- install python
pip install pytorchreference pytorch websitepip install unityagents- download reacher env from Reacher_Linux
- Run
components/env_tst.pyto test the environment work all right. - Run
train_maddpg.pyto train the agent. This module is for the vector state space. - Run
test_maddpg.pyto test the agent interact with Env. - The default config file is
components/config_maddpg.py. You can modify the default parameter value to retrain the agent.
- XUbuntu 18.04
- CUDA 10.0
- cudnn 7.4.1
- Python 3.6
- Pytorch 1.0
- yacs v0.1.5
- num agents: 2
- action space: 2 continuous action.
- state space: 24 states
- [version] The agents must get an average score of +0.5 (over 100 consecutive episodes, after taking the maximum over both agents)
MADDPG.
- Package agent contains the MADDPG agent.
- Package components contains the config files for agent, envs and util functions.
- Package network contains the agent policy network.