[Wronnay]

I don't agree with the categorization of “full self-driving” as "LEVEL 2: EXAGGERATED CLAIMS".

There is no proof that self driving cars will ever be able to navigate in e.g. European cities with narrow streets and without lines on them.

Full driving would mean driving autonomous everywhere and between a highway with visible signs and lines and streets without any markings is a big difference.

its more like "LEVEL 3: UTOPIAN FUTURES" because there is the potential for it but no current systems can handle difficult situations in streets without clear markings.

[Retric]

> There is no proof

That’s an odd requirement. There is no proof Intel will ever put out a faster chip. There is in general no proof for any new technology because it hasn’t been built yet.

Narrow city streets aren’t vastly different than normal city streets, and people used to constantly say sure operating on highways is easy, show me self driving cars inside cities. I think people underestimate how long self driving cars are going to take, but they are already good enough to be useful which means companies will continually invest in being just a little bit better.

Looking back automatic transmissions took decades and where seriously flawed for most of that time, but just a little bit better every year and eventually they became the default. Cruse control and every other major innovation went through that same cycle, it’s just that once something is good enough to be the default it has been around so long nobody wakes up and says “wow it’s finally ready” because it’s been in use for a long time.

IMO, take everything people are generally expecting self driving to be in 2030 and it’s going to be ready in 2060 and the default by 2090.

[Animats]

Here's WeRide testing in an urban village area of China.[1] The system is being very cautious, but is doing fine.

Here's Waymo, dealing with San Francisco streets.[2] Right now, Waymo is offering some rides in SF on a trial basis. Fully autonomous. No safety driver. 29 cameras, some radars, multiple LIDAR units. Full hemisphere coverage for each sensor.

It's getting steadily closer.

[1] https://twitter.com/i/status/1400772458991407113

[2] https://www.youtube.com/watch?v=2CVInKMz9cA

[Wronnay]

I agree with your comment, but I understood the author's level 2 as something which can be realized with existing technology. Meaning there is already a prototype.

There is no prototype for complete full driving, including very difficult situations - so I see that as level 3.

[Retric]

Level 2 says Tomorrow as in technology that’s in development. I read Level 3 as when people talk about the potential for say intersections without traffic lights and the end to gridlock. That’s not just exaggeration or extrapolation of current research but hopeful projections without anything even vaguely in the pipeline for development. Aka the semantic web and other rather grandiose ideas.

The first generation of cruse control didn’t shift gears making it useless for maintaining speed up and down steep hills but it provided the basic functionality on a reasonable subset of roads. It was still called cruse control. So, my point was a Tesla’s ability to drive from a limited set of parking spaces to another set of parking spaces by crossing public streets, covers the basic description of self driving. It’s a poor implementation but fulfills the basic premise of you not having to turn the steering wheel and ending up your destination even if it has a host of limitations. Doing that but in Paris is really more of the same.

(Edited the above several times.)

[danielmarkbruce]

How would one build a prototype for "complete full driving" ?

[inglor_cz]

Narrow streets can be equipped with some kind of helping beacons one day, much like they were equipped with street lamps and canalization in the previous centuries.

Anyway, this is a fairly low percentage of the total street network. Even in Prague the medieval center is smallish (and many of those streets are off-limits to regular car traffic).

[boopboopbadoop]

> Narrow streets can be equipped with some kind of helping beacons one day

That changes the goal posts in my opinion. The “self” part of “self-driving” no longer applies.

[hiptobecubic]

I mean, many people can't drive without safety features all over their cars, GPS to help them navigate, and road signs to tell them where to go.

I don't see how "helping beacons" for robots are unfair but street lights are not.

[boopboopbadoop]

Maybe I’m drawing conclusions about what these beacons would do, but if any of the things you mentioned stop working, I can still drive and won’t go full speed into a wall. I don’t have the same confidence in these beacons.

[hiptobecubic]

Why would a car that is having trouble orienting itself go full speed?

[boopboopbadoop]

Software bug? I think you’re dodging the point a bit - going into a wall at any speed is bad.

Alternatively, say it has trouble orienting itself and slows down. That can be just as dangerous if cars behind you are going fast (e.g. a highway).

[nexuist]

Does it matter? If investing in smart infrastructure is what it takes to recognize this reality, who is this a downside for? It's like saying we can't build any more intersections because we'd have to buy traffic lights.

[boopboopbadoop]

I’m criticizing the “self-driving” term, which seems to be having more and more asterisks applied to it. What’s next, the cars will need tracks?

[hiptobecubic]

Honestly, small streets where there's really only one thing you can do (go straight forward but don't drive into anything) are not as hard as where cars are already driving today because the speeds are very low and the distances you have to consider are are pretty short. This is not like making unprotected lefts across 6 lanes of traffic in Beijing.

The hype around "full self driving" is exaggerated because today it's not even "limited self driving," let alone SOTA.

[Wronnay]

Did you ever drove through e.g. city centers in Europe or Asia?

There are not only missing lines but it's often also used by many people which walk on it. A self driving system doesn't only have to figure out where the steet ends and walkways begins but also has to correctly detect many moving people and animals.

Additionally, there are frequently side streets or walkways which are not mapped in every map database...

[hiptobecubic]

Yes. My point is that detecting people is not hard and everyone but Tesla already seems very good at it. What's hard is understanding what they want to do and reacting to it. That part is much easier if everything is going 10 mph or less.

Detecting inconspicuous transition between sidewalk and road is difficult, but if you haven't noticed, human drivers don't seem to care much about it either.