Policy Gradient Methods

Return-to-Go

Sum of discounted future rewards from a given time step, used as a gradient estimator in policy gradient algorithms.

← Wstecz