Interesting project! I had fun playing with it. I wanted to test the theory (which I saw in an article posted here on HN a few weeks ago) that a single car slowing right down on a busy highway (e.g. to avoid hitting an animal on the road or similar) can actually cause a traffic jam that goes back miles and can last for hours.
I couldn't prove that hypothesis at this scale. I am assuming that the AI drivers have just too good a reaction time and they can co-ordinate their speeds to each other to keep things moving along fairly nicely. A thumbs up for robot driverless cars, I guess ;-)
EDIT: Actually, playing some more - it does prove it to some extent. If you increase reaction time to ~800ms and let the traffic blitz along at a steady clip, then slow down two cars side by side to almost a stop, then after the initial cascade of accidents, it take ages for the traffic to get back to the same steady pace. Interesting.
What is the number for the Simple Highway Police Department, I just witnessed 3 hit and runs back-to-back and would like to report them!
Neat idea and while it's still at a very basic stage it'll be cool to see this evolve into something that could eventually be used by traffic controllers!
What I'm missing is right lane preference, that would control how much people obey the rule that you shouldn't be on the leftmost lane unless you are faster then the cars in the right lane in front of you and if you aren't much faster then the car that's next to you on the right lane you should slow down and place yourself after it or speed up and place yourself in front of it.
It's horribly painful to watch two slow cars occupying both lanes next to each other.
Or indeed left-lane preference, for British and Japanese users.
But yes a more sophisticated simulation would use > 2 lanes, and have various controls for lane discipline, especially since poor lane discipline is a major contributor to congestion. This is a suggestion not a criticism btw, I had fun with this.
Some bugs (cars not always breaking when another is standing still ahead; cars crashing at the beginning if you hit launch too quickly), but overall a fun toy to play with. I would like to know what realistic values are though: I know that 200ms reaction time is a good case, but what are actual average cases? It should be set to that by default. Or acceleration speed, is 5km/h realistic? 10? When set to 5, I cannot simulate a normal highway (they'll never go over 100) because the track is fairly short.
I think the braking is a feature not a bug. See the "reaction time" I'd assume that means that some just aren't reacting in time to not hit the car ahead.
That might be, I don't know when they would have started to break but they were extremely late if ever. But perhaps it was well within the reaction time I set (350ms), for me it's easy to think they just kept driving because I saw them coming from afar.
From default settings, set speed variance to 50%, launch frequency to 120 cars/minute, and reaction time to 500ms. Unsurprisingly, this is chaos.
Then, keeping those settings, clear the cars and set the target speed to 150km/h. The traffic mostly clears up! Since the cars are going faster, they leave the road sooner, meaning fewer are on the road at a time, fixing most problems.
(Of course, in real life, this also makes the accidents that do occur worse.)
Something this sim doesn't model that plays a large role in real accidents is the speed differentials during merges. If an offramp moves too slowly and gets fully congested, it starts spreading onto the highway, compounding the problem.
Playing Cities: Skylines I've noticed a similar problem when I try to use roundabouts. They're idealized as a way to keep traffic flowing in busy intersections by imposing no stops, but if they surpass a critical mass of congestion, the whole thing stalls, creating a problem worse than stop lights.
(The Cities traffic model is pretty close to how a robocar might behave, as they are polite drivers able to execute apparently dangerous merges with 0% chance of collision, but also have the problem of not detouring when the ideal route becomes congested. As a result, you have to design your city to be decent at diffusing traffic early and not rely on wide roads to do that job.)
Have they fixed the problem where that 'ideal route' even includes not merging left to pass around a stalled lane? It's been seemingly forever since I built a city.
Its principles also hold up for drivers in China. Guess the chinese love copying american things right down to driving just as badly :-D
Anyway, jokes aside, this has a lot more uses than proof we are all bad drivers. I can this being a great simple way to show queues in a factory/shop/office coffee machine . Awesome work and i wish i as 5% as good at coding to be able to do this when on holiday. On my hols i would be lucky to get past hello world. Kudos to this guy.
This is cool! I always wanted to make something like this for intersections and randomize the response time in an effort to prove whether or not people who pull out slower at red lights cause others to miss the lights or not. Just a neat experiment as well!
I'm most interested by the umlaut in the domain name that apparently isn't respected by HN. Also a way to restrict your domain name to certain keyboards (easily)?
Yes, please! I suspect lack of lane discipline to be the single biggest disadvantage US freeways have to American's. I would love to test that at least via simulation. I would expect that two lanes with lane discipline are better than three without. Potentially even better than four.
Nice. I've wanted to do something like this, as I have a theory about the car pool lanes being worse for traffic. When traffic is bad, car pool lanes are just as backed up typically.
When someone wants to get in the car pool lane upon entering the highway, they tend to do so fairly quickly. This can cause a shockwave as someone flies from one side of the highway to the other. The same thing can happen in the opposite direction when a carpool lane driver needs to get off, and this issue is compounded even more since you can't always (legally) exit the carpool lane.
So I thought it'd be interesting to play with a simulator like this, add/remove carpool lanes, or lanes altogether, and see what results.
In the uncounted thousands of hours of driving on the freeway in my life thus far, I've observed:
* Drivers will improperly idle in a lane they feel safe in, refusing to keep right except to pass (this is actually state law where I live).
* Any deviation from a straight, flat, roadway (corners, hills, dips, anything) will cause the flow to slow; even when you can proxy the status of the roadway in front by the tail-lights of all the lanes of cars who /can/ see.
* Apparently no 'big' city anywhere balances jobs and housing. Cities too small to balance anything are also generally too small to notice the issue. (I say this because it seems like everyone is economically incentivized to drive as far as they can possibly stand from work for the cheapest housing; clearly there isn't enough housing near jobs to keep those costs competitively low.)
On your first point, that's the norm in my driving (most of both US coasts.) I've sometimes wondered what the saturation level is where the left lane should just be another lane to increase steady state flow rate. I guess it's when the lane becomes too crowded to cycle properly.
On your second point, I've always just assumed that that was a tall car thing because I figured they probably get more uncomfortable cornering. The same way I probably seem irrational when I speed around pickup trucks: they probably fail to recognize that their driving in front of me is like putting a wall in my face (and unlike semi trucks the drivers are no better than the rest of the people on the road.)
On your third point, the US has baked in too many policies that are prohibitive to sensible development patterns. The implicit subsidies for low density living and the federally insured mortgage cone to mind as two of the worst.
A bug in this simulator is that new settings apply only to newly-launched cars, incrasing the chaos as you explore a range of options. Increased speed shows an exaggerated immediate increase in crashes, decreased speed causes an exaggerated decrease in crashes, and so on.
The sensible conclusion here is that human drivers could be banned in order to safely minimize travel times (much faster reaction times, lower variance in speed).
What about prioritizing vehicles? Imagine - if all cars drive along in a well-defined path, will we end up with making assassinations easier? That's a stretch (I mean, not like exceptions couldn't be made for POTUS and such), but I wonder about what sort of consequences would arise from having predictable traffic.
Simple Highway Collision Simulator. Could be a great little game with an educational bent. Personally I set reaction time to 0 and jacked everything else to max. Then I spammed the launch button.
I couldn't prove that hypothesis at this scale. I am assuming that the AI drivers have just too good a reaction time and they can co-ordinate their speeds to each other to keep things moving along fairly nicely. A thumbs up for robot driverless cars, I guess ;-)
EDIT: Actually, playing some more - it does prove it to some extent. If you increase reaction time to ~800ms and let the traffic blitz along at a steady clip, then slow down two cars side by side to almost a stop, then after the initial cascade of accidents, it take ages for the traffic to get back to the same steady pace. Interesting.