This model is overlooking some aspects of traffic flow. The most important one is that flow does not increase and decrease smoothly. As density increases there starts to be shock waves that bounce longitundinally[0].<p>It would also makes sense to also distribute some autonomous vehicles along the train to minimize these shock waves. A much simplier solution is to autonomously centre between the vehicle in front and rear[1]. Doing so smoothly is simple local logic which may improve flow rates dramatically but I don't know if it's been implemented anywhere.<p><pre><code> [0] https://www.youtube.com/watch?v=Suugn-p5C1M
[1] https://www.insidescience.org/news/fluid-dynamics-explains-some-traffic-jams</code></pre>
On my commute all of the possible routes are congested. I can be as selfish as I like, but I won't get home any faster. I think we need more roads and more pubic transport and more work-from-home schemes. Autonomous cars may may allow me to chill a bit more in the gridlock I suppose...
Funny what apparently simple differences in mathematical models can imply.<p>I studied an old-school "Wardrop model" of traffic flow in one class in school. In that model, your speed along a road segment is a function of the flow (vehicles per hour), and it's usually assumed to be monotonic (or at least non-decreasing.) In the article, the images <i>never</i> show increased throughput coinciding with longer travel times.<p>In the Wardrop model, equilibrium traffic levels are unique, and in TFA's model they aren't. (The diagrams make this clear in a really nice way, I like them a lot.)<p>I wonder whether/when each model is a better description of reality. It's hard to imagine the road in that Office Space clip demonstrating especially veh/hr flow...
In my reading, this says that autonomous cars, if they all behave cooperatively, can be more efficient road users than manual cars driven by individuals acting selfishly.<p>While it works out in the Nash equilibria they describe, I don't see it working at any sort of scale. It's too easy to defect from the better equilibrium: drive the car yourself, or buy that aftermarket app which disables the cooperative driving mode.
I drove to work for a while and it was awful. I wished that there was some way to tell everyone what the average speed of the road currently was and limit their speed to that. Instead I found that people moved their cars forward to the next place to stop as quickly as they could every single time. It was infuriating.
What actually happens with human drivers is this: everyone wants to get home/work so the distance between the cars will decrease below safe levels when more cars are added to the road.<p>So the breaking latency of the autonomous cars must be better than this to fit more cars onto roads.
> <i>...when a modest portion of the autonomous cars are altruistic, it is possible for majority of the cars to use the shortest route without causing congestion. This can significantly (in fact, unboundedly) decrease the total latency.</i><p>how could it decrease the total latency unboundedly? I mean, doesn't the lower bound have to be zero?