Traffic flow measured on different 4-way junctions [video] | Hacker News Reader

Traffic flow measured on different 4-way junctions [video]

youtube.com

67 pointsltomic26 comments

This is from the game Cities: Skylines, which is basically the modern version of the old classic SimCity.

I think it would be good if the title made that clear, and didn't pretend this is some serious simulation. I mean, it's very cool and detailed for a game, but as you can tell from the video, it's far from perfect. It doesn't care about collisions, just about a rough approximation of traffic flow.

Yup. I'd also like to see a couple other figures alongside traffic flow like, "total square footage of structure", "collision rate", and "number of displaced residents."

Cost might be a good one too.

Cost can be very misleading, since most of the cost of big projects like these tends to be land acquisition.

For example, roads are on average much cheaper than rail lines per lane mile of construction, but lots of roads get built in rural or depopulated areas where land is cheaper, whereas rail tends to be built in denser corridors.

Yeah, I'm sorry the title is a bit misleading, I thought of it mainly as a motivation for a more serious implementation of traffic simulator.

Right from the start it is obvious this "simulation" is not a full realistic one. Behaviour of vehicles and amount of them is not realistic and it is not clear what the score represents.

Here's a great game that is just junction design:

https://captaingames.itch.io/freeways

You are a traffic engineer.

Draw freeway interchanges.

Optimize for efficency and avoid traffic jams.

I loved this game. As you complete boards, you get to see them all link up and can watch cars drive all over your city.

In the simpler variants it seemed like some cars passed right through the middle of semis or were my eyes deceived?

I don't think your eyes are deceiving you, I saw the same thing. My guess is (and excuse my incorrect terminology) that the hitbox and the mesh aren't the same size, so the cars are passing after the hitbox (maybe the cab) safely, and thus through the trailer mesh/texture.

It's a cool game, but it's meant to simulate an entire city. It only makes sense that you lose accuracy in details. It's just meant to look somewhat believable, and make traffic design in the game meaningful.

There is no collision detection here, and consequently no hitboxes. It’s just a visualization of some simple game simulation data

There has to be some though right? Otherwise there wouldn't be any back-ups. I guess they could be point masses that can't occupy the same space of another point mass + some radius around it.

You just calculate the total load of the road and render enough cars appropriately with defined positions. Doing that logic through a physics system would collapse the entire simulation.

In case anyone is confused (like me), the video is an simulation made using mods for the game Cities: Skylines - http://www.citiesskylines.com/

Here, two very much related papers, albeit the title of the first one wouldn't tell you that:

https://arxiv.org/abs/1612.07888 Kurauskas, Valentas - On the genus of the complete tripartite graph Kn,n,1 [2017]

https://arxiv.org/abs/1801.03860 Kurauskas, Valentas; Šiurienė, Ugnė - Symmetric road interchanges [2018]

And slides from 2010:

http://web.archive.org/web/20101127223543/http://www.saga-ne...

That "turbo roundabout" really got me. I feel like for that example (and probably others) you'd also need to measure the quantity of efficiency lost to people just trying to understand how the system works and maybe then the actual score would be lower.

In the same vein, for the few complex systems at the end, they should probably factor in human mistakes in the score because it looks like it would be incredibly time consuming to get back on the right way if you took the wrong turn (and could potentially add traffic to the equation).

My town just recently put in the Diverging Diamond Intersection (or at least something very similar) on a major highway and major road interchange and the extra throughput was very noticeable.

I'd like to see it model Powderhouse Square in Somerville, MA... traffic circle with red lights going both in and out...

I wonder what the theoretical maximum throughput is? Anybody know? The video got up to 1000, which seems pretty high.

It seems like, from that simulation, that the longer the curve, the higher the potential throughput. I particularly liked the Turbine.

I think it had more to do with minimizing the number of lane merges, which just so happen to correlate with long curves

The theoretical maximum would be all vehicles operating at highway speed limits of perhaps 70MPH, non-stop through all points, bumper to bumper, packed as densely as possible, according to a profile for vehicular mixture, achieving was is effectively the equivalent of laminar flow for fluid dynamics.

If everything is tractor trailers, the throughput is lower based on vehicle classification. If everything is passenger cars, variation between SUV and sub-compact car provides a mid-range. If everything is motocycles only the vehicle dynamic squashes vehicle volume and boosts maneuverability to inflate apparent throughput.

Somewhere in this, driver skill level and physical fitness comes into play though, so there has to be an index for that also, based on available residential zones, and an origin/destination matrix.

If everyone is driving from the retirement home to church bingo, the throughput will be lower than tractor trailers. Tractor trailers and other commercial vehicles are expected to operate at lower speeds due to inertial lag during braking and acceleration, prolonging periods spend at lower speeds. All other vehicles can only be modeled according to legal speed limits, even though the accuracy of the simulation is threatened by the fact that in reality many vehicles will be ignoring speed limits.

Time of day, however, is usually going to model rush hour, which consequentially alters vehicle profile and driver persona.

That was an interesting way to kill a few minutes.

My conclusion is that throughput is basically a function of space the intersection takes up. Also water is wet.

Sure given a fixed space you can optimize type to maximize throughput. If you really care about maximizing throughput with minimal complexity (the more rapid fire merges and lane changes you have the greater the chance of an accident that bottles up the whole system) big rotaries seem to be the way to go (above/below grade structures and traffic signals are a lot more expensive and cost more to maintain than some raised medians to divide right turn lanes).

In general, but there were some exceptions. The various roundabout solutions (rotaries, dumbbells, cloverleafs, etc.) seemed to punch under their size, while the continuous-flow intersection moved a lot of cars in a space not much larger than a traditional 4-way intersection.

Also, the video showed how effective getting the signaling right can be: intersections with dedicated left turn lanes (with the TPME mod) outperformed stock traffic lights by almost 50%. The 4-lane road with dedicated left turn lane outperformed a 6-lane road without protected lefts.

My own observations were that roundabouts tended to perform better than intersections with time division utilization and also didn't require any signaling infrastructure.

Modifications to roundabouts that added dedicated bypass routes for higher traffic / shorter path (merge to closest output without being part of the main queue) were inexpensive upgrades.

The same general observation also applied to higher scale variations where the core loop was either factored out entirely (every route has a dedicated path); applying more complexity (expensive) lead to better results.

Of course the real world has different needs than this over-simplified simulation.

Terrain and political situations impose limits on usable work area while use patterns can vary widely. Worse solutions that used to work in the past can be woefully incorrect as the world around them changes.

The example that comes to my mind is the pattern in urban areas, especially like Seattle with heavy geographic constraints that make alternate routes untenable, where the old "most commuters go in to the city in the morning and leave in the evening" break entirely as jobs flee the expensive urban core and leave workers trying to traverse the choke points where cross traffic was never designed in.

The cars move through each other. Looks buggy. But cool.

Something about this video was very satisfying.

Reply on news.ycombinator.com