Zipper Merge -- A Literature Review

[gfish]

It's getting common to see transportation departments 1, 2, 3, 4 and news stories 5, 6, 7 advocating zipper merge (AKA late merge), claiming that it is safer and allows higher throughput. As someone who naturally doubts the pronouncements of transportation departments, I thought I'd look into the research behind it

Comments

[caladri]

what we wanted to do in the first place

I mean, some people clearly want to do late merges, but I can't imagine why. It's like going a third of the speed limit on a highway on-ramp and then waiting at the bottom of an opening. The latencies, the narrow windows, etc., are all just awful and have really gnarly cascade effects that are not outliers, but common. Merging over when there's time and space, and when you're moving at speed, seems so much better. Now, if traffic is flowing freely and with adequate following distance to allow it, I can see how late merging would increase order and throughput, but you're also using up all your runway and increasing the chance of hitting the fallback case where you have to come to a total stop.

There is, it seems to me, an argument for consistency, though. Like, a mix of early merging and late merging can go worse than plain late merging; and as long as someone can late merge, they can worsen the situation overall, and also trigger everyone's happyfun responses to the feeling that someone

[randomdreams]

I feel like consistency is exactly what's going on here: they figure some people are going t late-merge anyway, so if everyone does it, everyone knows what to expect, and maybe that'll make things move faster.

but mostly I'm just cheering for level 5 autonomous vehicles.

[eub]

Yeah, I have no data, but this is my intuition -- that early opportunistic merge would work better if everybody did it, but it risks breaking down messily.

Late merge is a stable strategy because there's no way to "defect", there's no advantage to an individual in being the one who merges early. Early merge sometimes leaves an open lane stretch for some dick to skip ahead, then want to get back in, other people try to punish the dick, dick gets aggressive or plays near-chicken to get in -- risky situation.

Merge is a Prisoner's Dilemma where the idea is that the Defect/Defect outcome (all late merging) is actually collectively better than the messy Cooperate/Defect mixed state. The C/C would be best, but it leaves an incentive for defectors. Doesn't mean people will, but that's what I would look out for.

[noumignon]

The entrance to the express lanes on N I-5 in downtown Seattle is a prime example of this. There, a lane becomes exit-only that many people want to take. Many of them apply late merge techniques, zooming up to the front of the line and then trying to merge. In doing so, they block an entire lane that would otherwise be open, and this sometimes carries over to slowing down even the next lane, as people dodge out around the blockage. This is the core of my objection to zipper merge -- it encourages people to be jerks.

Which lane are they blocking? The one they're zooming up in? Big deal -- there's nobody in that lane, that's how they zoomed up it.

"Encouraging people to be jerks" is begging the question. Best to design a system around human nature, so that something tempting and efficient (like right turn on red) is also the encouraged action.