GridMM: Grid Memory Map for Vision-and-Language Navigation

TL;DR

This paper introduces GridMM, a novel top-down grid memory map for VLN that effectively captures spatial relations and visual clues, improving navigation performance in various 3D environments.

Contribution

We propose a dynamic, egocentric grid memory map for VLN that better models spatial relations and visual details compared to existing memory methods.

Findings

Outperforms existing methods on REVERIE, R2R, SOON datasets

Effective in both discrete and continuous environments

Demonstrates superior navigation accuracy

Abstract

Vision-and-language navigation (VLN) enables the agent to navigate to a remote location following the natural language instruction in 3D environments. To represent the previously visited environment, most approaches for VLN implement memory using recurrent states, topological maps, or top-down semantic maps. In contrast to these approaches, we build the top-down egocentric and dynamically growing Grid Memory Map (i.e., GridMM) to structure the visited environment. From a global perspective, historical observations are projected into a unified grid map in a top-down view, which can better represent the spatial relations of the environment. From a local perspective, we further propose an instruction relevance aggregation method to capture fine-grained visual clues in each grid region. Extensive experiments are conducted on both the REVERIE, R2R, SOON datasets in the discrete environments, and the R2R-CE dataset in the continuous environments, showing the superiority of our proposed method.