waterstein-distance

This paper introduces RQIQN, a robust quantile-based method for distributional reinforcement learning that uses Wasserstein geometry regularization to prevent distribution degeneration and improve performance in risk-sensitive tasks.