Microsimulation of Traffic Flow
traffic-simulation.de
traffic-simulation.de
I think it has more to do with the suddenness of the stop; each car behind it has to stop suddenly in response, as the change only spreads locally. I’m pretty sure all you need to modify is deceleration speed for braking and acceleration speed to catch back up, to see like 90% of the effect
I’d test it with the simulation but it was too difficult to setup on mobile :-)
Yes. And simply maintaining good following distance gives you a buffer that can absorb this sudden change and prevent the tailback. This is one of the reasons that I long for the day that humans don't pilot cars on the highway.
To-date, if you see a sign saying "40", that basically means there's no traffic flow and you will have to come to a stop very shortly.
If you see a sign saying "60" that means that you probably won't need to stop but you're going to slow to a crawl.
If everyone just did as it said things would probably improve massively.
To some extent this is my confirmation bias, but certainly as often as not the signage is spurious and that's why people don't pay any attention to them. I suspect most people agree with the principal of the things, but every time they obey a sign whilst everyone else flies past, only to discover it was inaccurate it does affect the amount of credence future signs are given.
Essentially you are doing a "smoothing" operation if you proactively reduced your speed ahead of a jam.
Even with regenerative brakes, the regeneration isn't fully efficient. You lose energy to heat. So smoothly traveling one speed is superior.
I respect the idea, but the actual implementation feels a bit half hearted to me. They need to kick in a lot quicker and be more severe, all they seem to do is make a minor improvement and slow people down perhaps a half mile from the actual congestion.
So if the cameras work, we'll have to do exactly what they say or get fined for speeding.
The ones I was referring to are advisory and look like this [0].
[0] https://www.drivingtesttips.biz/images/motorway-advisory-spe...
All / most / some people? Or just the person you responded to?
Just look at politics / marketing / dark patterns / mortgage contracts.
I successfully changed my sister's mind from variable rate mortgage to fixed mortgage, as the language of the contract doesn't explicitly write how bad the worst case option can be 10 years from now for variable rate mortgage.
https://mil.brandenburg.de/cms/media.php/lbm1.a.2239.de/stud...
Please tell me of one road/route in Germany where you can drive '200 km/h for hours straight". I'll settle for tiny values of 'hours', as long as they're >= 1.
For example when Swedes sees a lower speed sign or even just a warning before a road block they will slow and stay in one long lane, while in the mid eastern country I live in until recently everyone just speeds up trying to be first to pass the blocking. The results ? smoother flow and shorter queues, guess where ?
We can try to smooth out this ripple, but if the congestion situation exists for long distances, all it will do is push back the ripple to the end of the congestion.
If people can't follow the process maybe the process is wrong.
Both are valid points one could argue for, depending on underlying values that are not necessarily rational.
It works surprisingly well, both through modulating the speed of traffic and by allowing the hard shoulder to be used at peak times.
Wouldn't it?
If everyone is going 70 mph, it's quite safe. It's the one moron going 50 mph or the one asshole going 90 mph that will cause the crash.
As for wearing out brake rotors prematurely - even with aggressive braking they last more than 50k miles. So on average you'd replace them less often than every 5 years, and they cost around around $250 for a set of four. Doesn't seem like a very good get-rich-scheme for the mechanic.
Perhaps more surprisingly, /reducing/ comf(ortable?) deceleration keeps things running smoother than higher values.
This is basically what the dynamic speed limits do on motorways in the UK - if there's congestion, the speed limit is reduced on the section of road leading up to the congestion. It reduces the number of cars flowing into the traffic jam temporarily, which helps to smooth out the overall flow.
[1] http://vision-traffic.ptvgroup.com/en-us/products/ptv-vissim...
For example, how much time could be saved if at red lights, when the light changes green, if all cars accelerate at the same time and speed, instead of the accordion behavior. This is something basic driver AI could handle and might help with pollution and traffic, while fully autonomous driving is still underway.
I wonder why a "SimRoads" hasn't been created?
A mashup of this road microsimulation and something like TORCS might be a fun thing to play with, both from the "design" and "simulation" perspectives.
There is no easy answer. Aggressive driving = bad. Tailgating = bad. But see what happens when you crank up the acceleration and decrease distance between cars. Then do the opposite. Simple little simulations like this teach us that bad driving is actually good for traffic.
One very interesting lesson is that speed seems largely irrelevant to traffic flow. So long as people don't leave gaps, that they all follow each other closely and accelerate/brake like racers, net flow is divorced from speed.
Aggressive driving and tailgating are bad because humans are not capable of making good choices at those distances/speeds. If you could 10x our reaction speed, then those actions would not be as looked down upon (going 6-8 over the speed limit is technically illegal most states, but nobody on the road is going to think you're some kind of maniac).
Simple little simulations like this teach us that simple little simulations do not account for reality and the complex nature of the meatbags in control of each vehicle. A more realistic simulator might be able to roll the dice that a driver might not notice an obstacle in time, and crash straight into it - putting more debris and obstacles on the road for every other driver to avoid/crash into[1]. The gains in net flow from have everyone drive like madmen would probably immediately be wiped out from the hour(s) spent getting emergency vehicles onto the scene, redirecting traffic, treating the driver, towing the vehicle, etc. Not to mention that there would almost certainly be multiple scenes down the road - as there usually already are during rush hour - and we already condone this type of behavior!
Accidents do happen, but are actually a very small part of traffic jams. They are the rare case that do cause a big onetime backup but don't play a role in the daily slog. Eliminating every accident would not result in traffic moving appreciably faster on a daily basis.
This obviously can only work if necessary protocols get adopted. Industry might do it themselves, otherwise passing such regulation should be quite doable... little opposition from voters due to AVs mostly owned by big fleets but a significant prospective quality of life improvement.
Simulators like this (and seeing ideal simulation behavior deviate from correct real-world behavior) concern me re: the coming mix of AV and human drivers.
Second, removing all of the politeness has an interesting effect. I think I can learn some lessons from this, thank you!
Third, they should use this in driver education classes.
Personally though I’ve always wanted to make a genetic algorithm implementation that models this but where card can evolve when and what signals they give and how they respond to those signals. Fitness function would be speed to get to destination, crashing would be a failure.
I think it would be interesting to see what happened
https://www.google.com/maps/@48.4324714,13.0948723,637m
It was built about 20 years ago, and I was living in the area back then.
Before the "roundabout" was built, being a traditional intersection it used to be a major accident hotspot, despite having a speed limit of just 50km/h. Crossing the main road was nightmare, I can tell you that.
With the new topology, which is a very elongated roundabout, the speed limit on the main axis could actually be raised to 80km/h yet accidents now rarely happen there and traffic flow is much improved
Yet another example of LCD governance imposed by the crazy minority - in this case, those who would not yield correctly (see Skin In The Game, NNTaleb). Always remember that 50% of drivers are below average intelligence.
The purpose to blocking the view is to make traffic flow better. There are many people that will wait for any vehicle in the circulating lane regardless of that vehicle blocking their path. By blocking the view across the central island, those entering the roundabout can only see so far upstream as matters to their flow.
Stopping people from traversing the central island may not be a large issue in an urban environment but in rural settings that is different. The state I live in is very rural and we have some roundabouts that are not near a water supply for irrigation. As such the state DOT tends to use just a mound of gravel (roughly 3/4 to 1 inch aggregate.) It is very common to see tracks going right through the central island, likely lifted trucks thinking it is awesome. Having landscaping in the central island ads something to hit.
Well...technically not in this case, no.
It's usually misleading to use the median instead of arithmetic mean for measuring averages in population, because you can wind up with situations where, say, 90% of people are "above average."
That being said, IQ is specifically calibrated to follow a normal distribution. Ordinarily that would mean you're correct, and we can just call the arithmetic mean and median the same thing.
However! The normal distribution of IQ - average 100, standard deviation 15 - includes the bottom percentiles of the population who are severely mentally impaired. These people aren't usually drivers. It follows that when you're driving, the population of drivers around you will actually skew higher in percentile. In fact, it's probably more likely that the modal driver has intelligence equal to or greater than average.
Personally, I think safe driving, polite driving and technically skilled driving are all orthogonal with respect to each other and to overall intelligence. After a very serious accident years ago, I became the kind of person who will only drive at or below the speed limit. But I know several people at least as - and probably much more - intelligent than me who drive "erratically", to put it nicely. I don't think raw intelligence captures enough information to really dictate behavior here.
Besides that, many after-the-fact roundabouts are so small in diameter that the indicators on almost every car model are invisible, so there is essentially no way to actually tell if someone is taking an exit until they do it. Lots of unnecessary stopping and waiting due to that.
Another great thing is having a zebra crossing (preferably combined with a cyclist crossing) on each approach/exit, but planting a bunch of bushes such that it is practically impossible to see people there when approaching the exit, forcing people to crawl speed before exiting, and conversely forcing everyone behind them in the roundabout to brake as well.
I'm not sure if stuff like that is intentional or just accidentally stupid.
When an intersection light turns green - all cars should begin rolling together. Waiting for cars in front of you in succession is burning time.
Human nature is to pull as far forward as you can, within certain limits at least. But is that actually optimal? If not, with self-driving cars or some other technological assistance, maybe we could increase the efficiency of intersections that use signals.
I suspect it might be good for getting through that intersection faster, but in general I think it’s probably best for all traffic if stopped traffic takes up as little room as possible.
Even for long blocks, you'd want to figure out that the light will be green for X seconds, which allows Y cars to get through, so only Y cars (and no more) would line up with the required spacing between them, and the ones behind that would pack in more tightly to save space. Then while the light is right, the next Y cars could get themselves into the correct configuration.
I haven't done the maths, but it wouldn't surprise me if they're about the same.
Data Transfer allows for instant start/stop so the car can actually stop before the light turns, allowing the cars on the other side to go quicker and go immediately.
In the future, we may not even see self driving cars stopping at red lights (think of all intersections as four way roads)
Will not work. The safe distance between cars is negligible in traffic lights but a lot more in moving traffic. At least two seconds is considered reasonable although in reality much lower distances are seen.
The "rolling together" would only work if the queue to the traffic light would copy the distance needed by moving cars.
That is, the cars would be spaced 10-40 meters apart while waiting for green. Only then all cars could floor it and accelerate together without having to wait for the first units to run away first.
Almost...as if...they were mechanically linked...
Let's say we are in an area with a 60km/h speed limit.
Going 60 km/h you do need a wide gape between cars.
But once you stop at a traffic light you should leave enough of a gap to see where the types car in front of you touch the road. That way you are not too close to get around the car if it gets stuck.
When the light goes green, lets say the first car accelerates slowly and only reaches 30km/h by the time they cross the traffic light. If every car behind waited until there was 1.5 car length gap, then only half the cars would get through compared to if they all accelerated at the same speed at the same time, maintaining a smaller gap until they cross the traffic light.
You only need the large gap once you've accelerated up to a much higher speed.
Going from small gap while standing in lights to large gap at a higher speed cannot be reached in unison.
The later cars must inevitably accelerate slower or the first cars must accelerate faster in order for the larger gaps to appear. So even if all cars did in fact start rolling at the exact same time when the light turns green the acceleration of any tailing car is bounded by some safety margin as a function of speed.
I agree. But my point is that people can maintain the small gap at lower speeds until they cross to the other side of the traffic lights. (So that more people can get through) then they can spread out.
You can see clearly from the simulation that having the opportunity to move around an obstruction doesn't stop the ensuing slow down from occurring because you have to slow down and reduce the bandwidth of the road to move around the obstruction.
When you're stopped you don't need a buffer so much for when things go wrong, but at higher speeds you need a bigger buffer, and having space helps avoid mass catastrophes when some freak storm knocks a tree branch onto the road or a car malfunctions or mechanical failure/wear and tear.
https://www.scania.com/group/en/platooning-automated-driving...
https://platooningensemble.eu/news/using-its-g5-for-efficien...
Agreed. I've always suspected/known this. Just play with the acceleration settings in the simulation for proof.
I hate when the light goes green and people wait for a gap before driving. Just go, you are barely going 20km/h at this stage.
If everyone accelerated quickly until they were on the other side of the traffic light, you could get double the amount of cars through. No need to break the speed limit.
If they didn't move, then obviously it's retarded to jsut start driving.
My point is when the car in front of you _does_ start moving, you need to move too, at the same speed/acceleration. Not wait.
The default examples are good too. The kind of bunching you get due to merging traffic onto higher speed roads, and the kind you get from traffic slowing down to exit on a slower speed road, both of these are great examples that come up in the real world very frequently.
I always wonder if, at busy pedestrian intersections, it would make more sense to have the pedestrian flow stages and the vehicle flow stages be totally separated. Cars waiting to turn right for ~30s as pedestrians walk is kinda crazy.
New York uses Leading Pedestrian Intervals, which give pedestrians the green before the cars can turn right; by the time a driver starts turning pedestrians are actively crossing the street and no longer in the blind spot. Another good way to increase pedestrian safety is just to ban right-on-red altogether.
- Microscopic simulator: https://sumo.dlr.de/index.html
- Flow (abstraction of RLlib and Sumo): https://flow-project.github.io/
Have a look at Sumo, its free, open source, and comprehensive. Also Aimsun [1] but I've never used it and don't think it's free although I'm unsure about that.
Interesting to see this in practice and to just give the rule "ring has priority" for granted.
Do they use Poisson distribution, by any chance?
These are the 'best' settings I could find (almost 500k veh/h): https://i.imgur.com/eqM8nbb.jpg
To reproduce, set the timewarp to 0.2 and density to 0 then wait for 30 in-game seconds.
Always seemed really pointless.
Turns out that was the only way I could get traffic going again on the curved road with the on-ramp (even though I could only control the main traffic, not actually the on-ramp). Even dropping the speed limit had no effect on clearing the congestion at the junction.
They make a lot of sense. Traffic entering an on-ramp will usually come in bursts from traffic light. A burst of cars will have a much harder time merging with traffic than a trickle. So rather than creating, for example, a group of 20 cars every 60 seconds trying to merge at once, the on-ramp traffic light throttles that to 1 car every 3 seconds. Total throughput is the same, but the effect on existing highway traffic is significantly reduced.
As highway-driving AI approaches perfection I'd expect:
- Acceleration/nausea approaches zero
- Number of lane changes approaches zero
- Distance between cars approaches zero (when necessary in crowded areas)
- Maximum speed approaches physical limit
It will be satisfying to watch these happen over time.
edit: formatting
I've made something simpler yet similar way back (hit "refresh" after the initial load, sorry for that bug):
http://sinelaw.github.io/jammed/src/
In my simulation each car has a "personality", which affects lane passing, following distance/time, speed, acceleration, etc.
Did it as an exercise in JavaScript + HTML Canvas programming...
Being able to adjust the min/max and standard deviation of traffic speeds would be nice too.
You can set the maximum speed to the max. While this doesn't directly control on-ramp flows, the cars already on the highway will be moving much faster than cars on the on-ramp due to acceleration/distance constraints. Then a jam ensues
On the other hand, there is no lane switching when traffic on adjacent lanes clears up. In real life, drivers will quickly switch if a nearby lane starts moving. They can't see whether their own lane is also clearing up ahead, and most people won't take any chances.
Of course, for that to work every car has to trust the data it gets from every other car, so there's no place for bugs or user-modifiable software in such a system. And the stopping distance is the worst of any car in the platoon.