Forums | MacLife

One way to think of NPC simulation is to look at the scales we see in the 'real world'.

When I look at it, I think people can be grouped up as individuals, relationship groups (eg clans, families), political groups, cities, countries and maybe continents. Inside each group, all NPCs can be considered to be doing much the same thing.

To bring this back to game terms, imagine a player wandering through a city. The NPCs they directly interact with will be simulated as deeply as possible. The NPCs close around them will be simulated to a deep level, but slightly simplified. The NPCs nearby but out of sight can be further simplified, and  any NPCs outside that range, but still in the city can be greatly simplified.

NPCs outside the city can't have a direct bearing on events inside the city, so they can be simplified a lot, by treating them as groups. Distant cities can be considered one entity for the purpose of simulation, and would seem to behave as one entity when viewed from a sufficiently large distance.

You don't need to know what Joe Bloggs, Farmer, does when you're in another city - he can't possibly affect the player when they're so far apart. He can be factored into the overall city model as one more farmer. This model will just take inputs like this, and spit out the outputs, such as general health, general mood, number of births and deaths, etc. Only when the player comes closer do these outputs need to be applied to individuals.

So there are a set of discrete levels of AI, based around distance from player. Another level would be wandering 'adventurers' or hight level characters, who can do everything the player does and probably more. They'd need to be well simulated throughout the entire game. Some examples of these characters (in the Silmarillion) would be Turin, Morgoth, Fingolfin, Sauron, Glaurung, Ungoliant, and a bunch of (initially) random characters.

The idea I have is that the player can contribute to the story, by accepting missions or quests, or by doing things that they think help one side or the other. If they don't take up a quest, a wandering adventurer may (it's down to who's present when the quest is offered, or who accepts it first).

A goal of mine is that an NPC could beat the player to some goal, whether it be amassing gold (when playing the Dwarven role), amassing political power (when playing the Human role), defeating Morgoth (when playing as Elves or Edain) or even in rising in Morgoth's eyes (when playing as Evil or Orc). Just as in the 'real world', our actions can affect the world, but the world doesn't hang upon our actions. If we fail to pick up the sword, another may well do it. If we miss an opportunity, another will surely take it.

However - the experience that is central to any game is the player's experience, so the player may receive certain benefits not available to other NPCs, such as slightly better odds in combat, slightly better rewards after missions, etc. They are then better able to succeed where others fail, creating an heroic feeling.

As for the Beleriand terrain data - I'm looking at the entire planet, not just a small section. I've got a map that's about 60,000 by 60,000 points, which corresponds to 6,000 by 6,000 miles. Between those heightfield points, I'll interpolate using my own subdivision scheme. The map design is partly done, and I'm creating a designer tool to edit the heightfields and compress subsections into memory as I go. I finally get to use Bryce for something - mountain and hill heightfield generation!

Disc size is unimportant - this is a project I'm doing for myself. My original terrain map was just 72MB of uncompressed data, but I'm going for a better resolution, so I'm looking at 13.4 GB of uncompressed data, potentially more. I'm reading up on compression algorithms in preparation for writing some codecs, hopefully to reduce storage requirements down to by 75%+. I've worked out a caching scheme that only requires about 144MB in memory at once (maybe I'll bump that down to 36MB with in-memory compression).

This has certainly been on our minds here since we picked up a PS2 and got the GTA series.

It's incredible what people come up with while watching these games.  It starts with: can you shoot through a windsheild?  can you see into that building?  can you buy food and eat it?

But eventually the ideas start to gravitate towards the open-endedness that this article is predicting. Questions like: can you follow someone to his/her workplace?  what if everyone had a real position in the city?  what if you killed one person, and the effect rippled through the city because his position was important?  what if you could sit down in the park and play a game of chess?  or in your hideout and whip out your Nintendo?  what if you could actually talk to people?

The only other thing I have to say is this:  Escape Velocity.  Perhaps this game of all others is closer to the "completely open" label than anything else.  Plus, I think it would greatly benifit from the AI described above.  Imagine if the storylines were not actually written by people, but by the AI itself.  Crazy.  :!:

I like to hear that you're working on something, Gary, even if it's just a hobby.