This work bases on my master's thesis.
we propose a hierarchical DDPG algorithm to solve the Multi-Agent cooperation problem
Main task: StarCraft1 small scale combat(unit micro control)
Mian problem: Multi-Agent cooperation problem
Performance:
Voide link to youku:

gym-starcraft-modified
It's depends on TorchCraft, I warp an SimpleBattleEnv for multi_units battle, you can follow the installation above.

pytorch
gym

• remember change the paramater(MYSELF_NUM, ENEMY_NUM) to fit the map setting
• model will be saved every SAVE_ITERVAL episodes

python3 sc1_train_hierarchical.py

• change the saved_floder and episode_num to chose the model you want to load

python3 test hiernet.py