Substantial progress has been achieved in embodied visual navigation based on reinforcement learning (RL). These studies presume that the environment is stationary where all the obstacles are static. However, in real cluttered scenes, interactable objects (e.g. shoes and boxes) blocking the way o…