Robocars improve traffic even when most cars around them are driven by people | Hacker News Reader

Robocars improve traffic even when most cars around them are driven by people

theconversation.com

19 pointsgmays18 comments

I think it's plausible. I believe a lot of traffic would disappear if people drove the same speed. I do a lot of highway driving and use cruise control, and I see lots of cars go faster, then slow down, then go faster, then slow down, etc, causing more lane changes and probably annoying other drivers. Lane changes takes up two spaces and causes others to slow down creating a domino effect.

Traffic is one of those things where if everyone else looked out for each other, waited their turn, let people merge, etc, then everyone would get to their destinations sooner.

There's a group in UT Austin that has spend lots of time trying to model and analyze such dynamics: https://www.cs.utexas.edu/~aim/

My experience with Waymos is that I am much more cautious around them both out of curiosity, but also because if something does go wrong, I don’t know what signals I have to change their behavior (e.g. honking, polite gesturing, etc.). Once autonomous cars are more common, I wonder if people will relax back to relaxed/sloppy/aggressive patterns. Maybe most already people don’t notice.

As an aside, I left a baseball game a few months ago and was surrounded by six Waymos leaving downtown with the game traffic. How wild to see almost the same number of driverless cars as traditional cars. My kids are young enough that they won’t know any different, but from my mid-40s perspective it may as well have been the Jetsons.

"Optimize" traffic flow all you like it's not really relevant at all because it's not the dominant problem, cannot ever address the fixed spacial constraints of car and road network, and whatever gains are achieved will be instantly swamped by induced demand and the fixed road network.

There may be hypothetical efficiencies to be found in a closed network, but that is not real life. In real life people may choose to begin driving or stop driving at any time. We need to be aware of the impacts of people deciding to drive more.

In real life as soon as traffic improves and time spent in traffic decreases that is a pricing signal to the market, lowering the time price of choosing to drive instead of other alternatives, and encouraging people to switch to driving vs other alternatives. The amount of traffic will increase again until it reaches the fixed maximum throughput of the road network or the highest price people will pay in terms their time before seeking other alternatives.

This is the well studied phenomenon of induced demand. Previously of concern to engineers weighing whether to build a 10 lane vs 8 lane bridge, but also very applicable to this new technology.

When we consider that there is a significant amount of people that currently cannot drive that will now be able to drive with the introduction of self driving cars, such as children and the elderly, it's clear that the most likely outcome of prevalent self driving cars will be a significant increase in the amount of cars on the road. This induced demand for driving will necessarily make traffic remarkably worse regardless of whatever marginal optimizations may occur.

>When we consider that there is a significant amount of people that currently cannot drive that will now be able to drive with the introduction of self driving cars, such as children and the elderly,

Also, I'd expect the average number of people per car to decrease. I could even see the average dipping below one. Just think of things like sending your empty car to the vet, to pick up the dog. Or sending an empty car to Walmart to have them fill your grocery order. Or people who will have their empty car circle around instead of paying for parking (or because there are no parking spots available). And think about how no-passenger cars aren't really going to care too much if they are stuck in traffic.

I agree. But I also think that, then, the size of the vehicles will decrease to accommodate these new realities. Perhaps no steering wheel, smaller dashboard, no need to divide seats. There's lots of space savings to be had.

If you have more cars traversing the same road with the same commute time, you have increased throughput. While it might not mean much to the individual driver, it's beneficial for the community the road serves to be able to move more people and goods around.

There are also potential benefits to the individual besides reduced commute time. For many people the real pain point of congestion is travel time variability. If you expect a drive to take 30 minutes, that's not particularly onerous. But if a drive you expected to take 10 minutes winds up taking 30, that is terrible. When there is variability, either you are late to arrive or you need to add buffer to your travel time, both of which are highly undesirable. Even holding average travel time constant, if you can reduce the standard deviation through optimization, that is a huge quality of life improvement.

There is also a safety concern. Congestion causes increased rates of accidents for a variety of reasons - less clearance between cars, people driving abnormally trying to make up for lost time, overflow to alternate routes, road rage, etc - which obviously is a problem for both society and the individual driver. Getting traffic to flow more smoothly, even if there were no throughput increase at all, would be a substantial benefit.

Finally, induced demand is just demand that is currently unsatisfied. There are people who currently live where they don't want to, work where they don't want to, or avoid trips altogether that they want to make because of excessive congestion. Improving their quality of life while holding the quality of life of current drivers constant is a good thing.

> If you have more cars traversing the same road with the same commute time, you have increased throughput

Yes and then more cars appear.

> Improving their quality of life while holding the quality of life of current drivers constant is a good thing.

Sure absolutely people need to go places and we need to get them there. However we have neglected to consider the negative externalities of induced demand and the quality of life of everyone else who isn't a car driver. The geometric realities are such that induced demand can only be satisfied by growing the size of the road network. This has incredible costs and significant negative impacts on the community and environment. There are other alternatives that achieve the same goals.

While there may be "efficiencies" to be found they cannot make up for the fixed dominant variables, the geometric reality of ever increasing SUV size, and fixed road sizes. As soon as any efficiencies open up free road space the amount of cars will simply increase, quickly closing that free road space again.

So this doesn't mean that efficiencies can't be found, it just means that we need to understand the limits of efficiencies and we should not expect this to be a core solution to our traffic problems or as a key part of our broader transportation strategy.

> Yes and then more cars appear.

This is after those cars appeared.

> The geometric realities are such that induced demand can only be satisfied by growing the size of the road network.

We are explicitly referring to increasing the capacity of the road network without growing its size.

> s soon as any efficiencies open up free road space the amount of cars will simply increase, quickly closing that free road space again.

Again, we're not looking to increase free road space, the additional cars filling that space are the benefit.

There is some absolute limit on throughput where the road is completely packed full of cars all moving at maximum speed, and the only way to further increase throughput is expansion. But most road networks are no where near that limit. When you are in bumper to bumper traffic on a 60 mph highway moving at 5 mph, that road is carrying about 1/12th or 8% of its maximum capacity. That's an immense opportunity for improvement.

I feel 100x safer in a Waymo than an Uber in SF. That said I did have a Waymo lose its mind and take me to a random alley in Chinatown at 3am…that was fun.

It’s a theoretical result, not something happening with current self-driving cars.

“Our work is the first to demonstrate the feasibility of controlling mixed traffic via robot vehicles at real-world, complex intersections.”

> It’s a theoretical result, not something happening with current self-driving cars.

I'm having a hard time finding the specific reference, but Waymo has a solid study showing that the presence of their cars in the areas they operate in became overall safer.

I skimmed the article, but the only metric I saw mentioned was that "traffic jams were eliminated". What metrics do the hypothetical robocars improve?

I ask because time-to-destination is not the only metric I find myself caring about. I will often choose a much longer, scenic, route instead of the congested highway even though it adds 10 (estimated) minutes to my trip. But I'd rather be actively driving, moving through space and feeling like I'm making progress for an hour than sit in a barely moving queue for 50 minutes.

This video illustrates the principle: https://youtu.be/iHzzSao6ypE?si=maK5yPJn7LMhPDU1

Basically, people following too closely cause more braking, which cascades to more braking for an obstacle that was never actually there.

I can't seem to find the original source, but I read years ago that a surprisingly small percentage (I believe 4%) of drivers just leaving excess headway was sufficient to mitigate most congestion-causing traffic waves. It makes sense to have robots be the ones driving in that counter-intuitive fashion.

I recall another report that most traffic jams could be made much better if people slowed down when approaching and were accelerated their way out on the other side. Makes sense from a basic flow rate / bottleneck perspective, but people will still race around others to late merge 10 cars up and cause everyone to got to a full stop...

This is talking about purely hypothetical future robocars that don't exist today. Yes, great. Flying cars would improve traffic too.

Only when you ban pedestrians and cyclists. I.e. in a dystopia.

Reply on news.ycombinator.com