sjtu-marl / malib

A parallel framework for population-based multi-agent reinforcement learning.
https://malib.io
MIT License
499 stars 60 forks source link

Mappo+gfootball #30

Closed KornbergFresnel closed 2 years ago

KornbergFresnel commented 2 years ago

Summary: this branch implemented the MAPPO algorithm and a complex testbed google-football. moreover, some changes to the rollout mechanism were made in this branch. This branch will be archived and does not accept any new updates (https://github.com/sjtu-marl/malib/releases/tag/archive/mappo+gfootball).

Details

  1. add general RNN model support for the policy implementation.
  2. test cases for rollout, environment apis and dataset apis.
  3. componentize rollout process, but doesn't support sequential rollout yet. :(, @KornbergFresnel will work on it (see branch evaluation-worker)