Jump to content
  • Advertisement
Sign in to follow this  
ErUs

learning how to play a game.

This topic is 4651 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

are there any example of ai that learns to play a game? for example ai that is aware of some inputs and outputs but has no hard-coded knowledge of what is required to win a game. It would then try to work out why he was failing to reach an objective. sort of thing.

Share this post


Link to post
Share on other sites
Advertisement
I think you could make it similar to pathfinding. You could give the system a goal and a start but it has to find his way through the obstacles (in this case, mistakes)

-Zubair-

Share this post


Link to post
Share on other sites
you need to at least be able to tell your system something akin to pleasure and pain, right and wrong, yes and no.

then devise a system that will learn through trial and error how to react to different situations in terms of your pain/pleasure paradigm system.

in the case of pac-man i would hardcode my system to sense pleasure in eating the little yellow balls, whereas dying would cause pain.

pleasure merely re-inforces the last act that took place before the pleasure occured.

pain jumbles perception/thought giving a sense of void and anti-reinforcement in the last action.

it could end up being a huge project, but it would be worth it if you made it play well enough to play multiplayer, hehe.

Share this post


Link to post
Share on other sites
The biggest issue that see is when the actions that lead to a pleasure inducing or pain inducing state are complex and/or time delayed, it is difficult to decide what to reinforce.

In a situation where a single action produces a reward, such as "move towards food" = good, the implementation is simple. If it is a situation where you have multiple inputs and multiple rewards, such as "move left wheel left, right wheel right, apply gas, remove brake" => reduces distance to nearest food but also reduces distance to nearest enemy = bad if enemy is close, it gets more difficult to determine what to reward.

This is where it is nice to have genetic algorithms to do the fitness testing for you, but that doesn't solve the problem of teaching the current agent what to do.

Share this post


Link to post
Share on other sites
I don't mean to sound lame, but you should really check out some of the books on the subject of AI. There has been a ton of study in the subject of training an machine learning. In particular neural networks are very good at pattern recognition.

The concept of breaking the problem up into a state space is a good and simple abstraction of the problem, but I think that you will find that representing learning in a dynamic way is a very complex task. I do not know very much about it, but I hope that you are successful.

Share this post


Link to post
Share on other sites
Quote:
Original post by ErUs
are there any example of ai that learns to play a game?

"Game" is too generic a word. What type of game? Chess? Doom?

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement
×

Important Information

By using GameDev.net, you agree to our community Guidelines, Terms of Use, and Privacy Policy.

We are the game development community.

Whether you are an indie, hobbyist, AAA developer, or just trying to learn, GameDev.net is the place for you to learn, share, and connect with the games industry. Learn more About Us or sign up!

Sign me up!