2.4 Q-Learning & Artificial Neural Networks

Q-Learning & Artificial Neural Networks

An Artificial Neural Network in gaming AI (Using AI in Computer Games - Assignment 1, 5.2 Learning Technologies - Artificial Neural Networks) is a piece of software which is used to make the AI in games more intelligent, it works with the heuristic method meaning the AI learns from its previous accounts. But unlike heuristics the ANN gains more information which means it can utilise the data it has gathered from previous encounters, it works more like a brain and some go on to say it can begin to out think the human mind.

So where could I apply Artificial Neural Networks to the AI in my game idea? Well in my game I have one player character and two zombie AI's, now when the player character enters room B and gets attacked and killed by the zombie AI's. If I was to allow it so the player can kill the zombies, then the zombie AI would re-spawn in the same location as before but with ANN applied to the AI, the AI will have learned and will remember where the player character is hiding and wont make the same mistake again by running into your path. This would be enough evidence to show that the AI has learnt where the player was where he killed the AI which results in the AI spawning in either a safer location or in a better location in which to attack the player character.

Where are Artificial Neural Networks used in today's games? Well actually ANN's have been applied in games to AI's for a long time now, even games as old as Red Fraction used this technique to make the AI more intelligent and harder for the player character. Newer games such as Battlefield 4 have AI which learns from the way you act as a player, it can learn to predict what you plan to do and then learn to out think you.



Now Q-Learning in gaming AI (Using AI in Computer Games - Assignment 1, 5.3 Learning Technologies - Q-Learning ) is a learning algorithm and the way this works is that an AI is rewarded or punished based on its previous actions. Sort of like a dog, if a dog behaves well, that dog will get a treat, but if the dog pee's in the house, that dog will be in trouble. Q-Learning is so linked to human life and ANN's that it combines the human way of life and apply's that to a gaming AI but I have already explained all this in my previous blog, so what we really want to know is if Q-Learning can be applied into my game.

Well for Q-Learning to work, i'll need to combine the two processes of trial and error and the process breaking a problem down into smaller problems. But because my game idea is very basic, to apply Q-Learning would take much more AI involvement and a bigger plot to be set. It links in with Artificial Neural Networks because of its heuristic approach, if I could make the zombie AI's attack each other but because they are not supposed to, this would result in a bad god situation and the zombie AI being punished. Now the AI could lose health, points or even die of its own actions.

(8)

Q-learning is used in more modern games we play today because the advancements in AI is much more in depth then it ever was, a prime example would be Forza on Xbox One, you can create a drivatar which is an AI which learns how you drive in the game by imitating and repeating your driving style. In the video link above, you can see how drivatar is used to improve the players driving skill, which shows advanced AI now just going against the player but assisting the player.