NAS: A Neural Attention-based SLAM for Autonomous Navigation in Unknown 3D Environments

Abstract

Navigation in unknown 3D environments aims to progressively find an efficient path to a given target goal in unseen scenarios. A challenge is how to explore the navigation quickly and effectively. An end-to-end learning approach is proposed to extract geometric shapes from RGB images, but it is not suitable for large environments due to its exhaustive exploration with exponential search space. Active Neural SLAM (ANS) presents a semantic map to maximize exploration coverage to tackle the object goal navigation task, but ANS still visits the explored areas frequently since the unproper local target selection. In this paper, we propose a neural attention SLAM (NAS) model to explore unknown 3D environments. Spatial attention is provided to quickly identify colorful obstacles (such as tea table or floor). We also leverage the priority of unknown regions in the short-term goal decision to avoid frequently exploration with a channel attention. The experimental results show that our model can build a more accurate map than ANS and other baseline methods with less running time. In terms of relative coverage, NAS achieves a 0.5$\%$ improvement over ANS in overall and a 1.1$\%$ improvement over ANS in large environments.