Masked World Models for Visual Control

Younggyo Seo, Danijar Hafner, Hao Liu +4

Visual model-based reinforcement learning (RL) has the potential to enable sample-efficient robot learning from visual observations. Yet the current approaches typically train a si…