How California is changing its roads for self-driving cars | Hacker News Reader

How California is changing its roads for self-driving cars

scpr.org

65 pointskjhughes54 comments

The last paragraph is the most interesting one from a policy perspective.

"I have a concern about it increasing Vehicle Miles Traveled because I don’t have the capacity to handle that. We already don’t have enough capacity for the number of miles that are being driven today."

Of course, they'll increase miles traveled even if they're not allowed to drive around empty (which, of course, negates their use as taxis, etc.). Make something easier/cheaper/etc. and you get more of it. I expect this is still quite a ways off before it's a real question but it's going to be a problem for areas where infrastructure is already close to the limit.

I agree that Vehicle Miles Traveled will likely increase, but it's important to note just when that increase will occur. I suspect that increases occurring in off-peak hours can be handled by current infrastructure.

Even if this is correct, an overall increase would increase road wear proportionally.

Weathering and heavy vehicles do most of the damage that is done to roads, so increasing passenger vehicle miles won't increase wear proportionally.

I guess if shipping increases that would increase road wear. But that is also an easy to justify tax.

Not sure why that would be the case if people are still working roughly the same hours but commuting longer distances and potentially sending cars off to park in less congested areas. Furthermore, for a lot of major cities, traffic is pretty bad across broad areas and for pretty good chunks of the day and night already.

I've seen estimates in industry reports estimating a 400% increase in vehicle miles per year as a result of auto-piloted vehicles. We clearly need to rethink how we inhabit this planet as this is complete insanity.

A 4x increase, but a world without speed limits, stop lights, or traffic lanes. We could support 10 times the amount of traffic in that case.

That sounds hellish for walkers, cyclists, motorcyclists, and people who enjoy quiet.

That sort of thing won't happen on regular city streets. Maybe on freeways.

Sure it will. You'll be able to walk right out into traffic safely

Right. Because once you have AI the laws of physics no longer apply. The braking distance with an instantaneous reaction time at just 40mph is still about 20 feet.

All the cars around you will see you before you get anywhere near the road. As long as you approach the road to cross, perhaps as you've always done, you'll be granted access to cross with little fear.

https://www.youtube.com/watch?v=4pbAI40dK0A

It's already crap on roads in the US (the US has very few of what you'd call a "street" - "stroad" would be more accurate). Having more VMT even with current speeds makes it worse. It's shameful that EV's are being forced to be loud when urban noise pollution is already terrible.

It will still be loud as hell. Especially with rules mandating that EV's be as loud as their internal-combustion counterparts.

But there has to be a limit to demand for VMT at some point, right?

If car travel was easier and cheaper, would more people take more trips? Of course. There are obviously trips that people don't take because they're too cumbersome with current technology.

But at the very least we know 1 person cannot exceed more than 24hrs of equivalent VMT. So there's at least a hard limit there. And I doubt most people want to spend the majority of their lives in cars. So there's some natural limit on VMT that we'll approach.

I mean, think about it like this. If self driving taxis were a thing, would you be in the car 4x more than you are today? I wouldn't want to. Maybe I'd take more weekend trips to LA for fun excursions, but during the week my schedule would probably be about the same.

So the hope, I think, is that the improved efficiency of self driving cars will allow us to easily hit that ceiling with at least the current infrastructure, if not _less_. It's not hard to imagine that the current infrastructure could become at least 400% more efficiency with self driving cars. Cars that can drive bumper-to-bumper at 60+ mph, no accidents, no rubber necking, no aggressive driving, no brake checks, and no traffic waves from lane changes. Not to mention the space saved by using taxis sized for the task. Right now people drive 4 seaters for their daily, 1 person commutes. Imagine all of those people are now in 1 seaters. Maybe even self driving motorcycles!

Some limit although I could easily see it doubling especially if there's a lot of "deadheading" to reposition empty vehicles. I think about my own situation and I'd definitely go to more evening events and I'd be inclined to go further for weekend recreation if I didn't need to personally drive.

As for your other points, they mostly apply if 100% autonomous vehicles were the only thing on the road. Even if they were available today (which they're not really all that close to), it's hard to see that being a political possibility for many decades even if there were no other obstacles.

Yes, 100% will take awhile, but we're likely to see Self Driving Car Lanes first. In fact, in California, we can just replace the car pool lanes with Self Driving Car Lanes.

These would be lanes of the freeway or multi-lane roads that only self driving cars can drive in. That would allow most of the efficiency gains to apply.

Then we'll see a phasing out of bad drivers. People who get DUIs, X number of tickets, etc would be required to use a self driving car. So of the human drivers left, more will be competent.

So basically, as self driving cars add additional load on the roads, we'll be slowly giving them more and more efficiency advantages to offset that load.

Brad Templeton, who was also in attendance at the automated vehicle synopsum wrote:

"The AVS has a lot of governmental people, and they’re all very keen to imagine their role, which they see as making the infrastructure “ready” for robocars. There was a whole long session on the topic, and many people who imagine there is a lot to do. This is the wrong impression. Robocars are being designed to handle the infrastructure we already have, and only low-skill robocar makers are suggesting we need to make significant changes to the infrastructure to enable these vehicles."

I mostly agree. That said, there may be relatively small things that can be done which help out with assistive features like lane keeping or which help out in corner cases--like identifying construction sites--which may be helpful.

Holey hubris, Batman! I bet there will be large dark zones on the AV's maps, where they dare not go, for reasons they cannot comprehend with their own sensors.

Then I expect they won't be going to those large dark zones. I largely agree with his comment. It's not that we shouldn't do anything. (If it makes sense to make lane markings a bit wider, why not?) But if we really need to change large parts of infrastructure to accommodate automated vehicles, that's a non-starter just as it was when that sort of thing was discussed in the 1960s.

California already has made massive infrastructure changes from streetcar removal to building BART.

When we see articles about AI needing to be regulated by CEOs like Elon Musk, it's not because AIs are going to take over the world. It's because AIs alone cannot work seamlessly in certain environments and regulations are needed to make the environment(eg roads, signs, bicycles, pedestrians, other vehicles) conform to the needs of AI.

See https://en.wikipedia.org/wiki/Regulatory_capture

Hopefully, when these regulations are written, there will be rules added to ensure drivers and mechanics can inspect and debug the computer code in their vehicles.

> How do you delineate clearly?

> One area this comes into with California is we’ve been using those Botts' Dots. The automated vehicles [AVs] can follow lane lines. They can’t follow the Botts' Dots, so we’re actually changing our delineation standards to go away from the Botts' Dots which we’ve been using for decades because AVs have a difficult time following those.

That is very surprising due to claims of superhuman image recognition due to deep learning. Is it because of speed issues?

Lots of claims get made about AI capabilities that are... optimistic. From a paper on the topic: "Botts’ dot marker detection is more difficult than that for white lines because small, sparse, low contrast regions are mixed in with road-texture noise." https://trid.trb.org/view.aspx?id=1253090

No matter the technological progress, I know I'll hate the moment I'll cross a vehicle whom I can't interact with.

Why?

The same reason you get frustrated when you realize the customer service agent you have been talking to the whole time is just AI powered, and more then 3/4ths of the response have been pre-filled.

Loss of the human element. Kind of hard to describe, but very obvious when it is missing.

Loss of intelligence / skill is the problem. If that AI understood your goal, and had the ability to modify your account, then we might actually prefer the AI version.

Until the situation I am in requires compassion and or empathy to resolve. Maybe the outcome isn't desired from my perspective, but I feel better about it now.

A human customer service agent might have all the empathy in the world, but if it's not permitted under their company policy then it is unhelpful.

The point I'm making is that empathetic outcome in customer service interactions is the outcome of explicit and implicit decisions by the organization. Empathy can only be used if the employee knows he or she won't face a negative impact.

Sometimes bending explicit policy is part of providing exceptional customer service. Not every situation is the same. If companies never bent the rules, you would never get these wonderful PR stories that companies love -- as an example.

intrinsically this is part of the problem -- we are reducing decisions down to basically a hard set fast of rules that I am sure AI will implement and arrive to the correct rules better than any human can -- but we aren't robots. There is no correct set of levers to pull, and you can't take emotion out of decisions that are made. Nobody likes strict legalism, and you can't possibly make rules for every single situation.

I was driving to work one day and the light was broken at the busiest intersection in town. People were able to make it through the intersection using hand signals and flashing headlights. How does this work when half of the cars are fully autonomous? How are computers going to understand what those hand signals mean?

A bridge in my town is under construction and it is down to one lane. They put up full stop lights at both ends of the lane closure to manage traffic. Every two or three times I go over that bridge, someone coming from the other direction goes on red because they can see no traffic is coming. I have to sit and wait for that vehicle before going or no one will be able to move.

I love the idea of fully autonomous vehicles, but next time you drive somewhere count the number of ambiguous situations that occur and ask yourself if you think a computer could be programmed to handle all of it. I'm rather skeptical.

There will be a rather long period where auto-cars still have steering wheels and can ask their occupants to take over.

That's almost certainly the case.

However, as soon as you've said that, you now have a requirement that:

- There be a sober, competent, licensed driver in the vehicle

- There needs to be a reliable non-instantaneous handoff from the vehicle to the snoozing driver

- The driver is probably still ultimately responsible for actions taken by the vehicle

Don't get me wrong. I'll take a reliable highway autopilot system. But this does largely rule out many of the use cases that people are thinking about when they say self-driving car.

Those use cases can come after the technology has proven itself reliable. You don't have to have the perfect solution with the first production car.

Kyle Vogt, YC alum and CEO of Cruise said on Monday "It's looking like months, not years, that we're going to have the first commercial production of driverless cars". They're pursuing robotaxis which won't have the option of allowing an occupant to take over.

The initial intent with Google's Koala cars in 2015 was to deploy them sans steering wheel, but they were thwarted by hamfisted regulation from California's DOT. That's expected to be revised before the end of this year.

I'm very skeptical about this kind of timeline, unless we're talking about very well known/controlled routes. Let's see.

Cruise is in San Fransisco and Phoenix right now, so as well known and controlled as driving pretty much anywhere in downtown SF is what we can expect, and that's not easy driving.

Another interesting thing Vogt said on Monday "Looking just at disengagements, if you extrapolate, based on past data, and using our internal metrics, we think we'll actually be in the number 1 spot in 3 to 6 months"

So at last report in December of last year, Waymo's test vehicles had 1 unplanned disengagement every 5000 miles on average, which is impressive, but Cruise thinks they'll be beating that benchmark in very short order.

I agree. That is why we shoukd target Freeway AVsfirst. It is the most controlled and homogeneous environment of driving, and driving it is tedious.

I've contemplated driving across country to visit family, echewing flting. It would be great if some of those miles happened while I slept.

A broken stoplight is considered a 4-way stop under the law here. There's no need for hand signals or flashing headlights. Though, Police may direct traffic if necessary.

In the case of alternating one-way streets, I've always seen construction workers directing traffic. People will obey a person, even when they won't obey a light. Alternatively, consider installing a red light camera.

> A broken stoplight is considered a 4-way stop under the law here. There's no need for hand signals or flashing headlights.

This intersection is 4 lanes plus a left turn lane in all directions with constant traffic. Hand signals were absolutely necessary because 12 people were all approaching the intersection simultaneously.

> In the case of alternating one-way streets, I've always seen construction workers directing traffic.

It's nice that you have always seen that. I'm not sure it would make sense for my town to have construction workers directing traffic 24/7 for the next 3 months while they repair the bridge, so they put in a temporary light. Regardless, I don't know what your point is.

>This intersection is 4 lanes plus a left turn lane in all directions with constant traffic. Hand signals were absolutely necessary because 12 people were all approaching the intersection simultaneously.

There is a set of rules that describes which cars go in what order.

The big problem I see is human drivers are going to start driving more aggressively because they know automated vehicles will always back off first.

Right now there is not much interaction happening with drivers texting or watching videos on their phones. I cannot wait for the moment this doesn't imperil their and others safety.

I think the self-driving plus electric cars will indeed change the transportation model by reducing the public transportation appeal. As the result the demand for more roads will only increase. Unfortunately, many politicians (as many generals) are fighting the previous war and building public transportation projects that will be abandoned even before completion (e.g. Brown's bullet train). Instead, we should be building more roads or may be more efficient roads (multilevel?) to acomodate the upcoming increase in traffic.

Studies have shown that increasing road capacity just causes more drivers to join the party to meet the supply. [1] Multilevel roads would be sick though. I'm hoping as vehicles become more autonomous they'll be able to move more efficiently, reducing congestion. A lot of slowdowns are artificial, people looking at an accident, a group driving slowly backing up a couple lanes.

1. https://www.wired.com/2014/06/wuwt-traffic-induced-demand/

<widen the lane lines from four inches to six inches so they are readable.

Time for Time/Life to publish a coffee table book. My interstate and secondary roadway sysyem hasnt changed much. Accommodation will alter the feel.

"Today, we know we can’t put magnets in the pavement everywhere for cars to follow..."

Why not?

As someone from a Midwestern state, I fear that all the self-driving car development in California is going to miss the important edge cases we experience here. Snow, plows, salt, dirt, heavy rain, etc all destroy the visibility of lane lines.

What we need is something that can be a marker for cars that doesn't require line of sight, or else self driving cars aren't going to be able to navigate at the lightest snowfall.

We have those edge cases too. California is not all sunshine all the time.

> As someone from a Midwestern state, I fear that all the self-driving car development in California is going to miss the important edge cases we experience here. Snow, plows, salt, dirt, heavy rain, etc all destroy the visibility of lane lines.

You overestimate the quality of roads in California. :) In San Francisco and Oakland the roads are, quite literally, falling apart. Take a ride on 880 -- straight up four lane high way with virtually no road markings.

This is one of the many reasons Pittsburgh is becoming a hotbed for self driving tech for many companies testing there. It has a lot of bridges, plus all the environmental factors you described above.

I assume that comment is a realistic assessment about how existing infrastructure can't realistically be dug up/torn down to accommodate autonomous vehicles. Incremental changes can be made over time when other scheduled changes are being made. But, for the most part, AIs generally will need to be able to handle the physical and human world as it exists.

>You’ll be able to put more cars faster and closer together when they’re autonomous.

I hate this misconception about driving. That being closer together (assuming front to back) is somehow useful. That restricts lateral movement and is one of the biggest causes of slowdowns on freeways. Say you're in the fast lane and you need to exit. If there is a column of cars next to myou, driving less than 1 car space apart, all of those cars need to slow down to adjust and let you over (or you need to slow down significantly, slowing the column of cars in your lane until you find an opening). Leaving space allows for lateral movement with no adjustment of speed.

Reply on news.ycombinator.com