DeepMind Publishes StarCraft II Learning Environment | Two Minute Papers #182
Shared from (source: https://www.youtube.com/watch?v=St5lxIxYGkI)
Hurcelean task
To play the game, deep reinforcement learning observing environment and choosing next action to maximize score.
Starcraft 2 vs. GO
-
Dota 5.5
SC 1v1
– How do we know we are winning or scoring (post game: W / T L). This information is not driven
win battles
player with more resources does not have enough time to spend it.
To beat a formidible player:
T
feature layers encode: relevant information
terrain height
hitpoints
camera