Solving the scalarization issues of Advantage-based Reinforcement Learning algorithms

In this research, some of the issues that arise from scalarization multi-objective optimization problem in Advantage Actor Critic (A2C) reinforcement learning algorithm are investigated. The paper shows how a naive can lead to gradients overlapping. Furthermore, possibility entropy regularization term be source uncontrolled noise is discussed. With respect above issues, technique avoid gradient overlapping proposed, while keeping same loss formulation. Moreover, method noise, by sampling actions distributions with desired minimum entropy, Pilot experiments have been carried out show proposed speeds up training. approach applied any Advantage-based Reinforcement Learning algorithm.