I know a few traffic engineers and one of the things they regularly laugh about are zipper merge advocates (as well as traffic circle advocates, but that's a discussion for another time). It's not that they don't recommend the zipper merge, it's that the advocates think they are magical and ascribe magical effects to them that don't exist in reality.
Want an example of the perfect zipper merge? Look at an hourglass. Thousands of tiny grains of sand perfectly zipper merging at the last instant. And yet there is still a traffic jam. Inflows and outflows still dominate traffic congestion. Now imagine what would happen if you early merged those grains of sand. Your hourglass is now 6 miles tall. Interestingly enough, the early merge is more efficient for flow, but not by enough of a margin (1-2%) to overcome the drawback of the terrible space efficiency.
What the zipper merge is really good at isn't improving flow, it is improving space efficiency. The bottleneck uses far less of the highway when people merge last instant. This allows people to reroute or exit the highway before the traffic jam, reducing the burden on the bottleneck. That's it, nothing more. The idea that traffic would suddenly get so much better if people zipper merged perfectly is not rooted in reality...traffic volume is still the more important factor by several fold.
> This allows people to reroute or exit the highway before the traffic jam, reducing the burden on the bottleneck.
Many of those people could have already been intending to exit the highway earlier. Because of the back-up, their journeys are slowed way down unnecessarily. And then you get compounded effects - the highway backs up past another entrance, causing that feeder road to back up past an intersection... In the worst case you end up with gridlock.
I'm not arguing that zipper merges are magical either, but they can definitely have a significant impact.
Instead, most zoom up until the choke point, then late merge, and rarely do it right, leading to someone slamming their brakes - then traffic jam ensues.
I don't wait for the choke point if I can avoid it. When I see signs that say "construction ahead" or whatnot, I start looking for the signs that say how far ahead, and what lanes. If those don't exist, I try to see what the traffic is doing as far ahead as I can see - looking taillights, turn signals, and anything else - to tell me what lanes are closed, etc.
Then I get over as far and as quick as needed, so I am in the lane I need to be in long before the choke point.
I do the same thing in regular driving; if I know something is going to be on the left side of the street, I try to get over to that side long before I need to - instead of waiting until the last minute and cutting people off (or "begging" with signals to be let in).
I've always found it to be a pain to do it otherwise, even when I don't have the choice and I am forced to, because I have to change speeds and lanes, and that puts me more at risk of collision (it's not the speed that causes problems, but the delta in speeds and lane changes). I'd rather "merge early" (whether that means before real choke point, or just being on the right side of the road to where I am going or turning long before I need to be there) because it seems safer.
The problem is (mainly with choke points, or even on/off ramps) that other people now see the "empty" lane as "I can speed up and go around everyone" and then they get to the choke point and force their way in, or slam on their brakes and swerve, or cause an accident, or anything else. They could see that lane was empty, they had (usually) warning signs miles back saying that the left lanes were closed ahead - but apparently that didn't apply to them.
> reducing the burden on the bottleneck. That's it, nothing more.
That's precisely it though. By reducing the burden of the bottleneck on the greater traffic network, it localizes the congestion to a smaller area and reduces the disruption the rest of the traffic network has to feel. Overall improving congestion, even if locally the congestion still occurs.
> What the zipper merge is really good at isn't improving flow, it is improving space efficiency.
This is based on the human execution of a zipper merge. Ideally, the continuing lane of traffic will be going at a consistent speed with enough space between vehicles to allow the merging traffic to merge without causing a disruption in the flow of traffic. Theoretically you'd get this benefit from merging earlier as well, but as you mentioned the space efficiency causes cascading impacts to other aspects of the traffic network.
That being said, I only mentioned that as a recognizable example. However, a lot of congestion isn't due to a specific trigger such as a lane ending. It's due to bad human driving habits, and suboptimal human reactions to that behavior. Which create phantom traffic jams[1][2]. Self-driving cars will drive more consistently than human drivers, introducing a level of calm in a sea of inconsistent human reactions. Effectively buffering traffic in a way that will improve overall traffic flow, even before we have majority adoption of self-driving cars.
It is possible the reduced need for downtown parking could significantly reduce congestion in cities though.
Self-driving cars will not only be better drivers on the whole, but also allow human drivers to break through inhibitions related to "proper road etiquette", since it's a whole lot easier to do something subjectively rude to a machine than a human.
[1] http://transportation.ky.gov/zippermerge/Pages/default.aspx
[2] https://arstechnica.com/cars/2014/07/the-beauty-of-zipper-me...
[3] https://www.nytimes.com/2016/10/13/us/why-last-second-lane-m...
[4] http://jalopnik.com/stop-trying-to-make-the-zipper-merge-hap...
You're completely right that it could cause an increase in aggressive driving behavior. But it's just as possible that it'll decrease aggressive driving behavior. My conjecture's based off of three personal observations:
1) A lot of the aggressive drivers I know that have road rage-like tendencies take it as a personal affront when another driver does something like turn in front of them or using a less congested lane of traffic to get as far ahead as possible before getting over.
2) One of the most effective strategies I had available as a retail manager to diffuse a situation with an irrational/irate customer was to pass blame on to "the system/corporate policy" if they couldn't see reason. They'd still be irate, but they'd redirect their behavior towards something other than the immediate situation. The same principle applies when you go from blaming a specific driver to blaming "those damn autonomous Nissans. They always <insert some peculiar trait of Nissan's self driving implementation>".
3) Aggressive driving is hard to escalate into full-blown road rage when one participant is a computer. As you said, it'll react to the situation by potentially braking aggressively, but it won't then try to tailgate the individual and escalate the situation. It'll back off appropriately and assure that it's not put into that position again and be done with it. Aggressive drivers will get bored with antagonizing autonomous cars when they get used to them and to the fact that you can't get a rise out of them. As well, autonomous cars can likely get better at identifying indicators/patterns of an aggressive driver and account for that.
Once you remove the inconsistent skill level, distractedness, irrationality, and emotional reactions that accompany human drivers, you can increase road utilization considerably without a noticeable increase in congestion.