Problems of QTRAN about marl HOT 1 CLOSED

Sorry for the late reply. In the QTRAN paper, the author is only experimenting on Matrix game, Multi-domain Gaussian Squeeze (MGZ) and Modified Predator-prey (MPP). In the Appendix C.1. Environment,

• about MGZ, the paper described as,

This task covers a fully observable case where all agents can see the entire state.

• about MPP, the paper described as,

Each agent’s observation includes its own coordinates, agent ID, and the coordinates of the prey relative to itself, if observed.

So I speculate that in MGZ the global state is available and used, but in MPP, the global state is not available and not used.

from marl.