Multi-Camera Disentanglement for Reinforcement Learning to Enable Zero-Shot Generalization

A self-supervised approach to learn a disentangled representation from multiple camera views that enables zero-shot generalization to a single camera for reinforcement learning tasks.