick up groceries because it’s overkill.
That extends to things like zoning: the majority of people living in exurbs for financial reasons are doing so because closer in development was low density, often required by code, and significant amounts of land were required to be used for car storage.
Part of climate change mitigation will be reversing those problems, and that kind of thing seems like a more fruitful area for us to be spending time than trying to make high-pollution commuting more appealing.
Eg. Putting the current speed limit on the speedometer
Deliveries should be done by autonomous air drops.
Transportation by autonomous aircraft.
If this were any other startup it would be considered insane and never gotten to this point. Don't know why people keep dumping money into this project thinking they're just around the corner.
Even Tesla sold some cars when it raised a fraction of that amount, and $70 million came from Musk
> By January 2009, Tesla had raised US$187 million and delivered 147 cars. [0]
[0] https://en.wikipedia.org/wiki/History_of_Tesla,_Inc.
The product is being tested. Many things, including life changing drugs, are developed this way.
We both agree Waymo definitely doesn’t sell its half baked product calling it Full Self Driving before it’s ready.
They're whole plan lacks a feedback mechanism.
You don't even know the return if they succeed. Drug companies have a long history so they can anticipate the cost of development, odds of success and payoff if successful. They do all these things prior to any investments being made.
The closest Waymo has is # of truck drivers * salary = profit.
The top-down Waymo method is a failure.
They've explained this very often. Unlike Tesla, they don't think you can go through the intermediaries without the dangerous situation of drivers not paying attention because it usually works. Based on their experiences a decade ago of employees using it as drivers assist, people check out and it becomes more and more dangerous even if the software is improving.
I don't even know what their plan is. Do they want to sell cars, operate trucks, license out the software, have robo-taxis?
As far as expansion plans there's no reason to think each place they go will require thousands of engineers.
And although it does occasionally have issues, it is remarkably reliable.
It's just that a part much larger then 1% of the way is perceived to be just 1%. I.e. there is a insane lot of hidden complexity generally/often overlooked.
Also, often we'll change such titles (as I just did above), and then the shallow-title-objections become a sort of uncollected garbage, referencing something that no longer exists.
I'm going to make a stub reply and bundle those comments underneath it so as to collapse them. There are more interesting things to discuss here.
Yeah, didn't think so.
Also it doesn't account for their model which requires paying engineers and installing tens of thousands of dollars in every vehicle to make it work. And the fact that they need pre established HD maps to do anything.
Climate change is 99% solved. The last 1% is the hardest.
We are 99% of the way to ending all wars. The last 1% will be the hardest.
Scientists are 99% of the way to curing cancer. The last 1% will be the hardest.
etc.
I'm certain there will have to be smart roads before there can be cars without a steering wheel.
At first glance I thought the title was saying that a Waymo car carrying the CEO has been lost for days because it got trapped by traffic cones.
Thanks for posting it in the right place though :)
So it's ok to change titles to something that is nowhere near what the original title meant when posting a new url?
This is standard HN moderation—it might be good to review the site guidelines: https://news.ycombinator.com/newsguidelines.html. Lots of examples at https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que....
I will still take a thousand dollar bet from anyone who thinks self driving cars will be here in 10 years. No one ever accepts!
Or do you have some specific definition of what qualifies as "here" that generally makes people unwilling to take this bet?
My requirements:
- The time of day or night of my choosing
- Anytime type of weather - except:
- Will not be during snow or with excessive ice coverage (although this should be fine, IMO)
- Must not have a driver physically there who takes over, nor be driven primarily by remote drivers (remote support could be an option, but it's not a "remote control car" robo taxi)
If you accept, I'll arrange it and wager the $1000
I think this is one of the more interesting pieces in the article for me. We all kinda know by now that there are issues around AI handling all driving situations. I think it's interesting to hear Waymo having difficulty with the manufacturing aspect.
It does make a certain amount of sense. Waymo is a company that is mainly trying to solve the AI problem around self-driving cars, but other things can stymie those efforts like assembly. This isn't meant as a dig at Waymo, just more an observation that just because a company is excellent in one area doesn't mean that they will be excellent in all areas and that even excellent companies like Alphabet/Google/Waymo can stumble in new areas.
It doesn't matter how good your product is if you can't actually ship it.
Terrified, and behind by 10 years. And by terrified I mean their CEOs. Do you think the BMW CEO would still have his job if he didn't squander their EV efforts that started about 15 years ago?
The BMW CEO getting bounced was the warning shot that made every automaker executive sit up and realize: my company either has an EV platform program underway, or the investors will bounce me.
If they are still not taking Tesla (to a lesser degree Rivian and Lucid) seriously, they're in big denial at this point.
GM and VW have basically woken up. Ford and Mercedes are toe-dipping. BMW seems paralyzed by internal management dissent,
Consumer Reports 2021 report puts Tesla in 16th place among US car brands. In JD Powers 2021 Dependability Study Tesla came in 30th, followed only by Jaguar, Alfa Romero, and Land Rover.
Terrified? When Tesla will sell more cars maybe, they are badly lagging against virtually any other car maker, let alone all of them.
One of the reasons tanks exist is to protect the operators. If you’re not going to have boots on the ground in the first place (autonomous tank), you can rethink the vehicle completely.
>tanks exist is to protect the operators
operators of a large gun quite close to the target. If one can do without such a gun in such a role on the battlefield, then the need for tanks become less. We do see that in the recent wars - like Azerbajan/Armenia - the high-precision stand-off weapons fired from drones did that job perfectly. There were no countermeasures though, like say electronic communication and positioning blocking which would have necessitated autonomous capabilities for the drones.
Of course there are. Drones can fly given waypoints and return to base. Hell, even consumer quadcopters can do this.
I would, of course, love to hear if folks here think I am wrong.
Anyone committed to self driving cars needs to be a road building company and own the road network that will manage these mostly-automated efficiency lanes, with car provision (more likely tech/compatability licensing) as a secondary income stream.
Google running this as a pure Taxi substitute is seemingly a completely viable option today. They don’t need to be 100% autonomous as long as one person can handle a dozen or more cars that’s easily good enough to out compete traditional taxis or ride sharing services. Granted this is limited to a 50 square mile area, but collecting more data isn’t the difficult part.
Is this a viable business in a 50 square mile area? Is it scalable nationwide? Is it price competitive with traditional taxi and ride share?
The overnight use case is compelling as it opens entirely new transit options. Cars could be made competitive with air travel for the overnight use case. And traffic wouldn’t be nearly as large of a concern for most if they weren’t driving.
Yes to both, critically you can use a hybrid approach for outlying areas and cities lacking regulatory approval. By comparison DC is only 61 square miles and San Francisco is 47 square miles so this is plenty of area for local service and a reasonable benchmark for expansion.
What’s not mentioned is the cars do need someone to clean them and fill up the tank etc, but that’s hardly an additional expense.
Sure it's EXTREMELY limited but that doesn't seem like an accurate description since they trust it enough to be liable for it to drive by itself or stop safely when it can't. Given the number of their cars going around SF lately I think they'll expand some soon too.
It is actually less than that, since there is no Waymo car you can go out and buy. Driverless Uber rides are absolutely useless. When I order an Uber, how does it matter to me if it is driven by a human or a computer?
Modern cars depend on modern road infrastructure. Like traffic lights, lane markers, traffic engineering, etc, enabled the efficiency and convenience we enjoy today.
And this infrastructure was built for human driver. And they do not work well for software driver.
How much would it cost though for parallel infrastructure that is suitable for a software driver? (better signage, rfid tags in the roads, 'smart' traffic cones, etc.) For say the Bay Area. 10 billion to a few hundred billion? Speaking for myself, I'd be happy to pay say an additional 1% sales tax if it meant facilitating an area wide 24/7 autonomous 'public transit' taxi service, presumably with at-cost per mile fares.
On the other hand, isolating the roads for AV's from us may become a major inconvenience. I really don't see how this tech can evolve today, unless cars learn to fly already.
Seriously, though, check out the density of grocery stores in areas with good public transportation. Your grocery store is far away from you, and huge, with a giant parking lot, because of cars. I have 3 grocery stores within a 5 minute walk. I also have 3 bakeries, a general butcher shop, two chicken butcher shops, 4 convenience stores, 3 dry cleaners, numerous clinics, and too many restaurants to count.
Cars ruin your city and you're simply unaware.
Try to jerk the wheel into the woods and it won't obey.
It’s unlikely this will ever be the case.
If you think this is sarcasm, it is not. That horse was clearly more reliable, adaptable and performant, not to mention safer, than anything Tesla or Wayomo is offering today.
Horse manure is mostly dry and inoffensive, plus solutions to that problem have existed for hundreds of years in the form of bags hung from the harness, to collect the manure, which makes for excellent fertilizer.
I believe within the next few years we will see driverless long haul freight operations. Freeways are orders of magnitude easier than dynamic city streets, longer routes can be taken at little cost to avoid adverse conditions (a human rider isn't going to be ok with adding 10 hours to avoid a snow storm, a truck load of coffee beans doesn't care), and the cost of humans in this case is stupid high.
Once it is demonstrated in the real world that corporations stand to profit from eliminating drivers, you'll start to see pressure on the public sector to annotate the world. Stop signs will by law include an RF beacon, lanes will no longer be defined by paint but by transmitters in the pavement, federal law will mandate that personal vehicles transmit their location and dimensions to other nearby cars, etc.
Eventually once we have the infrastructure built, every new car will piggyback on it. Remember the Interstate Highway System was started in 1955 and completed in 1992 - progress just takes time.
But you can’t attach them to people like the homeless pedestrian killed by Uber’s test self driving car, animals, trash thrown on the roadway or falling off a truck, or a boulder falling from the side of a mountain.
And looking at the decades-long process it took to install PTC on train lines, it would take decades to install radio beacons on everything we can and by then technology should have advanced to the point where we don’t need them.
Say you're right and an ML model cannot know intrinsically the trash bag is safer to hit than to avoid, however, if a few car hit or avoid those for a little while they can learn of the consequence of both actions and then share them back to every other car.
I suppose as long as it can properly learn, it'll get there (and, more valuable: STAY there) eventually. The brain is an efficient machine that can work in highly complex network, but I dont know if the fact every brain is born mostly blank of experience is a good or bad thing to build much more complex specialized behaviour like driving with 0 death in an entire nation.
> Say you're right and an ML model cannot know intrinsically the trash bag is safer to hit than to avoid, however, if a few car hit or avoid those for a little while they can learn of the consequence of both actions and then share them back to every other car.
He is definitely, 100% right.
All this talk of networked AI/ML is a particularly insidious form of evangelism. Because it doesnt answer the questions that need asking.
Heck, I’ve seen kids mess with real cars. Like the bag you mentioned, some idiots were taking cardboard boxes, filling them with rocks and putting them in the street waiting for the drivers to think “oh just it’s an empty cardboard box, not gonna slow down for it”.
Somehow I can see an AI driven car, especially without a person inside, becoming a target of bullying like that.
If you tag each one with some sort of RF beacon, not only do you tell the car "this is the spot to stop" you also have an absolute fixed reference point and can correct GPS or other navigational systems.
At the point you know where everything is, and cars begin to share their location with each other, you have an absolute map of everything in the environment. If LIDAR detects anything that isn't part of the gold standard of what the world looks like, _that_ is an exception and handled as a danger (the person Uber should have detected had they had all the sensors turned on).
In most countries, every council does their own road markings and signage. Sure, there are laws, but they're implemented inconsistently. Worse, those laws are often ambiguous, intended for humans, and defined at the state level.
Ideally, all of the critical markings should have a single, consistent, internationally agreed standard. This doesn't have to be every road sign, but all of the really important lane markings and basic speed limits and stop signs should be consistent.
Then, these signs could have infrared 2D barcodes layered on top, which would guarantee machine readability.
Humans would benefit from the consistency, but wouldn't be able to even see the bar codes.
I've also seen vertical poles: https://cdn2.bigcommerce.com/n-arxsrf/vpk2ddje/products/119/...
I also regularly see these water-filled barriers, in various colours including red, white, blue, and yellow: https://www.bronsonsafety.com.au/media/catalog/product/cache...
I wouldn't be surprised if some countries have custom-made things built from bamboo or whatever local materials are available...
PDF page 20 - https://truckingresearch.org/wp-content/uploads/2018/10/ATRI...
The commercial trucking industry goes bonkers over a solution that cut costs by even .5%.
The cameras on the car can provide just as good an image as what a human eye can. The problem to be figured out is generating a mental model of the environment and reasoning about it.
Think "Brain in a vat" analogy https://en.wikipedia.org/wiki/Brain_in_a_vat
Yes they are. Although it's still not an easy problem. Occasionally I get sucked in to watching highway accident and road rage videos on youtube and try to consider which ones a ML could realistically handle. It's still a very complex random environment, even on interstates far away from cities.
I don't believe full self-driving has any chance of being a reality this century. It's far too complex a problem in an environment where all the edge cases do matter.
> annotate the world
This I do believe may start to happen in coming decades. But this is effectively an admission that actual self-driving isn't really possible, so the problem gets redefined to a smaller and easier scope.
Although even that will only work in limited access roads where nothing else can be present. Like barrier-separated HOV lanes but for these trucks.
If you stand or park on a railway track and get creamed, the pedestrian is going to be the target of blame.
We've come to the conclusion that the track belongs to the trains; at best the engineer can perform a limited, good-faith effort to stop and everyone knows that's not going to happen when you've got a kilometre-long, 10,000-tonne wad of physics rushing down slick metal rails at 60kph.
But if you walk into the middle of I-5 at rush hour, we still throw the legal and financial liability at drivers. Everyone is supposed to expect a rando suddenly stepping into the road and brake/steer around them. That's hard for humans and potentially harder for machines.
I could imagine a network of "autonomous only" roads which were much regulatorily much more like the rail infrastructure-- designed for minimum surprise (limited on/off ramps, limited turning points, predictable advance signing and marking, perhaps a block-control system to mitigate traffic flow) and limited recourse for pedestrians or human operators who enter the territory.
Maybe not something you could predict in a product with just a couple years of time, with just a couple employees, but we're talking over a decade, and thousands of employees. Oh and in Phoenix Arizona, one of those ultra uniform grid cities the US is famous for.
What's stopping them from having their AI operate in simulations in literally every possible traffic situation? Every location a pot hole can be in, every type of traffic cone arrangement, every kind of jaywalking, every kind of vehicle approaching any kind of street from any kind of location, possible speed, reaction time, everything? How could a government approve any autonomous agent that is not tested in such a way?
My world view gets rocked a little everytime I see a headline like this. I want to believe these companies are ran by the smartest engineers on the planet, but then I read about a Tesla running into an emergency services vehicle, or a Waymo being confused about traffic cones and I can't help but wonder what the heck are they doing?
This problem has to be solved end to end. It can be done with just vision as humans can drive with just vision. It's literally just a software, machine learning and data problem. Just takes time
Because this approach has huge limitations, principally that you’re now optimizing against a cost function that isn’t the real world. RL has a place for sure, but limitations.
It’s like saying that you’d expect someone who’s played 10k hours of Call of Duty to be equivalent to a Navy Seal. They’re just really good at playing the simulation, not the real thing.
Also simulations are definitely a part of military training, and if they perform badly in the simulation they'll have to fix that before going on a mission I'm sure. I don't know much about Navy Seals specifically, but I've got friends who built full body VR suits for special forces training.
> ...still getting stymied by traffic cones
Maybe you're nowhere done or near to that 99%, perhaps around 10% done considering that this doesn't work worldwide either so seems like 5% to me.
Still, though, how _do_ humans drive? One answer that comes to mind is "humans are smart." True. But I suspect a severely "dumb" person can drive safely in a variety of novel scenarios.
It makes me think of child development, all that time making sense of depth perception, distinguishing objects, fine motor controls, expectations of motion, etc. It takes years. Could a non-human animal trapped in a "car suit" drive effectively, if it could somehow overcome the panic? Maybe it could maneuver safely, but what about a four-way stop?
I have no idea what it takes to drive a car in general, and specifically I don't know how humans do it. Where do you even start when trying to model the task in a computer?