1 paper · 1 filter
SM Mazharul Islam, Manfred Huber
World models for visual control typically learn compact latent states by reconstructing observations, implicitly encouraging representations to preserve information across the enti…