Hey everyone,
Actually, I’m a software developper and I dont have any experience in game programming, so i need some materials.
Is there any good books about AI? I mean one with many concepts that you can use to build a specific AI?
I dont have any problem to code my ideas, but the problem is that I’m not use to any concept usefull for games like Platinium Rift. I just discover scent algorythm, but that seems hard to confugure (I’m around top 120 right now).
Do you guys build a grah of each possibely step? (like in a maze)
Do you use area algorytm?
Or only shortest path?
I you could just give me some keywords, ideas that would be great. I don’t need code, just some how-to-ideas that are widespread in the world of game.
I have a more theoretical background than games development - so there may be game-specific approaches I am missing - AI is basically making decisions automatically - so there are various ways to approach this in games:
- With infinite time and resources, we can play all possible games and pick the best path
– But you don’t have this and the state space for most interesting games is prohibitively large, not to mention the limited time you have each turn.
- To choose a good algorithm, you have to think about the type of game you are looking at:
– Is the game zero-sum? i.e. do all players compete for a fixed pool of resources
– Is there any element of randomness in the environment?
– etc.
- In addition to choosing a good algorithm, the big pay-offs are usually in devising a good representation of the game to minimise the search space/time and tweaking the algorithm for the particular game.
Things you might want to look up:
- Minimax/Maximin
- Monte Carlo Search
- Simulated Annealing
- The concept of an ‘any-time algorithm’
- Evaluation function (for estimating the value of a particular configuration of the board)
- Some systems use multiple simpler AI with specific tasks (strategy, tactics, resource gathering, etc) and combine the votes each turn, can’t remember the technical name for this.
NB - some games are so simple that they can be played ‘perfectly’ with a fixed set of rules e.g. tictactoe, connect four, checkers - I remember reading somewhere that these have been ‘solved’ in the sense that you can always win just by not making a mistake. I am not sure if you have to go first in these cases.
2 Likes
Here is a free class where you can learn more about AI, taught by Sebastian Thrun and Peter Norvig. https://www.udacity.com/course/cs271
Norvig was previously the head of Computational Sciences Division at NASA Ames Research Center, and is now the Director of Research at Google.
He is the co-author of: “Artificial Intelligence, A Modern Approach”, ISBN: 0-13-604259-7
used in many university level AI classes:
4 Likes