Aha, technological solutions for social problems, who could think that would not work :)
Driving a car in public traffic is very much a social situation. A wink here, a small hand or head gesture there. Even looking away and avoiding communication alltogether is part of social behaviour and tells something. Even the same gesture in a different context says something else.
Parking a car, driving in congested traffic and just keeping a lane is all doable by technology. Beyond that it needs a different and social way of thinking.
Even without seeing other people's faces, you start to notice car "body language". For example, a car that is sticking close to a lane divider is probably about to try to change lanes. I always wondered to what extent (if any) self-driving AI tries to take this into account and whether, even legal, maneuvers by self-driving AI might be surprising to other drivers looking for these kinds of subtle signals.
It's incorporated into autonomous vehicles in a few different ways (mainly through ML), but current state of the art isn't perfect. Humans are probably still better at reading those subtle patterns earlier.
Isn’t that what turn signals are for? Car body language is somewhat location-dependent. Rural Missouri has more cautious drivers and downtown NYC more aggressive drivers. Seems complicated to modify the car’s behavior at such a fine level, but perhaps it’ll happen once the bigger issues are worked out. On-board, personalized learning from your commute would be a neat selling point.
Even the enforcement of rules and the expectation on that are location dependent. Someone I know walked on a crosswalk in a southern european country, expecting that this will be honored like it is here. Ended up in hospital for two weeks. No driver there expected it.
AVs can seem to do that, but it's unintentional. I've never seen or heard anyone describe encoding intent with microbehaviors. The general preference is to make driving intentions as obvious as possible because some people don't pick up on subtleties and we still want to drive safely around them.
The vehicles are intended to be very predictable for other drivers. I don't consider that to be synonymous with encoding intent into subtle microbehaviors. Instead it means straightforward behaviors like turn signals, appropriate speeds, avoiding last minute lane changes, etc.
One of the big ways people predict what cars are going to do is by reading intent through both macro- and micro-behaviors. Reading motion intent through subtle motion indicators is not just part of driving; it's something that happens all the time, and not just for humans. Automated vehicles are entering into an environment where that's the case, and that will not change.
It really feels like it's driving just like a human would in so many situations, and understanding the intention of pedestrians and cars driven by humans.
Seems to me that they may be well ahead of Tesla, they just don't do much publicity or push it out to end users before it's actually ready.
The fact that driving involves social behavior is exactly the reason why 35.000 people die each DAY in traffic. If a driver is not able to pay attention or makes a mistake, consequences can be fatal.
We need to get rid of the social aspect of driving, and make it a pure, reliable system that doesn't depend on the mood or the alertness of billions of humans to not kill someone.
That might only be done if you make it separate from people who are walking or cycling. So maybe a lifted monorail with cars? :) As long as people are on the street, it is a social situation with social behaviour.
Honestly drivers who stop in traffic or hold up intersections to make hand gestures and yield their right of way should be shunned and ticketed.
This is a social norm of driving today that makes little sense. It’s actually quite dangerous to stop in the middle of traffic and wave at someone. And the person waving usually acts in an impatient and unpredictable way. They will give you about 5 seconds to lurch into the road before they get irritated and drive on. Or they could easily commit insurance fraud by luring you into an accident that appears to be your fault.
If all followed the rules of the road it’d be a lot safer and also slightly easier to automate
I think you are overestimating how clear the rules of the road are. For example two people arrive at a 4 way stop. The rules dictate whoever arrived goes first. Arrived at the same time? Yield right. Easy right? But are you sure the guy in the other car agrees you arrived first and therefore has the right of way? Or in their estimation did you arrive at the same time and you should yield to them? What if 3 cars arrive at the same time? 4?
Often it can be safer to establish some form of consensus instead of assuming the cars around you have both interpreted events the same as you and are about to behave in the way you'd predict.
I'm also not sure if this was what you intended in your example, but there are roads in my town where unless someone waits and yields to let someone on a side street turn onto or cross the main road, the person on the side street is literally going to be stuck until rush hour is over. You won't find it in the DMV handbook, but yielding in this situation really is a kindness.
That's a problem with 4 way stops. Equal priority four way stops are just bad road design and should be replaced. Similarly your side street situation is why traffic lights were invented. Alternatively, you could give priority to the side street and force cars on the main road to yield.
That being said, I do think standardising some form of "go ahead" message is worth considering due to the prevalence of these poor road designs, particularly in the US. In the UK, officially, you're not supposed to make any form of signal. If you must, you are supposed to just stop, but that can be ambiguous, so most people flash their lights. However, that is unambiguously against the highway code. If it was possibly to explicitly state, through a light or some other kind of means, "I am yielding my priority", that could be unambiguously understood by both AI and humans alike...
That being said, these situations where you need to rely on messages are actually reasonably rare on the UK roads and usually as a result of roadworks messing something up.
> there are roads in my town where unless someone waits and yields to let someone on a side street turn onto or cross the main road, the person on the side street is literally going to be stuck until rush hour is over.
It’s my view that this is a failure in road design, and leaving it up to drivers to make up their own rules is dangerous.
Understood this is the status quo, but I wish more would be done to address these types of situations because they cause stress and accidents
I'm not sure what you do when you have a busy main street and lightly trafficked side streets (and driveways) that require people to make unprotected lefts. You're not going to put a traffic light on every country lane dead ending at busy road.
>but there are roads in my town where unless someone waits and yields
Making a left hand turn into my driveway at certain times of day can take forever to do safely with cars either piling up behind me or somewhat dangerously passing on the shoulder. Even worse it's at the bottom of a hill so traffic tends to speed up. Sometimes there's a break but often someone will flash their lights and let me go.
Actually I think the "zipper rule" is taught in driving school in some countries in Europe and applied informally in more.
[I mean: when both the main road and the secondary road that has to yield are clogged, everyone on the main road lets one car from the secondary pass.]
I've certainly applied it and benefited from it countless times.
We have a slightly different problem: people don't understand that the full capacity of the road should be used, zipping only shortly before the constriction. People switch lanes far too early, other people pass them (because the lane is free until the constriction), people who switched early feel cheated, everybody gets aggressive.
The thing is about the zipper rule is that it can genuinely be difficult in practice sometimes. If it's a simple example of a busy slow moving two lane road becoming a one lane road, it's absolutely simple and unquestionably the best practice to merge late. But in other situations, especially ones involving exits and junctions, it can be down to judgement.
In very low density traffic on high speed roads, waiting until the lane is actually closed to move over can create slow downs, but moving left earlier doesn't. The point where the optimum behaviour changes isn't always clear and can change rapidly.
In the UK, the rule is that in slow moving congestion you should late merge but in other cases you should stick in the left lane unless overtaking. There isn't a specifically defined point where one behaviour should take precedence over the other.
This occurs in the US too. It drives me crazy when a two-lane highway turns into a one-lane highway for miles, despite a free lane leading up to the accident.
And any attempts to use the free lane will result in cars dangerously cutting you off.
Incidentally, those are often the Navies who lose ships due to getting rammed by a cruise ship [1], allied vessels [2] and shipping boats [3]. Might may make right, but can be expensive.
It’s useful but its importance seems exaggerated because often you can’t see the driver at all. For example, at night or in bad weather. The driver of a truck or bus can often be hard to see. Yet we can still drive if we can’t see other drivers.
I'd rather have an AI that can cook meals. Because that would actually save me time (I have to spend time in the car anyway, and there's not much I can do than stare out of the window, as I would get car-sick otherwise).
Shouldn't it be "pulling out", rather than "putting out"? Also, they already pulled out in October, this article is about their efforts to retain the developers that were working on the AI solution. And it's not "(all) german car makers", just VW and Ford (Ford apparently means Ford as a whole, not just Ford Germany?) - no idea how the self-driving plans of the others are doing...
Five years ago, car companies were saying around the world that in five years, their cars would be 100% self driving. I said it was bonkers, unless they create a new infrastructure for self driving cars (new roads that can be used only by these types of cars).
Google has been running self driving cars on public roads since 2010. There were driver interventions back then, but I doubt they would have any trouble running those cars in a warehouse after driving 7 years on public roads.
That tech probably wouldn't be economical to use in a warehouse though.
Driver interventions, on a few streets, with highly instrumented cars. But compared to public roads warehouses are easy. There isn't fog or rain or snow. There aren't potholes. The light doesn't change. And to prove my point: look at all these self-driving cars on the streets today.
It's a peculiar case of Bay Area Tech brain. "We very rarely kill pedestrians in SF and Arizona. I'm sure this tech will work fine in Ohio."
There are fully self driving cars on the road today. They don’t have even close to global coverage, but empty cars are driving on public streets.
Extending these limited trials to every road in the world isn’t going to be fast, but a few cities with zero human taxi drivers could happen surprisingly quickly.
Right, but those are fragile. Geofenced, unable to drive in inclement weather, speed capped, etc. To get to 100% self driving, you need a lot more than what we’ve seen from, say, Cruise.
Much of that is simply how conservative GM has been with their testing. Extending geofencing just comes down to cost, but it’s pointless to map most locations when they aren’t looking to start selling any time soon.
The technology can also handle speed and inclement weather at least as well a people do, but they have little reason to push when they have little to gain and a lot to lose.
And today, I have an app on my phone which will summon a self-driving car to pick me up and drop me off (within very strict parameters in space and time).
I also said this, self-driving will only become a thing if the roads are modified. Then I realised that this has already happened. The changes made to urban environments when cars first became a thing aren't to be underestimated. Curbs are a result of cars. Traffic lights. Pedestrian lights. Footpaths.
Back in the day, no one even considered these changes to be bad. Today if politican came up with the idea that we need to remove pedestrians (and bikes) from the streets so that self-driving cars could be safe, there would be an uproar.
> Today if politican came up with the idea that we need to remove pedestrians (and bikes) from the streets so that self-driving cars could be safe, there would be an uproar.
Rightly so, telling people they can't enter their own neighbourhood unless they have a $50k+ car is beyond awful. I think it would be better to remove cars from towns and cities.
> So they finally figured out that "self driving" was really just a divisionary/marketing tactic with lots of legal liability?
> When a "self driving" car kills someone, guess who shares legal responsibility?
When a rollercoaster crashes who takes legal responsibility?
(Answer in UK law: The rollercoaster owners are responsible for negligence if they do not maintain the rollercoaster adequately or follow safety rules. The rollercoaster's builders are responsible for negligence if there was a design fault or if the rollercoaster was fundamentally unsafe. The rollercoaster operators can be responsible if they do not follow operating procedures which they would be reasonably able and could be reasonably expected to follow and have been trained in).
Similar ideas have been established in self driving cars already with Mercedes taking legal responsibility in Germany if it's cars crash while in Level 3 driving mode (assuming operators are following the safety rules, for instance making sure that they are sober and able to take over driving if required).
When a rollercoaster crashes who takes legal responsibility?
Typically not the manufacturer. Because a roller coaster operates in a fixed and controlled environment within which it's design can usually be shown to be inherently safe.
The same can not be said for a "self driving" auto.
Given the current state of technology, one could make a convincing legal argument in many jurisdictions that just marketing an auto as "self driving" is itself a deceptive and inherently negligent act.
I don't buy the idea that in a world where self-driving cars are safer, we should have 'less safe' human drivers just because we can blame them when they kill someone.
Car manufacturers shouldn't have to show something is perfectly safe, and IMO they should just have to show that something is reasonably safe (e.g. as safe or safer than a human driver in the same condition).
There isn't a requirement that someone has to take legal responsibility for every accident that happens in the world, just that people act with reasonable care and attention within the law.
and IMO they should just have to show that something is reasonably safe
As it currently exists, it would probably be much easier to show that "self driving" is reasonably unsafe --- e.g. less safe than the average human driver.
> So they finally figured out that "self driving" was really just a divisionary tactic with lots of legal liability?
I don't think people get how utterly behind and unsafe the legacy carmakers products are.
I've rented current-year Cadillacs, Audis, BMWs, and it's all a terrible gimmick. They can't keep the vehicle in the center of the lane, "adaptive cruise control" repeatedly tries to kill you by accelerating into semi-trucks and trailers, the autonomous systems repeatedly disengage, and they certainly can't navigate around on city streets, can't change lanes, basically need 100% of your attention to operate in a remotely safe way.
Tesla may have their QC issues and their Elon issues, but the technology is a decade ahead. You can punch in an address and FSD Beta literally navigates you through the city with pedestrians, traffic, and so on to your destination in a fairly predictable manner. The adaptive cruise control, lane changing, and highway functions are predictable and reliable enough to handle a regular commute. It's easy to stand on the sidelines and be a critic but if you go out into the real world and actually drive all these vehicles it's obvious who is innovating and who is desperately playing catch-up.
My experience here is the opposite. I have an Audi A8 and my girlfriend drives a Tesla Model 3. In my experience, the Audi is way, way more reliable when it comes to the features. The features just work for me.
I've never had any problems with the lane assist, for example, and I drive between 1000km and 1500km a week through Germany (not sure if the features are different depending on the location). I'd like if I didn't have to put my hands on the steering wheel that often but that's pretty much it.
Tesla's offerings are a real letdown. And they don't work as advertised either, which is why they're not allowed to advertise them in the same way anymore. It's missing way, way too many things that the car should notice. It's kind of like Siri. When it works, it's amazing. It often doesn't so you can't really rely on it.
We also had to replace a rim on the right hand side because the car simply didn't turn the wheel back. Girlfriend won't use the feature anymore now because she's kind of scared.
The summons feature doesn't work reliably either even when the conditions are perfect. The car just stops and you stand there and don't know why it's not doing anything at all.
I can't speak for BMW or Cadillac but calling FSD a decade ahead is ridiculous. It fails at so many simple things and all videos online confirm this and they don't seem to be able to really fix this. Every new Beta shows the exact same problems again and again.
Like I said; the features don't work reliably. I've driven thousands of KM in a Model 3 and the features are impressive but not reliable. I'd go for less impressive but more reliable every single day.
FSD as a product is a gimmick. It is an advanced automation but nevertheless a gimmick because driving and sharing the road with humans requires capabilities that synthetic AI has not yet achieved, and there is no clear roadmap on when and how it's going to be achieved. The way FSD is marketed is borderline misrepresentation.
> FSD as a product is a gimmick.
> The way FSD is marketed is borderline misrepresentation.
Only one of these two statements is correct.
FSD is a gross misrepresentation of the function’s capabilities.
FSD is not, however, a gimmick. If it was simply renamed “advanced cruise control”, it would be very accurate, and Tesla’s product would be at the high end of quality for advanced cruise control.
No. It's not. Legacy automakers are decades of ahead of Tesla in measuring/controlling quality and reliability as part of their long-term cost structure. Tesla is a VC hack that wasn't even founded by Musk.
Insurance companies exist, so it's clearly possible for a corporation to take responsibility for liabilities arising from car crashes, and collect enough money from car owners to cover those liabilities.
Surely that is true. The legal landscape will have to be settled before self-driving becomes common.
E.g. should a self-driving car leave the road to avoid sudden accident? What if it then hits a pedestrian?
One advantage to software-driven decision making is, we can make a rule and they will all follow it. I.e. the law can decide no self-driving car should leave the road.
It's logical that the manufacturer would bear the liability--assuming the car has been properly maintained, software is current, etc. The driver (passenger really) doesn't have any real agency while the vehicle is driving itself.
It's also a sort of unusual situation in that we can reasonably expect an AI that is a far safer driver than humans will still kill people. That's not a normal expectation for products sold to consumers and used/maintained as directed.
> One advantage to software-driven decision making is, we can make a rule and they will all follow it. I.e. the law can decide no self-driving car should leave the road.
Rigid adherence to a law like that is a pretty major disadvantage to saving lives though...
It's one of those things where it's not ideal, but it's the only way to prevent legal liability. Basically if the AI acts in accordance to the law governing such behavior then the manufacturers are not at fault. Else you'll have different vehicle manufacturers doing what they believe is right (think of all the debate over the "trolley" problem), and each case will have to go through the courts.
There are all sorts of subtle ways you need to "break the law" while driving though. Right now, one sees tons of delivery vehicles pulled over to make deliveries and you have to carefully go over the double yellow line in the middle to get around.
Well, they could try not to leave the road. Just like humans the software can and will have errors. Then you still have the issue of who to charge for breaking the leaving the road law.
The fly in the self-driving ointment is accidents are NOT 100% unavoidable for humans assisted by any technology.
People can jump in front of cars or intentionally interfere with said tech, for example.
There will be accidents, and the AI systems can predict with great certainty that a crash will occur before the human at the wheel does. In Tesla's case, it disingages before the time of impact to mitigate Tesla's culpability. (seemingly)
In related news, I'd like to announce that I am ending my training for the 2024 Olympics. That car companies are giving up on this doesn't change the story at all because they were never real competitors.
It never made any sense that manufactures of cars thought they could make self driving systems. Experience with supply chains, manufacturing, and marketing have zero translation to the problem space of self driving: sensing, interpreting, and decision making.
It's kinda funny that we'll try to invent advanced ai that can perform a near miracle of technological advancement before investing in proper public transit.
As an American that works in the industry and advocates for improvements in public transit, I suspect the near-miracle of technological advancement is faster and cheaper than funding new public transit. Look at transit projects in the bay area for example. The 5 station BART extension to Santa Clara (planned since 1981!) will cost somewhere north of $9B and be complete by 2030 if everything goes to plan. The proposed Livermore extension would have cost $1B+.
Compare that with autonomous vehicles. Several billion has been spent on development since 2010, but test fleets are already deployed in multiple cities globally. They also don't have to deal with the excruciating approvals process for further deployment that new public infrastructure has to, so there's a reasonable chance they'll actually exist before the heat death of the universe.
Would fixing the entire political system be great? Sure, but I'm tired of getting screamed at in public meetings.
My comment is mostly because I believe self driving will not solve the mobility problem at all. "in the future, all traffic congestions will be of self driving electric cars" is a way of thinking it.
It's a decent last-mile solution rather than a replacement for things like trains. Think about the Livermore BART extension I mentioned that will never be built. That route is currently served by a 40m bus that runs twice an hour and sees approx. 860 riders / day (latest numbers I could find, covid ridership) across its 27 stops. Most of those buses are functionally empty.
An autonomous vehicle could do the same route in 15m (so people might actually use it) and it wouldn't be empty.
I don't mind driving for the most part and there are situations where I even enjoy it. But the congested drive into the 1 hr+ away city for an evening event or long boring highway drives? (And certainly a commute if I did one?) If there were an AI chauffeur I could engage for a price not that much elevated over driving myself? Sign me up. Not that I expect to see that anytime soon.
This is a solved problem. You can buy cars that will drive themselves on the highway and in traffic jams. The other situations are the unsolved problem.
I’m not sure where you’re based but I hate driving and so do many of my friends. I’m probably in the middle of the average age of a HN reader, and I’ve lived in huge cities and small towns (mostly Europe, never the US, which might be e factor here).
Give me more trains and then self driving cars only for the last few kilometres of the route.
As someone who has driven the equivalent of 23x around the equator I have to tell you, as much as I like driving, you see a lot more from the passenger seat. So if I had an AI chauffeur tomorrow I’d happily jump into the passenger seat and have it drive me down historic Route 66, then onward to Wyoming and Montana, and maybe the trans-Canada highway from there.
Unfortunately. I guess it will take 50-100 years more for necessary changes in society values/worldview to realize that it's really not a good idea. Just like it happened with smoking.
I, for one, would love a self-driving rv. I can see it now, first taking a shower, then sitting back on the sofa, drinking a cup of coffee while silently gliding to the office in the morning.
I want a self driving camper van not for the office commute but for travel. Imagine being able to leave after work on a Friday night, chill on the sofa with a few drinks, sleep and wake up parked anywhere within 14 hours range.
Then on sunday night you simply go to bed, wake up, have the shower and coffee and boom you're back at the office. Or if you work remote you could potentially work while the van drives you places.
I want the freedom to take a short trip to the store or a long trip to visit friends and family, without waiting to leave. I'm not a big fan of the actual driving part, but there's no way to avoid it without depending on another human being and their car.
That is what a functional transit system does. In Japan, train headways can be 5 min. You get to the station, wait a minute, hop onto a train, hop off, walk a minute, and you're there. The same amount of time it takes for you to put your seatbelt on, check the dash, and park. Transportation on demand.
Depends on what kind of want to drive actually. I doubt anyone wants to be stuck in traffic every day because they have to get to work / are out of groceries etc.
Now going on a long weekend and having fun on a winding mountain road is another matter...
Clearly not true in the US, where driving is (at best) a distant 2nd to the most popular activity while behind the steering wheel: looking at your phone.
I'd be very interested to see some data here. What's "plenty"?
But the big difference between the 1990s and now is that everybody already has perfectly good text chat. Plus free voice calls and video calls. Plus regular social media for non-synchronous interaction.
I can believe VR Chat is better for some niche of people. But I don't see much reason to think it's going to be enough better than all the other options that people will don special gear for it.
Someone would sure have to convince me that the VR made things a lot better. I'm skeptical given that I rarely even use video for personal calls outside of some group Zoom one group of friends or other put together maybe every couple months.
Yeah, video is a great example. In the movie 2001, which was made in 1968, video calls were The Future just as much as space travel was. But now that we actually have video calls, we've discovered that they're good for some things, but people prefer lower-capability mediums for a lot of things.
In the case of 2001, it also made the assumption that the futuristic video call would be made from... a phone booth as opposed to a mobile device networked to the station's computer.
I suspect many of us would be more inclined to just use video if we were sitting in a video booth--or, as is actually the case, from a place we've arranged to made video call from using a computer. As it is though, when I'm casually chatting with someone, I'll often be multitasking with making dinner or I'll be pausing the video I was watching on the TV. I'm not really setup for video and they're probably in the same boat.
But we know what each other look like anyway. Video is often better for a group call for various reasons but for a 1:1 chat voice is fine.
I've been video chatting my friends since Skype was a P2P app in the mid-2000s, though I was definitely the more introverted/less video-happy friend in the bunch. The friends most interested in video chatting were usually not STEM folks. My Gen-Z baby sibling has been using FaceTime to chat with school friends since they were 12. Even as early as 2014, so very pre-pandemic, 15-20 million FaceTime calls were being made per day [1]. I think you're colored by a STEM bubble, since STEM folks tend to be more private and introverted. Video chatting is huge.
Fair enough. Honestly, probably not so much about being introverted. I'm actually a marketing person who travels to a lot of events these days :-) But when I'm casually talking 1:1 with someone I know personally, voice-only is just a lot more flexible. I don't mind video but it's more convenient to chat audio-only while also doing something else.
I think where 2001-style projections of the future differ is that they hold up video chatting to be some end-all, be-all goal of future communication. As you say, audio only is more convenient. I'm a dirty tech person and I prefer text even when most of my friends prefer audio. But unless I'm putting my software engineer hat on, I'm not really thinking about "how am I communicating today is it video, voice, or text" I'm just summing up my situation (at home, in public, doing groceries, etc) and finding out how I can communicate with a person. With video, voice, and text all instantly available, none of us really think of one form as paramount, though we all have our preferences. Just like in real life, we all communicate in each other with different ways appropriate to our moods and our settings.
100%. And this isn't a new thing. 25 years ago when I was a product manager, I used to hate the call from a sales rep or systems engineer that involved discussing the weather etc. when all they really cared about was the answer to a specific question which could have been a five minute email. (Not that there is anything wrong with some inefficiency if it helps establish connections.)
VR for communication may not be. VR for training will almost certainly be in use by millions of people on a daily basis in the very near future. Anything that requires operation of a machine or performance of a physical procedure and some level of risk/cost will probably involve VR.
https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...