So in the first post I said "I remember that back in Albany, my peak traffic counts were on Marin or San Pablo, about 10.3 cars per lane-minute."

But I remembered something key last night: left-turn lanes. Both streets had them. Traffic on Marin could keep flowing smoothly through an intersection because cars going left could get out of the way. But that takes space. Assuming that each direction can carry 10.3 cars a minute, that's nearly 21 cars a minute, but spread over three lanes -- two travel, one shared turn. And we're back down to 7 cars per lane minute.( Read more... )

Layman's conclusion: wide roads with little 'turbulence' can get up to 10 cars per lane-minute. More complicated streets are unlikely to get above 8, after accounting for turn lanes. This will have consequences for stuff like "is it physically possible for everyone to drive to work from here?"

In the past, I've estimate the car capacity of city streets with "two second following distance, cut in half for intersections" which yields 15 cars per lane-minute. But what about reality?

Signalized

Today in Philadelphia, around 6 PM, I counted at various intersections. One lane of travel, 30 second light, reliably had 10 cars passing me before running out of light rather than cars. 3 seconds per car.

At much bigger intersection, a near-highway with 6 total lanes of travel, I counted 37 cars in 38 seconds, and 33 cars in 38 seconds. In this case it was the cars that ran out first, I suspect the previous light cutting off supply. 3 lanes in the direction, so once again about 3 seconds per car per lane.

Your big bottlenecks will be where two major streets intersect, each getting green for half the time, so 10 cars in 30 seconds of green is basically 10 cars per minute overall, on average carrying about 15 people per minute.

One articulated bus (120 people) every 8 minutes would double the capacity of a lane, while serving a lot of people who can't or don't want to drive. Or better, turn the lane to a bus lane, keep the same capacity while serving a lot of people etc.

Mostly, it was funny to count 10 cars before the next red light, and think "if one of these was a dinky bus it could be carrying 40 people." Heck, even a little passenger van carries 15 people, one van a minute doubles your capacity.

Stop signs

So much for signalized intersections; what about 4-way stops? I'm not sure; for one thing, my nearby intersections didn't feel like they were at peak traffic. For another, a busy intersection is messy. At first I paid attention to just one lane at a time, and got maybe 6 in a minute, then 7. Later I counted every car going through in all directions (actually the intersection of a two-way two-lane and a one-lane), and got 52 cars in 3 minutes; 3 lanes, so 5.78 cars/lane-minute. But the traffic definitely wasn't fully saturated... Of course, when orthogonal directions are saturated, that slows both down, as do pedestrians. Especially since Philly drivers make rolling stops when they can, so being physically forced to actually stop would slow them down.

I remember that back in Albany, my peak traffic counts were on Marin or San Pablo, about 10.3 cars per lane-minute.

When self-driving cars first started getting talked about, like 10 years ago, many people were enthusiastic and anticipatory. I was skeptical, because as someone who walks around dense cities, driving safely and effectively in such felt like a human-complete AI problem, needing theory of mind, social interaction, and a large amount of adaptation to unforeseen circumstances.

Also because while in some things like chess or Go, rather dumb computers beat humans through powerful search, a more common AI pattern is that a fairly simple system can get 60-90% of human performance, but then stalls despite a lot of effort. Which is fine when you're making models for targeting direct mailing, and poorer performance can be balanced by much faster turnaround time and it's just moderate amounts of money at stake anyway. Less fine when even a missing 1% of performance may mean people die, or alternatively that traffic is frozen as cars can't figure out how to safely push through busy streets.

(The direct mailing example is from my first full-time job; we could build a decision tree, to predict response rates to a direct mailing, that was said to be 60% of a hand-crafted model but took a few hours instead of a few months to create. A machine translation course in grad school included various systems that could do 60-95% as well as humans, on fairly narrow word tests, but improving that was Hard. Statistical translation, rule-based, hybrid, all stalled.)

Basically an application of the Pareto principle: 20% of the work can get you 80% of the performance. Except it might be more like 1% of the work gets you 80% of the performance; since we don't *have* human-equivalent AI in most of these domains, we can't even say how much work it actually takes.

Early articles were along the lines of "we're making lots of progress! (but can't drive in the rain or snow and are tested mostly in low-density sunlight)", which for some people sounded like "we're almost there but for a bit more work" but to me sounded like "we're already spending years on the *easy* stuff, imagine what the hard stuff will be like."

More recent articles have been more like "wow, this is harder than we thought", with even the executives in charge of developing and selling this stuff saying like "thirty years away" or "never" or "far in the future", or "decades away".

Singapore reportedly has deployed them, as someone on Facebook likes to keep saying, but a friend there observed various caveats: 10 MPH, a bounded area, not mixed with other cars, safety driver, and attendants trying to shoo pedestrians out of the way. Also see. And this is the state of the art!

So, "ha ha!"

I'll also include a FB thread I made two years ago about predictions, and include just one example of receding predictions:

2014: Volvo promises fully self-driving cars by 2017, 3 years later.
2017: Volvo promises partial self-driving cars by 2021, 4 years later.

This is worth calling out from the previous post: look at https://en.wikipedia.org/wiki/List_of_countries_by_traffic-related_death_rate and sort by the various columns. In particular the third one, road deaths per billion vehicle-km. US is 7.3, Japan is 6.4, not hugely better. Most rich countries are better, down to 3.4 (UK) or Norway (3.0) Many rich countries are least 1/3 better than the US (5.1 or lower).

So when we talk about the 40,000 car crash deaths a year in the US, and how preventable they are, there are two dimensions: reducing the amount driven, by increasing density and mass transit and bikeability, and improving the safety of cars as they are driven, by I don't know what means exactly but roads can clearly have only 40% the death rate of US ones.

Between the two, well, Canada and Australia (large car-loving countries like the US) have less than half the road deaths per capita of the US, so 20,000 American deaths/year are easily preventable. Looking at the UK or Nordic countries, 30,000 deaths/year are preventable.

Shortly after I got here, my sister was driving me home and some SUV nearly merged into us on the freeway. I was trying to figure out if it would be tactless to say "this is why I don't drive" when she went and said it for me.

Then Martha was driving me home, and some car in a driveway nearly backed into the passenger side, aka me.

And finally today, I was exploring Oakland's Chinatown on foot, when there was a screech-thump in an intersection followed by two cars slowly pulling over with their hazard lights on. I wasn't actually in danger, being on a corner, but if it had happened a few seconds earlier than I would have been rather close to whatever happened.