~hed~

Sharpie: I think main problem with dynamically generated environments is trying to balance the fact that you don’t want to make everything look like it’s constructed from the same set pieces, when everything is in fact constructed from the same set pieces. The time to generate the world could also perhaps be split - generate new world content only when getting too close to an edge. What about game play, too? Libraries of missions? Missions with different NPCs that all have the same objective?

What I imagine is that the less important a variable is, the more it is left to chance. I don’t know a better way to explain then to give some examples;

Suppose, for instance, you take your warrior-person and run into a clearing. You trample some of the grass, and leave some light footprints. You kill a monster, and leave its corpse on the road. You thwack a tree with your sword a few times, taking a sizable chunk out of the bark.

Now, suppose you return to that area, I dunno, three game-days later. What the player is expecting (or at least, hoping) to find is the marks still on the tree, and the monster’s corpse (or some evidence that it was removed); the footprints are negligable because it’s been three days. Conventional game theory would say that, in the simple case, those marks would be forgotten; and of course, in any current game they would. A more robust game might try to remember everything exactly; how the monster fell, exactly (since it’d be using rag-doll, obviously), exactly what the marks on the tree look like. What I’m suggesting is that the game would remember a sort of summary of what had happened; it would remember that there was a mark on the tree that was roughly X deep, Y inches from the ground. It would remember that there was a monster corpse in the street three days ago, and then use a formula (based on elements of the game world) to randomly decide whether the corpse is in the same place, if it’s been moved to the side of the road (because someone passed through), or gone altogether. The advantages of this method is not only a much smaller amount of information to remember, but a much more dynamic game world; after all, if you killed a monster in a forest and left it alone for three days, you wouldn’t expect it to be in exactly the same position when you came back.

Likewise, an NPC doesn’t have to remember every little thing you said, to the letter. Instead of remembering that you said this, this, and that, and bought such-and-such for that X gold and tried to attack him Y times, it would remember a summary of its feelings towards you, such as “I am a gruff merchant, and I gladly sell things to Player, but I don’t wholly trust Player.” Again, this takes up much less space and is much more evident of an actual thinking being, and not a pre-programmed NPC.

The most apparent drawback with things like these are that, quite simply, they require a lot of work; for an NPC, for example, you would have to mold a realistic personality, make sure it remembers in detail the more important events (so that it doesn’t, for example, swear an oath to help you one day then forget about it the next), and carefully test it to make sure the balance of static and dynamic information is natural and believable. Like I said, this stuff is hard, but eventually the benefit will outweigh the cost.

Erm… I feel like saying more, but most of it’s redundant. I have a tendancy to get rather romantic when talking about dream video games, forgive me.

It’s kind of eerie that you should bring this up. I was re-thinking my initial ideas and contemplating this exact same concept on my way into work today. Modeling, detailing, and rendering a complete recreation of–for example–the city of Singapore, would take years of work and occupy an ungodly amount of disk space. It’s just simply not cost effective to do.

IMO, the best option to pursue (economically and creatively speaking) is to generate a skeleton world the first time a game is loaded on a platform. This randomly generated skeleton would remain for the duration of the game’s life on said platform. As the player travels to different areas of the world this skeleton is continually fleshed out with more detailed objects via the AI and random generation. This system would ensure that every individual installation of the game would be different (say goodbye to walkthrough guides). Obviously, once a section of the world has been generated to full detail, the majority of it would have to be retained and recorded on disk otherwise a player would be immersed in a continuously changing and random world (if you count on a given NPC being at a certain location, they had better be there when you return from somewhere else).

Obviously the initial load time of this game would be considerable (an entire game world has to be built from scratch) but the benefits far outweigh a wait of three to four minutes the first time you pop it in. If this game were to be developed for the Xbox 360 then it would most certainly require the hard drive variety so the generated world could be stored for retention (since the information is not pre-recorded on the DVD). The processing power for such a task must be available as well. Hopefully a next-gen console or high end PC could handle the task (Sandstone and his AI speciality would be better qualified to answer this question).

If not, then a game of this caliber might be farther down the line then I’d hoped.

Yes and no. AIs of any shape require an extreme amount of programming and finessé. But, eventually, the benefit will outweight the cost. In the future, near and not, with our Xbox 520s and our Playstation 4s, the desire for more detail (and the means to implement it) will increase exponentially; we’ll want individual blades of grass, individual dropletts of water, suitable NPC responses to anything we do or say, destructable environments on everything that makes sense. In the future, hand-modelling an entire game world will become absurd; the number of towns, the shape of the buildings, the different NPCs, multiplied by the detail, will take eons to produce and terabytes to store. That’s why we create an AI to make one for us; we tell the AI how people build towns, what farmers talk like, what happens when you hit a tree with an axe. Then to create a richer experience, what is required is not more content, but more information: we tell the AI first how a tree reacts when being hit with an axe. Then we tell it about velocity and force. Then we tell it about bark, and wood, and steel; if we were to continue indefinitely we would eventually tell it about molecules and atoms, and it would then know infallibly how to react when that axe hits that tree.

As inevitable as these advances are, they’re remarkably absent from modern games… the only important example is of the randomly-generated forests in Morrowing: Oblivion (according to a video preview).

To get a really good taste of what this kind of theory encompasses, here’s a link to the experimental game .kkreiger, an FPS that takes up 98 kb of space. (screenshot)

The player controls would have to stay the same no matter how you play it (hard core shooter, spy thriller, thinker rpg, etc). The best option would probably be to implement a solid Halo control scheme but take advantage of the unutilized d-pad (and ‘Back’ button) to implement a comprehensive command menu. The variety of weapons and gadgets would be controlled in the same fashion as a gun but be context sensitive (i.e. hand-to-hand combat automatically switching the player to 3rd person view).

By the way…I found a location and a game engine I wouldn’t mind exploiting.

I really feel that total open-ended gaming is the future. Grand Theft Auto only scratched the surface. There’s a reason why GTA and other open-ended games like Morrowind, and the Sims are best sellers - people want to play their own story, not someone elses.

How long before this all becomes reality is hte next question. I hope during this coming generation.