It has often been said that the true star of Rockstar's groundbreaking Grand Theft Auto IV was not lead character Niko Bellic but, rather Liberty City itself, a fictional metropolis modeled after New York City.
Creating a breathing, living city is one of the toughest challenges facing developers of open world games today: get it wrong, and your player feels like she's walking through a lifeless television set with cheap props and false fronts for buildings. Get it right, and she'll be absorbed enough in your world to momentarily forget she's playing a game at all.
Ubisoft Montreal is among the best in the business at creating believable cities thanks to its Assassin's Creed series. We sat down with Alex Hutchinson, creative director of the upcoming Assassin's Creed III, for his tips on making a city feel alive. He tells us it's "one of the hardest things you could possibly do," but managed to offer the following.
1. Two-tiered reactions
It is a given that NPCs in open world cities will react to your actions: fire into a crowd in any Grand Theft Auto, and pedestrians will run screaming, cops will come after you, and some of the rougher types might even fire back.
What Ubisoft Montreal has been focusing on for the Assassin's Creed is in two-tiered reactions: NPCs that understand not only the action the player has committed, but how it relates to them. Hutchinson explains:
"If someone hires you to kill Character A, and you do it, and they react like, 'Good, that's what I asked you to do.' But then you go and kill some other random person and they're going to freak out, like you've lost your mind. Whereas someone who wasn't involved in that conversation would freak out at both of the instances."
It sounds simple when explained that way, but Hutchinson insists that it is "one of those things that no one in games yet has truly solved," including his own team.
2. Random behavior from a pool
The Assassin's Creed series has perhaps the most believable crowds in games today: walk through an environment, and you'll see countless people going about their lives in realistic ways.
The trick is that these NPCs are spawned semi-randomly. As pedestrians populate an area in the game, they are randomly assigned behavior from a large pool of possibilities, meaning it's unlikely you'll notice too many of them going through the exact same animation cycles.
It's an expensive part of the series' development, Hutchinson admits, but it's a key ingredient in making a city feel alive.
3. Variety over complexity
Though a minority of your players will be troublemakers trying to break your game and peek behind the curtains (including this author), most of them will play your game just as intended. And in an open world environment, that means running quickly through the crowds from point A to point B.
Take advantage of this! Most players will only look at your random NPCs for six seconds, max, Hutchinson tells us. So instead of focusing on a complex AI routine full of animation cycles that last much longer than that, put more emphasis on a wide variety of different, shorter actions.
4. City first, props later
Every Assassin's Creed to date has laid out its cities based on actual historical maps, worrying about filling it in with gameplay setpieces later.
It's unlikely that your game will strictly adhere to an existing map, but it's a good takeaway: even a game like Assassin's Creed focuses on a believable city layout first and worries about making that city fun to play in later. Which brings us to our final tip:
5. Save your polish for mission areas
It's tempting to make your entire city a playground for explorers, and adding little touches will be rewarding, but remember that the places in your city where actual missions take place will give you the most bang for your buck.
Have fun decorating the city and placing clever props where you can, but focus most of your energy on making the mission areas as fun as you can, as that's where players will be focusing the most attention.
Thanks for composing this article; It will be coming in handy very soon.
Too bad the promo didn't actually show off any NPCs. :)
The game I am working on will rely on persistant NPCs that will have some kind of consistancy to their actions, but for this I think I'll just extend tip #2 to choose behaviours randomly from a pool of actions suitable to that NPC's role in the town.
This variety over complexity point really concerns me. It doesn't concern me for the AC series, but rather for Watch Dogs. If they apply this short animation cycle to NPCs in Watch Dogs it will be disastrous. Not unless they nerf the toolset and number of pedestrians considerably, so that players don't end up following an NPC which is not part of the essential plot only to realize they have no depth beyond a simple animation cycle and some background text. Thats where i felt conceptually RDR got things right, they had sparsely populated areas (read: on purpose), and could therefore focus on complexity over variety, although i'd argue that their NPCs werent that great either. I mean even though they could travel inbetween towns, start bar fights and kill other innocents, they still just flipped switches most of the time between 1 and 0.
Most random NPCs (not characters the player interacts with) the player interacts with will be a shorter period of time.
How's an NPC going to react when you raise a bridge? Not drive over it like a ramp, they will stop. Unless they are in a police helicopter giving chase. Or if the player started/stopped a train, they are not going to do anything drastic.
But we saw in the E3 demo for Watch Dogs that crashes seemed to be an area where they have a lot of NPC actions. And for the most part those NPCs are not critical to what the player is doing.
Variety is better than complexity. It is easier to have some NPCs call 911, another to run up to a car and check the driver, and lastly a NPC in a car in shock over the chase doesn't move, but talks to the driver. These are short, but varied actions.
They helped create a chaotic scene without having one of the NPCs try to grab the gun from the player and shoot the bad guys. Or have one NPC try to get spectator NPCs to help him remove people from each of the cars. That's too complex and not needed.
Most often the short, but varied actions are better than one long series of actions.
One also has to remember that players will often fill in the blanks on their own. Seeing the NPCS, as John put it "create a chaotic scene", will often result in players remembering things and inferring things that never actually happened. People love narratives in that sense.
Now, you're correct in that what if the game or the player demands or desires to follow someone for extended periods of time. I don't think the article is saying don't let them, but rather, instead of creating a schedule for everyone (99% of whom the player will never follow) but instead have many many possible actions the NPC can perform in response which then result in more branching (and the player again will build their own story). They can: duck into a nearby store and trigger store animations. When those are done, they trigger another animation of walking on the street, etc. Alternatively, the NPC can notice the player follownig them and stop and call the player out and not move (and call the police - in a game like Watchdogs, this might be a bad thing). Alternatively again, give them a set of points to walk to around the city or a car they'll get into.
None of which need to be special, unique or require any sort of complex system. Even for a game like Watchdogs, if say, an NPCs' car gets noticed only when the NPC actually has a car, I don't think most players would consider that odd.
I agree about variation. A pet peeve of mine for years is in sports game crowds ( now they are getting better about this) but nothing breaks my immersion as seeing the exact same "person" in the crowd with the exact same animation every 20th seat or whatever. Either do it correctly or dont do it at all. This is simply cosmetic yet for years developers were lazy in this regard.
I also was and am a big fan of games like AC and Red dead redemption and , loved the world they created there, but I do have a pet peeve in those games as well . How many times do I need to encounter the same " random" encounter"? How about ONCE. Any more than once it becomes annoying game world spam and breaks the immersion of the world, just sayin..
Anyway not to ramble to much, I do appreciate all the hard work you guy put into these "living" world.
"This is simply cosmetic yet for years developers were lazy in this regard"
You know it pisses me off when people make assumptions like that. I can understand players that dont know any better but on a developper`s community like here its just rude to assume lazyness.
I worked on console sport games for a couple of years in the past. In some of those games (depending on the court size, typical camera angle, number of seats, etc), animating and rendering the crowd takes a huge chunk of a fixed, limited budget of CPU time, GPU time and, especially, memory.
You said it yourself, its purely cosmetic. You can't interact with the crowd in sport games, unlike open world games like AC and RDR. Should sport games developpers take away ressources dedicated to actual gameplay and physics to improve a cosmetic crowd that players dont even see 90% of the time unless they go into replay? In AC the crowd is right in your face, but not in sport games. How much of the ressources should be spent on the crowd? 5%? 20% ? What are you going to cut on the court or field to make some guys eat hotdogs in stands you can't reach?
Its not lazyness its a very informed choice to spend your limited ressources on actual gameplay, in the context of a sport game.
@Mathieu
actually in baseball games the crowd is right in your face, and for many like myself an important element. We simply disagree, you think the crowd is not an important element, I believe it is especially in baseball games etc. Agree to disagree.
No one is asking for some complex algos running in the background for crowd animation, we are simply asking for a few more art elements so that not every 7th person is not the same reused element. The human eye and mind are very adept at picking up and spotting patterns.
With 5000 people visible in a stadium's crowd, you will never fit enough animations in memory to make them all unique. It has to cut at some point and with current hardware, that point is unfortunately not very high.
Unless you go with "realistic" baseball crowds, but I doubt the MLB would let you ship a licensed product with empty stands.
Again no one is asking for 5000 unique elements. They only come into question when they are at such a close distance as individual elements become recognizable. That is much much much less than 5000. With realizing that the problem lies with the humans eye and mind's ability to spot trend, and repetition etc, you also need to only worry about repetition in any given frame area, again lowering the amount of unique elements needed.Plus given that people are sitting watching an event, the majority of assets dont even have to be animated.
I just find it curious that so much time and resources are used to create such realistic individual stadiums, down to small small details, then corners are cut when populating said stadiums. IMHO it undermines all that hard work reproducing the stadiums.
But as I mention, it seems as if my concerns are being addressed as in more and more sports titles, crowds are finally being given the attention IMHO they deserve in order to create the illusion especially in sports titles in which the crowd is indeed a "part of the action" IE. Baseball, Tennis, and the first several rows in Hockey.
I know I'm kinda late with replying, I hope that isn't some kind of 'unwritten rule' around here.
TC's point just made me think of an interesting logic. He mentioned that baseball fans are "in your face". I understand the resources are being used for each fan, but what if.. just what if extra resources were used for the fans ONLY when they're in close proximity of the player, OR if the camera were close enough to the fans (sort of like 'clipping' when objects are far away compared to being closer in view).
This 'method' or 'process' or engine would only be called if the player is at a certain distance from the fans. Subtle to bigger animations like a fan sitting up straight, pulling up a hot dog/soda from off the floor and taking a bite, etc. could all be fired when the player comes closer (which fans the player will be looking at at the moment, while all others are in 'waiting' mode). Is that possible? This sounds like an interesting way to confront the problem TC sees in sports games nowadays.
Maybe instead of having every person mechanically shout out some pre-written and mostly insignificant dialogue as you pass (which would quite frankly be frightening in real life, even more frightening if it's the same exact thing said in the same exact way), how about people who don't know you don't pay much attention to you as you pass (like in real life, at most they may look up momentarily, and then avert their gaze so as not to be awkward). The ones who do know you from previous interaction may just give a slight nod as they look up from what they're doing, maybe a slight wave if you're friendly with them. If you actually approach someone in real life, they don't just start a conversation with an elaborate discourse, they may give a 'hey what's up?', but not much more.
I agree with this wholeheartedly. Subtlety is a problem in all aspects of game development, but in a game like Assassin's Creed where you're often walking around slowly through crowds, having people quickly glance at you, give you a once-over, and then ignore you entirely would do a lot for my immersion.
Another thing I'd add to this is that people don't just react to one person. I'd expect npc's to glance, wave, nod at eachother just as much as they would at me. Even without getting into complex details such as whether or not someone finds someone else attractive, distracting, intimidating, etc, just knowing that people around you are acknowledging eachother would add a lot to the credibility of the experience in my opinion.
The population size of the community makes a big difference in social inattention. When I lived in Seattle, people from Fairbanks would complain about how cold big city folks acted.
In some developing countries shop keepers are more aggressive and yell at you, stalk you, and try to physically drag you into stores.
The two preset actions for shock and alarm and the note of their costs shows how far we have to go to get decent simulation and acting in games. Too bad this cant be more modular. It would allow acting studios and mime studios and acrobat schools to sell behavior packs.
The Assassins Creed macho strut is very stylized. Ideally an assassin avatar would use body language to blend in with civillians. However, this would be cost prohibitive unless it was based off of modular bodylanguage tech tools or better yet modular kits from: The Juilliard School, The Oxford School of Drama, Yale School of Drama, Royal Scottish Academy of Music and Drama, New York University’s Tisch School of the Arts, Guildhall School of Music and Drama, American Conservatory Theatre, Birmingham School of Acting, American Repertory Theatre, and London Academy of Music and Dramatic Art.
I think its more a solution to common mistakes than a '"recipe" (like: "city first, props later", i made that mistake once when I was developing an small game in a university project).
3 and 5, though... those don't sound like they'd lead to games that exploration-oriented gamers will enjoy much.
I understand them from a cold economic viewpoint. Finite development money has to go where it's likely to have the highest payoff. Still, following that slavishly might lead to lowest-common-denominator gameplay.
Something some people might already be doing (I hope) would be to have a "focusing" concept to determine the depth of mob behavior. An NPC you're directly interacting with in a non-violent way should have a rich set of actions; a mob in sight within (say) 20 feet has basic actions plus a few role-appropriate random actions; a visible mob beyond 20 feet has just a few basic "big" actions; and all other mobs are only simulated at most.
I'd think that would meet the goal of not coding behaviors that players won't need (i.e., NPCs in a crowd), but people and places in the player's immediate vicinity still have the potential to be worth exploring. No doubt there are practical problems (there always are), but it still sounds better to me than just a laser-like utilitarian focus on setpiece content.
Too bad the promo didn't actually show off any NPCs. :)
The game I am working on will rely on persistant NPCs that will have some kind of consistancy to their actions, but for this I think I'll just extend tip #2 to choose behaviours randomly from a pool of actions suitable to that NPC's role in the town.
How's an NPC going to react when you raise a bridge? Not drive over it like a ramp, they will stop. Unless they are in a police helicopter giving chase. Or if the player started/stopped a train, they are not going to do anything drastic.
But we saw in the E3 demo for Watch Dogs that crashes seemed to be an area where they have a lot of NPC actions. And for the most part those NPCs are not critical to what the player is doing.
Variety is better than complexity. It is easier to have some NPCs call 911, another to run up to a car and check the driver, and lastly a NPC in a car in shock over the chase doesn't move, but talks to the driver. These are short, but varied actions.
They helped create a chaotic scene without having one of the NPCs try to grab the gun from the player and shoot the bad guys. Or have one NPC try to get spectator NPCs to help him remove people from each of the cars. That's too complex and not needed.
Most often the short, but varied actions are better than one long series of actions.
Now, you're correct in that what if the game or the player demands or desires to follow someone for extended periods of time. I don't think the article is saying don't let them, but rather, instead of creating a schedule for everyone (99% of whom the player will never follow) but instead have many many possible actions the NPC can perform in response which then result in more branching (and the player again will build their own story). They can: duck into a nearby store and trigger store animations. When those are done, they trigger another animation of walking on the street, etc. Alternatively, the NPC can notice the player follownig them and stop and call the player out and not move (and call the police - in a game like Watchdogs, this might be a bad thing). Alternatively again, give them a set of points to walk to around the city or a car they'll get into.
None of which need to be special, unique or require any sort of complex system. Even for a game like Watchdogs, if say, an NPCs' car gets noticed only when the NPC actually has a car, I don't think most players would consider that odd.
I also was and am a big fan of games like AC and Red dead redemption and , loved the world they created there, but I do have a pet peeve in those games as well . How many times do I need to encounter the same " random" encounter"? How about ONCE. Any more than once it becomes annoying game world spam and breaks the immersion of the world, just sayin..
Anyway not to ramble to much, I do appreciate all the hard work you guy put into these "living" world.
You know it pisses me off when people make assumptions like that. I can understand players that dont know any better but on a developper`s community like here its just rude to assume lazyness.
I worked on console sport games for a couple of years in the past. In some of those games (depending on the court size, typical camera angle, number of seats, etc), animating and rendering the crowd takes a huge chunk of a fixed, limited budget of CPU time, GPU time and, especially, memory.
You said it yourself, its purely cosmetic. You can't interact with the crowd in sport games, unlike open world games like AC and RDR. Should sport games developpers take away ressources dedicated to actual gameplay and physics to improve a cosmetic crowd that players dont even see 90% of the time unless they go into replay? In AC the crowd is right in your face, but not in sport games. How much of the ressources should be spent on the crowd? 5%? 20% ? What are you going to cut on the court or field to make some guys eat hotdogs in stands you can't reach?
Its not lazyness its a very informed choice to spend your limited ressources on actual gameplay, in the context of a sport game.
actually in baseball games the crowd is right in your face, and for many like myself an important element. We simply disagree, you think the crowd is not an important element, I believe it is especially in baseball games etc. Agree to disagree.
No one is asking for some complex algos running in the background for crowd animation, we are simply asking for a few more art elements so that not every 7th person is not the same reused element. The human eye and mind are very adept at picking up and spotting patterns.
Unless you go with "realistic" baseball crowds, but I doubt the MLB would let you ship a licensed product with empty stands.
I just find it curious that so much time and resources are used to create such realistic individual stadiums, down to small small details, then corners are cut when populating said stadiums. IMHO it undermines all that hard work reproducing the stadiums.
But as I mention, it seems as if my concerns are being addressed as in more and more sports titles, crowds are finally being given the attention IMHO they deserve in order to create the illusion especially in sports titles in which the crowd is indeed a "part of the action" IE. Baseball, Tennis, and the first several rows in Hockey.
This article and discussion is all about how to use resources in order to create a realistic environment all in order to make a better game.
---------
as is just about everything else in this industry as well. It goes without saying.
thanks for the chat, its been fun
TC's point just made me think of an interesting logic. He mentioned that baseball fans are "in your face". I understand the resources are being used for each fan, but what if.. just what if extra resources were used for the fans ONLY when they're in close proximity of the player, OR if the camera were close enough to the fans (sort of like 'clipping' when objects are far away compared to being closer in view).
This 'method' or 'process' or engine would only be called if the player is at a certain distance from the fans. Subtle to bigger animations like a fan sitting up straight, pulling up a hot dog/soda from off the floor and taking a bite, etc. could all be fired when the player comes closer (which fans the player will be looking at at the moment, while all others are in 'waiting' mode). Is that possible? This sounds like an interesting way to confront the problem TC sees in sports games nowadays.
In some developing countries shop keepers are more aggressive and yell at you, stalk you, and try to physically drag you into stores.
The two preset actions for shock and alarm and the note of their costs shows how far we have to go to get decent simulation and acting in games. Too bad this cant be more modular. It would allow acting studios and mime studios and acrobat schools to sell behavior packs.
3 and 5, though... those don't sound like they'd lead to games that exploration-oriented gamers will enjoy much.
I understand them from a cold economic viewpoint. Finite development money has to go where it's likely to have the highest payoff. Still, following that slavishly might lead to lowest-common-denominator gameplay.
Something some people might already be doing (I hope) would be to have a "focusing" concept to determine the depth of mob behavior. An NPC you're directly interacting with in a non-violent way should have a rich set of actions; a mob in sight within (say) 20 feet has basic actions plus a few role-appropriate random actions; a visible mob beyond 20 feet has just a few basic "big" actions; and all other mobs are only simulated at most.
I'd think that would meet the goal of not coding behaviors that players won't need (i.e., NPCs in a crowd), but people and places in the player's immediate vicinity still have the potential to be worth exploring. No doubt there are practical problems (there always are), but it still sounds better to me than just a laser-like utilitarian focus on setpiece content.