FAQ

Why ma-gym?
It was created from the need to have simple set of multi agent environments which could be easily accessed from other projects much like gym. These environments could be used to understand core learning challenges in multi agent settings.

Why do we have a list of action spaces?
This ensures that each agent can have a different action space if required.

Why do we have a list of terminals?
In some cases, some agents may die before others which means actions performed on these agents wouldn't have any effect on the environment. However, you can still control the remaining agents. An episode is considered to be terminated when all agents die. One may be able to exploit this early death of some agents into their agent learning process.

Why do we have a list of observations?
This involves local observation of each agent. Also, It's makes it possible for each agent to have a different shaped local observation.