11–20 of 143 posts
Perhaps these questions are appropriate for this post. I'm beginning to delve into this space and I've not managed to get any satisfactory answers to these questions, despite my month(s) long search. 1) How does one build a traffic (i.e state-space system), forgetting the visualization aspect of things ? Just generating sparse matrices, and adding elements to interact (add/subtract) from these matrices would be a gre…
Regardless of how I change the map layout, traffic in/out flow or truck/car ratio I seem to get the best results when I crank accel, speed and right bias for trucks to the max and lane-change threshold and following distance to the minimum. Which makes sense since those are basically the ideal conditions for reducing the number of obstructions low speed (merging, exiting, trucks on grade) traffic poses to higher spee…
You can almost always fix the traffic by maxing out acceleration. Sadly, this vindicates my real world aggressive driving behavior.
Perhaps these questions are appropriate for this post. I'm beginning to delve into this space and I've not managed to get any satisfactory answers to these questions, despite my month(s) long search. 1) How does one build a traffic (i.e state-space system), forgetting the visualization aspect of things ? Just generating sparse matrices, and adding elements to interact (add/subtract) from these matrices would be a gre…
This doesn't exactly answer your question, but what I settled on during my dissertation (granted, that was a few years ago now) was SUMO: https://www.eclipse.org/sumo/ IIRC you could import maps from from open street map, but I'm not sure if it has a "headless" mode, without all the visualisation.
Regardless of how I change the map layout, traffic in/out flow or truck/car ratio I seem to get the best results when I crank accel, speed and right bias for trucks to the max and lane-change threshold and following distance to the minimum. Which makes sense since those are basically the ideal conditions for reducing the number of obstructions low speed (merging, exiting, trucks on grade) traffic poses to higher spee…
You can almost always fix the traffic by maxing out acceleration. Sadly, this vindicates my real world aggressive driving behavior.
This way, agents would adapt their attributes to maximize their own local advantage, and therefore could see how large scale trends develop. I've driven in many countries and regions within countries, and it's just wild how "driving cultures" vary so much. Some places the drivers are so polite to the point it really makes things dangerous, other places the drivers are vindictive (distinct from being aggressive driver). It would be interesting to model how driving cultures emerge.
Edit: I guess the short story is I'm really curious why some places, like Bangkok driving is chaotic, dangerous, gridlocked, and no one adheres to traffic laws much or yields to pedestrians, but relatively devoid of road rage (and people will let you merge over if you need to). In DC, people carefully adhere to red lights, stop signs, pedestrians, etc but will go far, far out of their way to block you from changing lanes once they see your turn signal go on.
Also, exploring the game theoretic (probably prisoner's dilemma?) aspects of tailgating and how that seriously depends on the driving culture. When someone tailgates, the tailgatee can slow down (everyone loses), speed up or move out of the way (tailgater wins/tailgatee loses), etc. If everyone slowed down when tailgated, there would be no benefit of tailgating; if enough people give in when tailgated, that keeps rewarding the behavior. Anyway, that would be something fun to explore in one of these traffic simulations.
Looks like maxing out acceleration solves all other problems. I guess we should require a minimum power-to-weight ratio for cars from now on.
Looks like maxing out acceleration solves all other problems. I guess we should require a minimum power-to-weight ratio for cars from now on.
Things like this are cool. What I would be curious to see though, rather than setting global parameters (i.e., politeness), would be to have each agent have its own politeness score (as well as other attributes). Models like the one linked to here work as if each agent is a particle, rather than autonomous decision-maker adapting to their circumstances locked in long-term, iterated games with other anonymous drivers.…
> Some places the drivers are so polite to the point it really makes things dangerous
I've known people like this, who will approach a 4-way intersection and stop even when they don't have a stop sign and the perpendicular street does.
Looks like maxing out acceleration solves all other problems. I guess we should require a minimum power-to-weight ratio for cars from now on.
In practice that would mean a lot more head/tail accidents, I've seen that happen plenty of times in traffic jams from over-eager drivers, sudden stops (either people trying to go too fast or not paying attention), etc.