Sumo – Simulation of Urban Mobility(eclipse.dev)
eclipse.dev
Sumo – Simulation of Urban Mobility
https://eclipse.dev/sumo/
10 comments
A very large subset of the population bases their commute and travel path based on best time. If using the stop sign once in 1995 was faster than the stoplight then they will always use the stop sign even if it’s almost always slower. These are the groups of people who always seem to be late.
Other groups of people are incredibly pessimistic and always take the least bad route.
Other groups of people are incredibly pessimistic and always take the least bad route.
It's unlikely that it was irrational. More likely people don't have perfect information, or it was somehow the result of poor road design.
Of course people don't have perfect information, but this town is on a 1/10th mile grid and the "happy path" was a single block away. Let's just say I didn't have to fire up TomTom on my Palm T5 to figure this out after the first time I got stuck in the backup.
Much "irrational" behavior isn't. People make decisions with incomplete information and the subset of information they have differs between people. People also have different objectives when engaging in similar behavior. Asserting that a choice is rational or not is often making the unwarranted assumption that their objectives are the same as your objectives.
For example, people often have a preference for certain flow patterns in traffic that they prioritize more than the route with the shortest time.
For example, people often have a preference for certain flow patterns in traffic that they prioritize more than the route with the shortest time.
One take away from my behavioral economics course is that its application is pretty narrowly defined, such as default settings for 401k contributions or making visible potential losses from an action.
I regularly have to take my grandmother to doctors' appointments. The path I take to drive there and back is definitely sub-optimal, but it has the advantage of being burned into my brain at this point - and it's simple and avoids many other pain points, such as an un-protected left turn at a busy intersection, etc.
It's unsurprising, customs and habits play a really large role in behaviour of crowds. It's obvious from the point of view of some individuals, but a crowd of people functions less like a cohesive and intelligent system and more like a fluid. I'd be surprised if this wouldn't hold true by crowds of people in traffic and other situations where they have to self-organize.
If they didn't simulate sustained irrational behavior, there wouldn't be people driving cars in cities.
There's a highway interchange in my city that was a 55 10-15 years ago, maybe a half mile turn, used to be 1 lane. It's now built out wider into 2 lanes and is a 70mph zone to keep traffic flowing.
To this day, people still slow down like they used to as if it were a 55.
To this day, people still slow down like they used to as if it were a 55.
In my master's thesis we used SUMO to model a small part of our town and hooked it up to the latest and greatest reinforcement learning algorithms to learn traffic light control. Eventually we beat all the other built in conventional algorithms in most parameters; Average speed. Emission. Etc.
You might be interested in Google's IRL implementation: https://sites.research.google/gr/greenlight/
I've been wanting to build a city builder using urban planning libraries like this
Imaging the simulation being running headless, decoupled from the GUI client
Imaging the simulation being running headless, decoupled from the GUI client
Slightly tangential, but there's this game: https://store.steampowered.com/app/1134710/NIMBY_Rails/
"Early Access"
"Release date: Jan 26, 2021"
actually looks really fun tho
"Release date: Jan 26, 2021"
actually looks really fun tho
You can use it headless, apparently: https://github.com/eclipse-sumo/sumo/blob/53cdfa4b595500047e...
Seems like an obvious must-have feature for research oriented simulation software like this.
People are going to sweep parameters on the cluster.
People are going to sweep parameters on the cluster.
You read my mind. City builder games are unrealistic in a bad way that defies intuition and prevents you from modeling a real city. This has been my huge frustration with Cities Skylines, which looks great but where ambulances take three months to reach patients.
I think the problem is that a realistic city builder is going to require too much processing power and would probably be unfun.
If you want traffic that moves in real time, then your day/night cycle needs to also be real time: 24 hours. That means that if you're only playing in 2 hour gaming sessions, it's going to take nearly 2 weeks to simulate a single day. For those that truly want realism, that's great. For most gamers though, that's gonna be a non-starter.
So of course there's a problem: You can either have traffic moving in real time, or you can have a reasonable day/night cycle length. You can't have both.
The compromise that city builders make ends up with what you said: Ambulances take 3 months to reach patients, even if they're only traveling a mile.
That said, I do think that Cities: Skylines could do it better. The amount of semi-truck traffic in that game is absolutely insane. Easily 4x what it needs to be. You make an industrial district that's only ~1/2 of a square mile and it's an absolute zoo of semi-trucks, requiring some crazy traffic engineering to make it not be a gridlock.
If you want traffic that moves in real time, then your day/night cycle needs to also be real time: 24 hours. That means that if you're only playing in 2 hour gaming sessions, it's going to take nearly 2 weeks to simulate a single day. For those that truly want realism, that's great. For most gamers though, that's gonna be a non-starter.
So of course there's a problem: You can either have traffic moving in real time, or you can have a reasonable day/night cycle length. You can't have both.
The compromise that city builders make ends up with what you said: Ambulances take 3 months to reach patients, even if they're only traveling a mile.
That said, I do think that Cities: Skylines could do it better. The amount of semi-truck traffic in that game is absolutely insane. Easily 4x what it needs to be. You make an industrial district that's only ~1/2 of a square mile and it's an absolute zoo of semi-trucks, requiring some crazy traffic engineering to make it not be a gridlock.
With those timescales in mind, people must go decades between annual checkups.
How much do the various "Maps" apps change things? I have a longer commute, and when the freeway is clogged, Maps will direct me to an exit where I weave around town and country. There's usually a convoy of cars with me, but the freeway also seems to stay clogged.
This looks really polished. I've always found crowd and traffic simulation fascinating.
The Projects page is worth looking at too.
The Projects page is worth looking at too.
This is fascinating. Even supports simulating multiple modes of transportation (ped, bicycle, car…).
I’ll have to give this a test run later.
I’ll have to give this a test run later.
> Even supports simulating multiple modes of transportation (ped, bicycle, car…)
Anything serious needs to include those - what would Amsterdam look like if all those people stopped riding their bikes and got in cars? It'd be a disaster.
Anything serious needs to include those - what would Amsterdam look like if all those people stopped riding their bikes and got in cars? It'd be a disaster.
yup - I live in the US and used to be involved in promoting a diverse range of transportation options. In meetings with state transportation departments and reviewing their simulations they often lacked this perspective.
The state/city engineers only cared about efficiency of car traffic. bike and ped infrastructure is not included or meet bare minimum spec (ie, "share-ow" lanes/paint/signs on busy streets as a very poor substitute for bike infrastructure)
The state/city engineers only cared about efficiency of car traffic. bike and ped infrastructure is not included or meet bare minimum spec (ie, "share-ow" lanes/paint/signs on busy streets as a very poor substitute for bike infrastructure)
I ride rental scooters almost 10k minutes per year and would really like to get my hands on my own ride data to plug it into something like this (or simpler) to find the optimal routes for my regular trips.
Google Maps (or others) works good to find a resonable route, but I can do better on my own. One-way streets where bikes are allowed to go do opposite way is sometimes missing, short desire paths connecting bike ways, crossings where it's safe to do an (illegal) right-on-red etc.
Tried a GDPR data claim from Voi but got nothing back :( But I hope the data is somehow available for urban planners, think it would be a great source of truth to use in tools like this.
Google Maps (or others) works good to find a resonable route, but I can do better on my own. One-way streets where bikes are allowed to go do opposite way is sometimes missing, short desire paths connecting bike ways, crossings where it's safe to do an (illegal) right-on-red etc.
Tried a GDPR data claim from Voi but got nothing back :( But I hope the data is somehow available for urban planners, think it would be a great source of truth to use in tools like this.
Unfortunately it is not that easy to simulate traffic, especially not on city scale or larger.
The most important input to a traffic simulator such as this is the so-called "traffic demand", i.e. the routes that vehicles follow. Typically this is provided in the form of origin-destination matrices, but this data is not freely available.
Next up is the way in which traffic lights work. Reality is very hard to model here, again because the data is not freely available.
And then, due to numerous modeling errors in vehicle density, in the way that roads differ from e.g. OpenStreetMap, and how traffic behaves, the simulations are highly unrealistic, unless one spends some time to calibrate it.
It costs quite a lot of money to set up a realistic simulation, and most governments use commercial tooling that is easier to use, such as Vissim or Aimsun.
The most important input to a traffic simulator such as this is the so-called "traffic demand", i.e. the routes that vehicles follow. Typically this is provided in the form of origin-destination matrices, but this data is not freely available.
Next up is the way in which traffic lights work. Reality is very hard to model here, again because the data is not freely available.
And then, due to numerous modeling errors in vehicle density, in the way that roads differ from e.g. OpenStreetMap, and how traffic behaves, the simulations are highly unrealistic, unless one spends some time to calibrate it.
It costs quite a lot of money to set up a realistic simulation, and most governments use commercial tooling that is easier to use, such as Vissim or Aimsun.
Did my PhD with that - good/hard times
Any plans to deploy to the web?
Not to be confused with Suno - Simulation of Musical Ability :)
Example:
For over a decade, the freeway on-ramp nearest my work had two main ways of getting to it from downtown. One of them involved a stop-sign crossing a road that had the right-of-way (i.e. a two-way stop). The other had timed traffic signals. Every evening around 5pm, the traffic would backup from the stop-sign for multiple blocks. Meanwhile the route with lights was completely smooth.
Eventually the stop-sign was replaced with a signal, but I marveled at how many people persisted in making their daily commute much worse than it needed to be.