Framework

Our VMAgent’s framework is mainly based on the pymarl. It consists controller, learner, components, modules and utils.

Controller

The controller plays the role on output actions for sampling.

Learner

The learner plays is to update the agent’s policy

Components

It provides key components for learning and sampling. It includes the implementations of replay memory and action selector (i.e., epsilon-greedy action selection).

Modules

It provides different networks of the agent, including critic network, policy network etc.

Utils

It provides several utils for reinforcement learning.