Lessons learnt applying AI to levels

[Pixel Perfect]

It's been an interesting experience applying AI and NPC movement to what I thought was a relatively realistic game level design and some of the pitfalls of using 'off the shelf' static and character models.

 

Some of the issues that emerge are as follows:

 

The realisation that no matter how nice some of these models or sets of models might look when you first see them and even place them in your level design, how fit for in game purpose are they really and have they ever really been tested in game? The reality I'm finding is that there are often inconsistencies or problems inherent in the design which produce problems when walking characters through my world.

 

A typical one will be differences in door heights between buildings (even in supposedly matched sets of buildings that come as a pack).

 

Another example is geometry that simply doesn't suit the physics. So you get steps deliberately modelled to look worn ... very authentic ... problem is, without a collision mesh there is no way your character controller is even attempting to walk up them, and of course ... no collision mesh provided!

 

Walkways provided on walls which in practice prove to be too narrow for two NPCs to practically pass each other.

 

Characters that lack certain animations which, once you actually have them interacting in the world, becomes blatantly obvious. Hey, even one character model I converted which seemed ideal for the part until till I discovered it had no Idle animations!!!

 

It's not that any of these issues are insurmountable but they do cause a lot of additional work or rework and considerably extend the time taken to reach the desired goal!

 

... and the moral of the story is ... work directly in conjunction with an artist or artists where a coherent approach to style and requirements is established and issues that arise can be quickly rectified. Or pay big bucks to have bespoke work done where you can go back to the artist where issues occur.

[AggrorJorn]

Nice post Pixel, you might want to conside this for a blog. It is really interessting to read about your experience. It is something that every (AI)programmer will come accross.

[cassius]

I agree with every word pixel. I have had similar problems. I design my own levels and buildings so that they will suit the characters that I purchase. I had to make doorways smaller because they were three times the height of the character. I also had to make passageways wider so the camera would not get stuck behind a wall and lose sight of the (third person character. I also need characters to pick up objects but theres no supplied anim for that.

But as a hobbyist with no commercial interests I don't want to have partners. Solving the problems and doing almost everything myself is part of the hobby.But it puts restrictions on my game which iritate me.

[Pixel Perfect]

@Aggror I'll certainly consider a blog at the end of this process as there is a lot I've learnt and presumably still a lot more to be learnt

 

@cassius Nice to read your feedback and experience. I accept it's not always possible to team up with others for various reasons but as you say it places restrictions on what you can achieve. I can handle simple modelling tasks but anything more complex is beyond me.

[knocks]

Did you bump your head climbing to the second story? It sure is disappointing when you come across such issues. I personally have put together a great list of average sizes, be it a door size, street width, ceiling height, brick size, height of a cat the list is always growing for every project I take on. It’s a same so many ignore the importance of such details.