Hacker News Reader

Comment by mattlutze | Hacker News Reader

There's a fantastic amount of value in cities and urban areas. Self-driving cars can work as a massive network to reduce or eliminate congestion and give people back hundreds of millions of hours spent stuck in traffic.

https://edition.cnn.com/2019/08/22/us/traffic-commute-gridlo...

They can also, if Level 5 becomes a thing, reduce traffic collisions. And if they all end up electric, through route optimization they can significantly reduce energy consumption requirements for vehicle usage.

View on HN

How would they "reduce or eliminate congestion"? Congestion is inherently caused by too many people trying to go to or through the same place at the same time. That's not possible without reducing the amount of people trying to travel during a given time period in a given place.

There's lots of reasons for congestion existing.

https://www.geotab.com/blog/traffic-congestion/

Given the same number of cars on the road, a self-driving fleet would eliminate many mechanical failures and many human failures, and be able to adapt to sub-optimal infrastructure at a network-wide level.

Infrastructure would also be easier or more efficient to improve, because you'd have removed much of the human variability that makes identifying choke points difficult.

Freeways can handle much more traffic, but people get jumpy beyond a certain density and cause traffic jams "for no reason"

The problem is that in many cases, many people want to get off at the same freeway exits, and the local network doesn't have the road capacity to match.

AI doesn't really have the power to change that, and might actually be worse depending on how it reacts to pedestrians and cyclists.

By offering laminar rather than turbulent flow. Take a two lane highway where traffic moves at 100 kph. It contracts to one lane at 100kph. Humans will pull up close, stall in one lane, do a flaky and awful zipper merge. Latency will be higher than necessary and throughout lower.

Self driving cars can plan the whole exchange, negotiate (or be directed by an intersection controller) to decelerate, align, merge, and maintain high throughout at low latency. ATC does this for planes already; we have the coordination technology. It’s just about getting drivers to listen to precise instructions without deviation. For everywhere I know of, that’s going to take automation.

Reading up on standing waves taught me a lot; you might like adding that wrinkle to your model.

> By offering laminar rather than turbulent flow.

That's only going to work on roads that are AI-only.

There are already cities where buying a car is difficult and/or expensive because of traffic concerns. Singapore, Beijing, Shanghai, Tokyo, ...

I don’t see an AI requirement to be a huge hurdle for those cities.

There are other actors on the road, namely pedestrians and cyclists. Unless we're going to remove the right of citizens to the street at all, they will be present and will need to be accommodated for.

Computers can or will deal with the other actors as well as humans do now, which is quite a low bar anyways.

My point is that unless you get rid of them completely, traffic flow is not going to increase. You can't have free-flowing traffic everywhere and also have pedestrians safely crossing the street, and the lower bound on a traffic light cycle today is the amount of time it takes an old person to cross the street from curb to curb.

That neglects limited access roads and wide boulevards. In either case, even if pedestrians aren’t banned, they are rare enough to be an exception rather than a rule. For example, the old guy trying to make their way across 8-lane Chang-an Road in Beijing...well, there is a reason Chang an road forces pedestrians onto bridges.

By and large, today the congestion problem is not really on the massive roads themselves, but where they interface with low-capacity local networks. Most destinations are not located on a limited access road by definition, and the busy exit into city center with lots of pedestrians is not going to get any less congested, AVs or not.

In modern Western urban planning, large car-capacity streets aren't considered a good thing to be running through urban areas anyways. Most of the large cities (Paris, Berlin, New York) are taking steps to reduce road widths to more equally distribute the road between pedestrians, cyclists and transit users. As it turns out, building pedestrian overpasses at a convenient enough spacing for pedestrians is extremely expensive.

It's hard for me to guess how much something like this would reduce delays caused by traffic compared to the increased number of cars driving due to the fact that some driverless cars would be out with no passengers heading to pick people up.

Self-driving cars cannot plan an entire exchange in a dense urban areas because there will always be actors like pedestrians, cyclists, and stray plastic bags that do not participate in planning and act adversarially to their shared model. Optimizations will be highly local, spatially and temporally, and I suspect they will end up looking a lot like humans trying to coordinate on the same problem. And even if AVs can technically plan and execute faster, their actions will need to be artificially slowed to be legible to humans. Likewise with their raw speed — cars, self-driving or not, are already moving too fast in urban areas. Reaction times might improve but braking distances will not.

So if AVs can't increase the throughput of city streets, I'm skeptical that they can increase the throughput of off-ramps which are bounded by city streets, or urban freeway segments which are bounded by the off-ramps. And even imagining that significant (2x?) throughput is achieved, it's not going to meet the induced demand ceiling; there would be the same amount of congestion, only with more cars.

L5 is dead on arrival as congestion-mitigation technology and I hope at least some of the billions earmarked to be spent on researching and deploying it are redirected towards better walking, cycling, and transit amenities instead.

Self-driving car networks have the benefit of not being bound by the same finite capacity of a network of humans trying to manage real-world traffic.

Neither would they have to operate at artificially handicapped speeds, as no human operator would really be able to keep up regardless with the 10,000 or 100,000 cars running around.

Part of reacting and mitigating acute issues is being able to reroute traffic early enough, or slow cars down early enough, to minimize disruption (instead of a traffic slinky blocking everyone up).

There's no reason so far here that self-driving vehicles can't improve traffic and overall throughput, and it's a bold claim to declare L5 dead on arrival.

In a dense urban area, the speed limit is generally not dictated by road design but by the simple physics interaction between a person made of meat and bones, and a multi-ton metal vehicle.

Follow distances in urban areas with low speeds are already really small. I don't think AVs would make them meaningfully tighter.

> Part of reacting and mitigating acute issues is being able to reroute traffic early enough, or slow cars down early enough, to minimize disruption (instead of a traffic slinky blocking everyone up).

This was supposed to be what Waze and Google Maps were set to achieve, and they did not, but they did piss off a lot of people when cut-through drivers started jamming up their formerly quiet residential neighborhoods.

When roads belong exclusively to self-driving cars, they can increase speed, reduce gaps between cars and eliminate traffic wave effect.

Imagine highway where everyone drives 80mph with 5ft of distance between cars. No one suddenly changes lanes, no one suddenly slows down for no reason.

Here is illustration from Waymo testing facility: https://youtu.be/hKfEivMfDPU

Even if there is no reaction time, 5' of space at 50mph is silly. You still need room to maneuver if whatever caused the vehicle ahead to throw on its breaks comes back or halts the vehicle ahead faster than your breaks can stop you.

It’s basically a virtual tandem semi. The presumption is that you are delegating command and control to the lead vehicle, and there is bidirectional communication between the lead and follow vehicles.

Well, why don't we go a step further then and physically connect the vehicles? Maybe we can also make them longer and fit more passengers? And we could optimize the rolling friction, replace rubber on tarmac with something better like steel on steel, and use some sort of tracks to provide the sideways forces?

You jest, but I'd love it if we could automatically group cars into physically connected trains for long distance travel, and have them automatically split off when they approach their destination. Maybe there's some way to do that with rails, carrying peoples' cars, but not sure if the efficiency gain is worth the increased complexity of including another set of wheels/motors/etc, and making the automated separation/joining work with all of that.

Are the other passengers quiet and housebroken? Does the steel track go to my office? (Broken or misused rail switches cause a lot of accidents.)

Lol that is going to lead to the mother of all pile-ups not if but when something just so slightly goes wrong. Imagine somebody dropping shit from an overpass or really anything. Better build a giant cage over all the highways!

Those kind of pile ups already happen now when people drop crap from overpasses, how would it be any different?

Bigger. Faster.

because unless the driver is a raging asshole usually the person behind while you are going 80 has more stopping distance than 5 fucking feet.

That's because you're thinking of human drivers whom are so slow to react that it is recommended they keep a whopping 2 or 3 seconds [0] worth of trailing distance to respond to changing road conditions like the person in front of you slamming on their breaks. The hypothetical clustering of self driving cars requires them to react at much faster times so they don't need to keep as much space between them.

[0]: https://en.wikipedia.org/wiki/Two-second_rule

Also https://youtu.be/iHzzSao6ypE

Local streets will always have pedestrians and cyclists. The main limit on road capacity is generally where highways interface with the local network, and the local network itself.

I believe you'd have better throughput by dedicating roads exclusively to buses and bicycles, which are technologies that already exist.

Congestion is not just caused by reaching capacity. It is also caused by human behaviour:

- selfishness eg stopping or parking in no parking zones, driving aggressively

- error eg causing accidents

- ignorance eg not knowing the traffic flow in an intersection, driving slowly to find your bearings.

Also, if we have fewer cars, but higher utilisation of vehicles then we will require less parking.

When I drive to work there is an empty transit lane for cars with 2+ people which seems to imply that most rush hour cars have just the driver in. If everyone is ubering in robot cars then they’ll probably share. It’s not that weird as people share public transport and choosing between $10 to get to work and $3 is a no brainer.

They might even Uber to the train station then Uber from the destination train station to work. Lots of ways of cutting traffic!

> "they'll probably share"

The people who are driving, or at least a significant portion of them, pick driving over public transport because of perceived privacy and control.

I don't think uptake of shared mobility is going to be so popular where that becomes the norm.

Values change. No one would take a taxi where the driver is just some unlicensed taxi driver. Back in the 90's that would be consider akin to hitchhiking in danger! Now we have Uber.

Also I drive because it's faster, and because once you go from A->B getting to B->C is probably easier. Not for privacy. Most people just need to get to work or school or their kids to school for the bulk of their car use. Kids to school might work by sharing with other parents you know.

Cars are different than a lot of goods though, because in and of themselves they are a wealth signal (at least in the US).

Most Americans don't really need an SUV or a pickup; a minivan or a sedan are more practical. Yet most Americans are buying them, because it makes them feel better about what they're driving and they want to show off. You may only take your car to the grocery store, but if you meet Becky from PTA in the parking lot and she has a really nice car that she owns, that's still a big deal.

Unless the ownership cost of a car were to spike significantly, this part of American culture seems hard to change, particularly when automakers spend lots of money making sure that Americans see premium gas-guzzlers as status symbols.

Jones gonna keep up

Easily scheduled, ubiquitous micro-buses or other vehicle form factors may change the 20th century compulsion of having to have your own private car.

Self-driving cars will of course increase congestion because it will be possible just to chill out and do something else while car is driving itself. In Silicon Valley normal story will be “I worked in the car for 3 hours while commuting to the office”.

Work also needs to be decoupled from inner city offices (so you don't have to go into the city/office/etc) and local and regional legislation needs to be enacted in concert to lower overall personally operated vehicle density.

This is where the Chinese companies are digging. Instead of a self-driving car, they are trying to make a car driven by an external server.

Half of the work on "self"-driving is actually traffic dispatch.

If self driving cars make driving in any way more attractive than now, they'll increase demand which will increase congestion.

Reply on news.ycombinator.com