wholemoley
Joined 8 years ago
Joined 8 years ago
They're rewarded by exploring, not by values in game.
Maybe alone they're not enough, but in conjunction with other things, I bet we could beat Zelda. I had the bot exploring enough to find the first dungeon of LoZ.
I used it to train an AI to play Mario Kart ( https://www.youtube.com/watch?v=A8oSnh0M864 )
I also had it playing The Legend of Zelda (got as far as finding the first dungeon, but with more power, I'm certain it could explore the whole map.).
Here it is: https://www.youtube.com/watch?v=rCggOcKZn-c
(There's a fun interaction at 25:52)
Exploring this topic is currently my primary hobby. Specifically, I've been using OpenAI's retro (Sonic, Contra, Mario, Donkey Kong and, more recently FZero) and comparing the ancient NEAT with more fashionable stuff like DQN, PPO, A3C and DDPG.
With my extremely limited experience, NEAT seems to outperform all of these other algorithms. I believe the advantage is the potential for strange/novel network structure.
And the best part is that NEAT doesn't require a powerful GPU.
Apologies for the shameless plug but here's a link to a series on youtube I made about using Retro and NEAT together to play Sonic. https://www.youtube.com/watch?v=pClGmU1JEsM&list=PLTWFMbPFsv...
Ghostbusters!
I even made a tutorial series. https://www.youtube.com/watch?v=pClGmU1JEsM&list=PLTWFMbPFsv...
Apologies for the shameless plug.