Autonomous Car Progress

[diplomat33]

No we do not. (We have such a thing for partial driverless which are very carefully monitored at huge expense and it is not clear how it will turn out.)

Well I have to disagree. The proof that driverless tech exists today is that we have hundreds of cars that are driving around with no human driver. So clearly, the tech exists. Obviously, the tech is not perfect yet. Driverless cars do occasionally require remote assistance. But that does take way the fact that they are driverless.

And "partial driverless" is an oxymoron. A car is either driverless or it is not. Remote assistance is still driverless. Driverless simply means no human physically in the driver seat or remote controlling the car.

I do believe at some point cars will drive themselves. I used to think within 5 years from now (5-10 2019). Way way too optimistic.

It’s just such an incredibly difficult problem.

Well, I agree with you that autonomous driving is an incredibly difficult problem. First, we needed to build a perception stack for the AV to understand its environment. Then we needed to build a prediction and planning stack so the AV can reason about what other objects will do and plan its own actions. Those two things alone were very hard because they required a lot of data and complex machine learning and there were a lot of difficult problems to solve. And they required new machine learning in order for the stacks to actually get good. But now, I think the real challenge is dealing with those exceptions when the drivable space is changed or you need to read social cues. For example, we see AVs like Waymo and Cruise generally drive great under normal conditions but sometimes struggle in a construction zone or a road blocked by police or a downed power line. So it is in those cases when you have to read the context of the scene or follow human directions where AVs can struggle. But IMO, once those cases are solved, fully autonomous driving won't be far behind.

But, I guess I am more optimistic than you. I do think fully autonomous vehicles will be common in 20 years.

 

[Doggydogworld]

If they are all available to me I’d probably choose Uber over either of them when going to catch a flight.

Sure, if you leave late and are in a hurry. Lots of women love the idea of getting into a car that's empty car vs. an Uber/Lyft with a driver they never met. Especially at night. Who picks them up or drops them off at home and thus knows where they live. Then you've got minors. And introverts who'd rather be by themselves. And people who don't want to tip. And.....

"Most roads" will be open to human drivers in 20 years. Countless miles of rural roads, for example. But I can see cities making a bunch of roads off limits.

 

[AlanSubie4Life]

And "partial driverless" is an oxymoron.

Partial in this case means that it is available in a partial portion of a typically required ODD.

Cars are driverless in some places. Yeah, of course.

But being illegal in the US or even being viewed as anachronistic in 18-20 years? That is what I was amused by.

 

[diplomat33]

Partial means that it is available in a partial portion of a typically required ODD.

Thanks for the clarification. But I might nitpick with your term "typically required ODD". I guess you are saying that driverless needs to be available in a bigger ODD for you to consider it "full driverless". But driverless is the state of not having a human driver, it has nothing to do with the ODD. Moreover, we can debate what ODD we think a driverless car should have but the manufacturer gets to pick the ODD. They get to decide what a "typically required ODD" will be for their L4. In the case of Waymo or Cruise, they can designate whatever ODD they want for their ride-hailing service. And the L4 is fully driverless inside the ODD and not driverless outside the ODD. So it is not "partial driverless" inside the ODD.

 

[AlanSubie4Life]

But driverless is the state of not having a human driver, it has nothing to do with the ODD.

Sure, this is just semantics. I don’t care about that stuff, though of course it is super important (again, I just don’t care, and I was just clarifying what I meant).

It’s equally important in my mind to clarify that “fully driverless” cars per some definitions require drivers, in the general case.

 

[JHCCAZ]

First, I think it's very unlikely, and not a desirable outcome nor a sensible target, that driving conventional manual cars will become illegal on most roads in 20 years.

Second, I think that if you're an advocate of the technology (as I am), you would be making a very big mistake by telling people they will have to give up driving their cars in 20 years.

Third, I fully expect that the mistake mentioned above will not only be proposed, but will be the topic of another divisive political debate. I'll be surprised but gratified if this doesn't happen.

If it's allowed to develop along sensible lines, the first moves will be to designate AV lanes on major highways, similar to HOV lanes today but possibly with incrementally increasing physical separation - creating essentially parallel divided roads where you can see the smoother and higher speed AV lane. Over time, the proportion of the manual driving lanes will decrease until they basically become like today's frontage roads.

Dense inner cities will take a different path. There isn't as much space to permit coexistence and separation; also urban residents are more accustomed to regulatory restrictions. So I could see a trend of increasing AV/robotaxi and AV delivery vehicles in cities. There'll be an earlier onset of restrictions that push manual vehicles into a sparser and sparser network of drivable streets within the city, eventually becoming so restrictive that they mostly disappear. In some cities, we could see the move to a largely underground network of vehicle traffic, with walking streets and various kinds of people-mover technology coexisting on surface land. The details will vary widely and there is no one-size-fits-all.

The vast amount of suburban and rural routes will be the last to give over to AVs. Again, manual cars will eventually get pushed onto increasingly secondary and tertiary routes. If v2V/V2X technology develops (it's just been given a new encouragement regarding spectrum standards), I could see the availability and eventually a mandate for plug-in accessory modules that will bring some level of driver assistance/notification and control intervention to existing manually driven cars - the newer ones that have electromechanical steering and OBD/CANbus interfaces - thus allowing them to to participate more fully in the transition.

This is what could and will happen, if it's allowed to proceed in a more natural and locally managed fashion. If it's pushed too hard, centrally mandated and politically divisive, it will actually take longer and be a much bigger mess along the way.

 

[powertoold]

There's too much discussion about this unclear future.

Fact is, AV cars are not yet ubiquitous, convenient, reliable, user-friendly, cheap, and safe enough for wide adoption.

Once they get to that point, there are plenty of ways that manual cars will be phased out:

1) Make it expensive to own or maintain a manual car
2) Restrict manual car roads
3) Ban them from certain cities or times of day, etc.

People will definitely use AVs once they are easily accessible. For transportation, the population tends to gravitate towards the lowest priced and convenient forms of transportation. When I call my robotaxi, I want it to come on average < 5 minutes, be cheap, safe, and comfortable. That's all.

We can see from the airline industry that people will simply go for cheap, safe, and reliable, even if they're packed like sardines.

 

[AlanSubie4Life]

creating essentially parallel divided roads where you can see the smoother and higher speed AV lane.

* where you can see traffic come to a complete halt due to a scary paper bag in the road.

 

[JHCCAZ]

* where you can see traffic come to a complete halt due to a scary paper bag in the road.

Lol. Still I stand by what I wrote; I didn't fill it up with disclaimers about the AV technology needing to have better perception and decision making than it does today.

 

[diplomat33]

First, I think it's very unlikely, and not a desirable outcome nor a sensible target, that driving conventional manual cars will become illegal on most roads in 20 years.

Second, I think that if you're an advocate of the technology (as I am), you would be making a very big mistake by telling people they will have to give up driving their cars in 20 years.

Third, I fully expect that the mistake mentioned above will not only be proposed, but will be the topic of another divisive political debate. I'll be surprised but gratified if this doesn't happen.

If it's allowed to develop along sensible lines, the first moves will be to designate AV lanes on major highways, similar to HOV lanes today but possibly with incrementally increasing physical separation - creating essentially parallel divided roads where you can see the smoother and higher speed AV lane. Over time, the proportion of the manual driving lanes will decrease until they basically become like today's frontage roads.

Dense inner cities will take a different path. There isn't as much space to permit coexistence and separation; also urban residents are more accustomed to regulatory restrictions. So I could see a trend of increasing AV/robotaxi and AV delivery vehicles in cities. There'll be an earlier onset of restrictions that push manual vehicles into a sparser and sparser network of drivable streets within the city, eventually becoming so restrictive that they mostly disappear. In some cities, we could see the move to a largely underground network of vehicle traffic, with walking streets and various kinds of people-mover technology coexisting on surface land. The details will vary widely and there is no one-size-fits-all.

The vast amount of suburban and rural routes will be the last to give over to AVs. Again, manual cars will eventually get pushed onto increasingly secondary and tertiary routes. If v2V/V2X technology develops (it's just been given a new encouragement regarding spectrum standards), I could see the availability and eventually a mandate for plug-in accessory modules that will bring some level of driver assistance/notification and control intervention to existing manually driven cars - the newer ones that have electromechanical steering and OBD/CANbus interfaces - thus allowing them to to participate more fully in the transition.

This is what could and will happen, if it's allowed to proceed in a more natural and locally managed fashion. If it's pushed too hard, centrally mandated and politically divisive, it will actually take longer and be a much bigger mess along the way.

Thank you for the thoughtful response.

I would add that I think one reason there is talk about some day banning humans from driving is because human drivers are fundamentally bad at driving. We don't have 360 degree perception. We are not always fully attentive all the time. And we can drive in an impaired state (tired, distracted, sick, drowsy, drunk etc). Human drivers can also be impatient and reckless. And sometimes, human drivers just miscalculate and make bad decisions. For example, we think we can make the merge but the other car was going a bit faster than we thought or the other human driver makes a sudden change, resulting in a collision. AVs don't have any of those flaws. Before AVs, we had no choice since humans were the only ones who could drive. But when AVs are everywhere and they are proven to be far safer than humans, what possible reason would there be to still allow flawed human drivers? Why would we still allow humans to drive around and cause accidents (and kill people on the road) when we have a safer alternative?

 

[diplomat33]

https://twitter.com/x/status/1653094853146705921

 

[AlanSubie4Life]

I would add that I think one reason there is talk about some day banning humans from driving is because human drivers are fundamentally bad at driving.

Yes, this is a huge misconception which is relevant to the level of competence required for autonomy.
Humans are incredibly good at driving.

Exclude:
Drunkenness
Lack of attention
Incapacitation
Mechanical failure
Etc.

I am not sure, but people seem to make the mistake of thinking that the autonomous vehicle just has to be much much better than a drunk or inattentive driver and then somehow everything will be fine. This is very flawed thinking, for obvious reasons.

AVs don't have any of those flaws.

They don’t? I am so confused.

But when AVs are everywhere and they are proven to be far safer than humans, what possible reason would there be to still allow flawed human drivers?

If this ever happened and they were also equally efficient and fast as humans and broadly available and affordable and it was included as base equipment in personal vehicles, I think we would be potentially to that point. Though I think human driving would still be allowed (because why not? Most people would choose not to!). Few would care about a few unsafe drivers who can’t keep up with their AV.

 

[diplomat33]

Yes, this is a huge misconception which is relevant to the level of competence required for autonomy.
Humans are incredibly good at driving.

Exclude:
Drunkenness
Lack of attention
Mechanical failure

I guess it kind of depends how you look at human drivers. Certainly, a very attentive human driver will be very competent. And at the moment, human intelligence and intuition can figure out certain difficult cases better than an AV. But I don't think we can just ignore all the bad driving from humans. We have to look at all human drivers, the good and the bad. You can't deny that human drivers do have some serious flaws as I mentioned above: no 360 degree perception, not always attentive, emotions that can interfere with driving, impaired driving, reckless and poor decision-making. AVs have none of these flaws.

And regardless how good a competent human driver might be, the whole point is to reduce all those accidents that are caused by drunkenness, lack of attention etc... There are a lot of them. And AVs have the potential to reduce those accidents. Now if you are suggesting we keep all the good human drivers and only replace the bad drivers with AVs, maybe that could work, to get the best of both worlds.

A lot of people make the mistake of thinking that the autonomous vehicle just has to be much much better than a drunk or inattentive driver and then somehow everything will be fine. This is very flawed thinking, for obvious reasons.

I am certainly not suggesting that. AVs have to be as good as the best human driver, better even. In fact, that is precisely why Waymo created their NIEON benchmark. NIEON stands for "Non-Impaired, Eyes always ON". Basically, Waymo measures their AVs collision avoidance responses to a simulated human driver that is fully attentive and non-impaired.

Waypoint - The official Waymo blog: Benchmarking AV Safety: Demonstrating how the Waymo Driver outperforms the collision responses of always attentive human drivers via industry-leading assessment methods

 

[AlanSubie4Life]

And AVs have the potential to reduce those accidents.

Yes they do. (And of course they must also not increase accident rates in the other cases, which comprise a much much larger portion of driving miles!)

The Waymo benchmark is the important one, in a broad, nearly fully inclusive ODD. (Though after reading the first few paragraphs of their blog I have no idea how exactly they came up with the metric, so I am of course naturally suspicious. )

“The goal of this study was to examine a novel method for how a reference behavior model benchmark can be used to evaluate the performance of an autonomous driving system; not to establish exact specifications for it. “

So yeah there is currently no way to figure out the benchmark, it seems. Unsurprising of course.

 

[DanCar]

I wonder how Cruise will survive 2023 by spending $2b a year.

Waymo is close to $4b a year. They fund spending by selling share in the company. Stock value increasing has helped in the past.

 

[diplomat33]

I would further add that human drivers and AVs have different failure modes. Humans are very good drivers in ways that AVs are not. And AVs are very good drivers in ways that humans are not. As I mentioned above, human intelligence and intuition is better than AVs. So humans tend to be better drivers than AVs in complex situations, especially cases that require intuition or reading social cues. AVs tend to be better than humans in those mundane driving cases where humans might get tired or distracted. This is one reason why some, like Mobileye, argue for advanced driver assist + driver monitoring as a way to get the best of both worlds: you have an advanced driver assist that will not get tired or distracted but you also have an attentive driver that can intervene if the driver assist does not know how to handle a complex situation.

 

[powertoold]

So humans tend to be better drivers than AVs in complex situations, especially cases that require intuition or reading social cues.

Almost none of these "complex" situations are related to safety though.

At the end of the day, if the AV drives much safer than a human, and has all the right conveniences, then it doesn't matter if it can't see that the person at the crosswalk is waving it to proceed. Who cares if 5 seconds are wasted by a rare situation of an AV not understanding inefficient human behaviors.

Here's a prime example:

 

[ThomasD]

How many years will it be for AVs to be operating in every city in every state? How long before AV are allowed to transport people on and off Military Installations? Will the AVs be wifi hotspots so they can operate in areas where there are no cell signals? How many years are we from level 5 AVs? When will an Autonomous Van be able to pick up the Amish people and take them to Dr appointments and to the Grocery Store? We can't get Cities and States to rebuild their infrastructure let alone creating separate lanes and roads for AVs.

 

[kabin]

AVs don't have any of those flaws. Before AVs, we had no choice since humans were the only ones who could drive. But when AVs are everywhere and they are proven to be far safer than humans, what possible reason would there be to still allow flawed human drivers? Why would we still allow humans to drive around and cause accidents (and kill people on the road) when we have a safer alternative?

Until it becomes commoditized there will be differences between various manufacturer AV system performance, cost, reliability, driver etiquette, so some might not want to use them. Plus I'm not sure how to solve the problem of AVs being submissive to more assertive drivers wanting to get from point A to B faster than the next car.

 

[JHCCAZ]

...But when AVs are everywhere and they are proven to be far safer than humans, what possible reason would there be to still allow flawed human drivers? Why would we still allow humans to drive around and cause accidents (and kill people on the road) when we have a safer alternative?

Eventually this will be the state of affairs, just like today we don't allow horses and carriages (nor bicycles) to drive down themain lanes on the interstate highway. But we do allow them to use the roads in less disruptive ways. When this all started, horse traffic on to the routes and automobiles were dangerous, offensive and annoying. A bit later they became common, still dangerous and threatening to an established way of life - but they enabled mobility, vast economic growth and increased personal freedom.

The parallels are not perfect but they are apt enough to make us think. In my opinion, the best trajectory for a transition will minimize the amount of banning - and maximize freedom of choice within a sensible framework. This principle is really at the core of successful civilizations in a very general sense. The default should be personal and economic freedom, with justifiable and largely agreed-upon limits where necessary. Obviously this is easier said than done, but I would say you should ask not "Why should we allow it?", but "Why must we disallow it, and is there another way?"

Even if one doesn't appreciate like this kind of generalized social and political philosophizing, it's still an important point that the tactics matter when promoting and managing a big transformation. People don't like to to have their customary behavior banned, or threatened to be banned. (And even worse is trying to mollify them with condescending explanations and tricky vocabulary redefinitions.)

My advice is to allow the deployment of the new technology even if it's imperfect. Don't disallow AVs and don't disallow conventional cars. Manage the infrastructure development. People will be annoyed at the new technology initially ("Get a horse!"), will adopt it in increasing numbers, and eventually will see it as normal and expected.