Configure an example with bad human driving --> better robot driving
cathywu opened this issue · comments
In the other plots, we have separate examples from separate simulations or a single simulation that shows poor robot driving --> somewhat improved robot driving. The goal here is to demonstrate a more dramatic improvement.
Preliminary example with ACC and 85 vehicles. The tau is reduced from 3 to 0.5 at the midway point.
Whereas if tau=0.5 for the entire simulation, then it looks like:
So this is a good start, but why can't the robot vehicles improve the first scenario up to the level of the second?
Perhaps with a larger reaction time (tau), the capacity of the loop is just lower. With 55 agents instead, we see:
So what's the largest number of vehicles such that the humans can achieve 25m/s?
Looks like 20 vehicles.
A full switch from human (no controller, just default behaviors) to ACC halfway:
With humans with random lane changes (randomChangeLaneFn):
Sweep of varying proportions of humans + changeFasterLaneBuilder vs robot + ACC:
https://www.dropbox.com/sh/ikc4skzpifhqxtx/AADZfROr-xRViTbW-6TC9Yrda?dl=0