Multi-Agent Opponent Trajectory Modeling
-
this folder is used to train RL agent, including (dpiqn / mbom / rbom) with (qplex, vdn, qmix); where dpiqn, mbom and rbom are represented as drpiqn, MA-OTM and MA-RDOM in the paper, respectively.
-
please install all packages and sc2 properly in the beginning (https://github.com/oxwhirl/pymarl/);
-
'python ...' commands can be seen in run_exp.sh
-
uncomment 'sys.argv' in main.py to debug
- Please pay attention to the version of SC2 you are using for your experiments.
- Performance is not always comparable between versions.
- The results in SMAC (https://arxiv.org/abs/1902.04043) use SC2.4.6.2.69232 not SC2.4.10.