Tag
GPLD introduces a gradient-penalized latent dynamics regularizer for DreamerV3 to enforce local smoothness in transition learning, improving sample efficiency on continuous control tasks, especially complex locomotion.