Policy Gradient with Advantage Function Formula

A2C Loss Function Formulation

In a reinforcement learning scenario, an agent is in a particular state. The estimated value of being in this state, averaged over all possible actions the agent could take, is +10. If the agent chooses a specific action, the estimated value of taking that particular action in that state is +8. Based on this information, what can be concluded about this specific action?

If an action has a positive advantage value, it means that taking this action is guaranteed to result in a higher immediate reward than any other action available in that state.

Interpreting Action Advantage