next »

[lithander]

http://vindinium.org/

This looks really cool. To quote the site: "You have to take the control of a legendary hero using the programming language of your choice. You will fight with other AI for a predetermined number of turns and the hero with the greatest amount of gold will win."

The graphics are kinda cute and I actually like the game mechanics. Simple but deep enough to allow for a variety of strategies.

The only thing stopping me from writing my own bot is that it seems like the community isn't very active at the moment.

[quantumpotato]

Interesting, sort of a zelda-esque robocode

[lithander]

About 10 years ago back at university we had a graded robocode turnament! That was fun

[quantumpotato]

About 10 years ago back at university we had a graded robocode turnament! That was fun

Haha neat. So for AI in these kinds of games, you're following what other people are doing and in this case, trying to predict if someone's going for a gold chest then maybe you can't get it so you should go for another one? In the Robocode examples they seemed very reactive, I'm curious if there's much predictive AI going on..

[lithander]

In Robocode the main challenge was to predict the future position of enemies good enough that your shots would have a high hit-rate while at the same time moving in a erratic, random to make yourself a hard-to-hit target.

In Vindinium the first step would be to implement basic pathfinding and just walk from chest to chest to capture them and drink beer whenever you need it. A bot that could do that reliably is a decent single player bot. Then you add rules to make your bot react to the presence of your opponents. Wether you'd rather try to avoid enemies or seek them out deliberately to kill them - one strategy is not clearly better then the other strategy. So maybe you make it depend on your and the opponents health? Or on who's winning? Trying different behaviors under different conditions, balancing them against each other and improving your bot iteratively is the "classical" approach. If you'd want the AI to "learn" the pefect combination of parameters you could employ evolution programming or machine learning techniques. But I'd wager that most people wouldn't go that far. Classical AI is very much about figuring out the rules for good play and then implementing them in code, explicitly.