This list describes the planned features including breaking changes.
- 10 diverse environments available
- 18 algorithms available strictly under CDTE and Dec-POMDP
- learning curve released on over 1000 experiments cover four popular multi-agent tasks.
- API-based usage
-
test suits + coverage
-
support Windows OS / MacOS
-
manually inference, refer to issue: #86 (comment)
-
new environments
- MATE: https://github.com/UnrealTracking/mate
- Go-Bigger: https://github.com/opendilab/GoBigger
- Voltage Control: https://github.com/Future-Power-Networks/MAPDN
- Overcooked: https://github.com/HumanCompatibleAI/overcooked_ai
- CloseAirCombat: https://github.com/liuqh16/CloseAirCombat
- HideAndSeek: https://github.com/jiayu-ch15/Variational-Automatic-Curriculum-Learning#3-train-hide-and-seek
- SISL: https://github.com/sisl/MADRL
- Melting Pot: https://github.com/deepmind/meltingpot
-
support Transformers
-
support MultiDiscrete action space