Comment by potatolicious | Hacker News Reader

Comment by potatolicious | Hacker News Reader

Parent Full thread

This is a huge problem, as big as the server issues and the DRM.

The game is, IMO, mostly unplayable at this point. I have no idea how they let this one out the door with the pathfinding problem, which in my book is an absolute showstopper.

Clever, dedicated, and somewhat masochistic players have found ways to "hack" the poor pathfinding behavior. The most popular format right now is to build a city with no intersections, since the AI deals so spectacularly poorly with them. Which is to say, the entire city is a single, long, winding road. This forces your dumb sims to have no choice but the right one.

Some apologists have claimed this is simply the rules of the game, but I still maintain that SimCity fails unless it maintains some semblance to real-life cities. If the only way to play the game effectively is to build something that bears zero semblance to any real city, then it has failed.

The funny part is that this loosely resembles the recent Heroku fiasco. In the game your civic services (police, fire, ambulances, garbage) are supposed to intelligently service the city - in reality their pathfinding results in basically random behavior, vastly increasing the amount of capacity you need to build to statistically serve an area. You have to grossly over-build your fire and police departments because their pathfinding is awful and random.

I'm not usually this hard on others' hard work - but this game is a travesty that should never have shipped in this state, even disregarding the server issues. There are core gameplay mechanics that are still fundamentally broken.

"in reality their pathfinding results in basically random behavior"

It's actually the opposite, sims always go to the nearest source for whatever it is they need despite traffic or congestion. Some randomness in picking a destination might actually solve the problem without requiring any particular intelligence.

It's semi-consistent when it comes to city vehicles like garbage trucks or school buses. What you get is groups of vehicles traveling in a pack, bumper-to-bumper, taking a seemingly random path between bus stops/garbage pickups.

At various points in this one or two vehicles will peel off from the pack in a random manner - often towards stops with nothing to pick up.

So the aggregate effect seems to be a clump of city vehicles moving randomly around the city, neither towards nor away from places where they're needed, and you're relying mostly on the random "split from pack" behavior to get any real work done.

It's SUCH a better idea to simply weight paths based on their congestion. Like, that's A* path-finding 101.

I'm not sure they're actually using A*. In the following video, some cars are going directly right, although it isn't the shortest "bird-view" distance to their destination.

http://www.youtube.com/watch?v=l_ufAd79bOA

Edit: actually, they seem blocked by cars from the left lane. They'd go left otherwise!

Though I can't find any official sources to back it up, various internet citizens seem to claim they're using a D* Lite algorithm. I find it hard to believe. Either they're using some much cruder algorithm or they've implemented D* Lite horribly.

D* Lite Source https://twitter.com/MaxisGuillaume/status/311218564600315905

  A developer has said that they are using D*

They should be using min-cut max-flow graph algorithms rather than dist based metrics.

From past experience with traffic routing in real life, the best bet is to encourage a heavily monitored subset of traffic to take alternative paths. If this traffic continues to route faster than, you gradually increase the share of traffic to the alternative path until you observe it run slower than the primary route, or until there is no more traffic to route.

I have no idea how they have let this go out without putting more work into this...however maybe the same people who wrote the AI are responsible for running the servers and this is why we're seeing problems there too?

Luckily, All of this calculation is done on the server, so you won't even need to download a patch when the fix the pathfinding. /snark

Could you confirm this?

He even added "/snark" to show it was a facetious comment referring to the whole 'servers are required to do the heavy calculations!' argument made to justify the DRM.

Looks like they didn't even consider using some basic weighted graph algorithms. It would at least look less dumb if the sims wouldnt choose to take a dirt road with heavy traffic jams.

This is what I was truly shocked by. Even the most basic shortest path algorithm simulating traffic should include weights.

I get the impression that SimCity uses unweighted shortest path to encourage you to upgrade roads.

They should use poorly chosen weightings. A part of the point of Sims games is that you have to be the Sims caretaker.

Some of the test cases in the OP made me think, "Man, they've perfectly simulated Houston's road design". Then I was rather impressed.

The road design? It's more that the most of the drivers behave like mindless automatons.

"I have no idea how they let this one out the door..."

End of the quarter is two weeks away. EA needed the revenue bump in the earning report this quarter. It is sad that stock market issues drive release decisions.

But things like pathfinding and game mechanics are / should be amongst the first things they make for a game like this, even before deciding and developing a visual style / graphics and sound and the like. So rushing on developing / polishing pathfinding doesn't make sense to me.

IIRC, all these problems where there in SimCity 2000. You had to have police and fire every two blocks and the best possible layout was one long, back and forth winding rail line. Maybe it's in the bug database as "Closed as Feature - Nostalgia".

2000 didn't use any pathfinding for fire and police. Each station simply covered a certain area when it come to suppressing crime or fire. When it came to fighting disasters, the position of a station didn't matter. You could simply deploy police and firefighters wherever you wanted, instantly.

I don't know about the one long winding rail line being the best possible layout, but more realistic layouts worked very well. It's not surprising that a simulation can be gamed to work better with unrealistic layouts (e.g. search for "Magnasanti"), but a city simulator should work reasonably well with realistic layouts, even if it's not necessarily the best. Real cities surely aren't the best layout either.

This is correct. In SC2k, you can see the region of effectiveness police and fire stations have, the region is a circle surrounding each station. I don't remember the name of the tool, but in the tools palette thingy there is an option to show a map and then toggle the various information about crime, fires, health coverage, etc.

I don't remember this particular issue either, but I do remember that an optimal layout for one of the early games (it was either SC2000 or the one that came out for SNES at about the same time) consisted of small patches of roads, just large enough to join 3 zone types.

SimCity has always had its quirks like this, but I think this version really does take it to a new level.

It's always rewarded some fairly unrealistic layouts. Like I said, inappropriately rewarding certain weird layouts is much less important than inappropriately punishing normal ones. The former is only to be expected in any complex game, but the latter destroys realism.

I agree. While this behavior was apparently known and explained somewhat prior to release, I didn't know about it. I was prepared to give the a game a go once the connectivity issues were taken care of, but with this information I'm not even remotely interested anymore.

An ecellent and entertaining summary. I was looking forward to this game based on the previews but it seems like one big disappointment after another. Guess I'll stick with Netlogo...

It is a problem, in that the game is a lot less sophisticated than EA would have you believe, but does that make it huge ?

I've built agent level simulators and have run gate level simulations on fpga designs and the thing they share is amazing complexity which goes up really really quickly. Does is make SimCity less 'fun' ? Certainly. Does it rise to the level of failure? I'm not sure I agree.

I built a simple action selection mechanism based system for a predator / prey type robot setup. It was pretty simplistic (the prey ate/slept/hid/ran the predator ate/hunted/slept) assuming each of our sims has something like that I can't imagine how you would make that 'real time' unless you had like 100 machines dedicated to each city.

Cost effective web products have anywhere between 1K to 10K users per machine, not .01 user per machine. So the "real" (or real-er) thing would be pretty cost prohibitive as a game I suspect.

> "in that the game is a lot less sophisticated than EA would have you believe, but does that make it huge ?"

Sure. It's one thing for the game to fail to live up to the complexity the fanbase expected, it's another for the game to be broken when played the way the designers intended.

SimCity is a game, it is only a simulation insofar as simulations are entertaining. When the nature of your simulation means that players are constantly having to dodge and weave around the dynamics of your simulation in a distinctly un-fun way, you've failed at the basic task.

> " assuming each of our sims has something like that I can't imagine how you would make that 'real time' unless you had like 100 machines dedicated to each city."

Indeed, and I commented on this earlier - the choice of using agent-based simulation was IMO a disastrously wrong one. The limitations on agent population and agent complexity (due to hardware constraints) results in a simulation that is unintuitive, inaccurate, and most importantly, not fun. Previous iterations in the series had macro-level simulations which created less unintuitive emergent behavior and whose performance allowed a greater level of game complexity.

At the end of the day, the goal wasn't to create an agent-based simulation of a city, it was to create a pseudo-simulation of a city that happened to be fun to manipulate. On that front IMO SimCity 2013 is a huge failure.

Players want to come in and build cities - cities that resemble their real-life experiences to some degree. Whether they're out to create a farm town, a sleepy suburb, or a skyscraping metropolis, differs, but practically none of them are looking to study the quirks and edge cases of pathfinding implementations and model their gameplay around arbitrary rules that have no bearing in real life.

There are exceptions of course, the guy who created Magnasanti in SimCity 3000 took optimizing around the extreme parameters of the simulation to heart - he created the densest, most populous city possible given the game rules. It resembled nothing like what a real city would look like, but it functioned within the game's rules.

Players like that are also few and far between. Most people are there for the cities, not the metagame.

Reply on news.ycombinator.com