I’m a computer developer and I’m passionate about artificial intelligence.
So you probably understood the subject of this topic, being a developer you’ll also think that I’ll talk about programming flaws, or current problems with the game, this is not the subject so please disregard this side (I’m just warning you).
Today I’m calling on the community to push the devs to create an API in a common language (Python, C++, Rust…) that will allow us to develop a powerful AI based on current research (Basically Q-Learning, Long Short Term Memory, Reccurent Neural Network, Genetic Learning etc…) because yes your language is cool but no.
Why not?
Because :
It doesn’t allow us to have matrix or variable (float & numpy etc…).
Well, I won’t explain more about why because the first point defines everything.
Conclusion:
I think that the game offers new availability to AI, that SC2, on this one the micro-management is “faster”, the decision making less punitive due to the production time or the harvest time which is “shorter”, the fact that there are only 2 resources, 3 civ and that the maps are “smaller”. We have DeepMind AlphaStar which are a real masterpiece. But I am convinced that AOE II will not be as “easy” to implement because for me AOE II is much more punitive, in the sense that the game is slower and we have more stratagem to carry. This means that the complexity of the choice of the strategy to apply at time T is much more difficult and varies much more than on SC2 which is more monotonous.
Moreover it will allow to learn new technique, new approach, new build order, even beat the top ladder, which will motivate many players, because an AI better than the top 1, allows us to hope to be one day
So let’s get out this API !
Let’s show what AOE:II’s devs are up to.
Thanks for reading.
Please support us with a comment or a question.
AIs APM needs to be limited. Else it just abuses its infinite APM by exploiting ranged units (especially monks).
APM of Starcraft 2 AlphaStar AI was limited in order to make it play more similarly to humans. It is also more interesting if AI wins by being smart, not by doing 10 000 things in one second.
Here’s what you need to know, here’s what you need to practice! Thank you for your support. If you want to communicate with me, mp I will give you my discord.
I think I have found a solution to create a Recurrent Neural Network, with the game recs. This allows us to have an ai to create the best build order of the game !
So don’t hesitate to contact me if you want to participate!
A machine learning based AI very difficult to pull off for aoe2, but recording hotkey key presses and mouse clicks from experts games, and train it based on that data could be a start. These can be extracted from rec games
The challenge is also that there needs to be a ‘fitness’ in order to steer the AI in the right direction. Ultimately this needs to be ‘winning the game’ vs ‘loosing the game’ but this seems too far stretched since aoe2 is:
Hey, I’m here to try helping this issue, have you created a repository ? If not, let me create one to upload our possible solutions. I’m tryin to “replicate” the same way that sentdex did with his “Ai plays gta v” series, you should check out.
I think that the hardest challenge is to retrieve real time info about the game, did you came up with somithin?
