Moving Beyond Navigation with Active Neural SLAM

Anonymous

Keywords: navigation, reinforcement learning, domain generalization

Abstract: The ability to effectively harness autonomous control in real-world 3D environments largely depends on learning realistic navigation techniques for embodied agents. Advances to classical robotics call for efficient methods to navigate, viz., explore unseen simulations while transferring the attained performance to the real-world domain. Any autonomous agent in the wild should have the ability to make decisions on the fly, having explored its surroundings and being cognizant of sudden changes, as in the real world. This blog post breaks down the method of Active Neural SLAM for exploring 3D environments, moving away from end-to-end exploration, by learning policies in a modular and hierarchical manner. Active Neural SLAM enabled Domain Generalization to previously unseen environments and Task Generalization in navigation while curbing the high sample complexity of previous methods. We also suggest ways to leverage this method for scene understanding and learning agent-scene interaction by combining navigation with motion prediction or synthesis.

ICLR Paper: https://openreview.net/pdf?id=HklXn1BKDH, https://openreview.net/pdf?id=SyMWn05F7