[VikingCoder]

"“Our experiments show that with as few as 5 percent of vehicles being automated and carefully controlled, we can eliminate stop-and-go waves caused by human driving behavior,” Daniel Work, assistant professor at the University of Illinois at Urbana-Champaign and a lead researcher in the study, said in a statement."

https://www.rdmag.com/news/2017/05/study-shows-self-driving-...

[benjaminjackman]

Per the article: > “Before we carried out these experiments, I did not know how straightforward it could be to positively affect the flow of traffic,” Jonathan Sprinkle, the Litton Industries John M. Leonis Distinguished Associate Professor in Electrical and Computer Engineering at the University of Arizona, said in a statement. “I assumed we would need sophisticated control techniques, but what we showed was that controllers which are staples of undergraduate control theory will do the trick.”

If it really is that simple shouldn't it be possible to put this into an app that if 1/20 drivers had installed would also solve the problem?

[Phlarp]

An app can't convey near the granularity of control to the driver compared to what a fully automated solution could do, and you run the risk of making the situation worse due to the driver splitting focus between paying attention to the road and following the app instructions.

To some extent we already have "the app" (google maps) and I wouldn't be surprised if it's running for close to 1/20 of drivers actively in traffic, and a whole lot more passively collecting and reporting location info that is trivial to map to major roads. To that argument Waymo somewhat belongs to Maps (or Maps to Waymo if you'd like).

[benjaminjackman]

> An app can't convey near the granularity of control to the driver compared to what a fully automated solution could do

Yes I agree, however it might not require that level of granularity, that's what I was wondering given that the researchers were surprised by it working as easily as it did in their experiment. Since, That's why I am wondering, maybe it's something a human could be taught or directed to do by an app.

Open questions to me: Is central coordination required which observes the current state of the system and doles out specific instructions depending on that? If it's simpler is possible to craft a simple set of heuristics we could be taught to follow that would make things better if enough people knew them and followed them? Could they perhaps be implemented with simple road signaling targeting know problem area where jams frequently occur?

[mkl]

I'm pretty sure central coordination is not required for at least some significant benefit.

If you drive at the average travel speed steadily instead of filling gaps in front of you as soon as they open, that smooths traffic out for the car behind you as well. Traffic waves of stops and starts can still form back a bit behind you, but it's unlikely to happen immediately behind you. If even a small fraction of drivers drove like this, there would be much less opportunity for spontaneous jams to form.

I saw an impressive video demonstrating this a few years ago, but just spent a few minutes failing to find it. It showed a guy driving at the average speed of the car in front, so that the gap opened and closed as traffic in front slowed down, sped up, stopped, etc. There are articles about doing this, and animations, but what made this a compelling video was where it was filmed: a long straight concave-up road full of cars that was coming down a hill, so you could look out the back and see an long line of cars moving smoothly and not jammed, while in front there was a long line of cars stopping and starting chaotically.

[guiambros]

I think this simulation[1] explains why even a small change in traffic patterns can have huge consequences to the stop-and-go behavior.

(experiment holding one of the buses for just a couple of seconds...)

[1] http://setosa.io/bus/