Dreaming Smoothly and Sample Efficiently with Gradient Penalized Latent Dynamics
Signal
72
Hype
15
In three linesGPLD adds gradient-penalized latent dynamics regularization to DreamerV3 to encourage smooth transition learning in latent space. Tested on DeepMind Control, GPLD improves sample efficiency, with strong gains on complex locomotion and quadruped tasks.Read source
Your take?
Summary generated by Claude — human-verified