SEDualVLN: A Spatially-Enhanced Dual-System for Vision-Language Navigation

2026-05-17

AI Summary

This work addresses the limitations of existing vision-and-language navigation (VLN) methods, which often lack dynamic reasoning capabilities in long-horizon tasks or suffer from inaccurate spatial localization and low inference efficiency under zero-shot settings. To overcome these challenges, the authors propose a dual-system collaborative framework: System 1 leverages a vision-language model (VLM) endowed with both global and local spatial awareness to generate actions, while System 2 performs zero-shot path planning by integrating a multimodal large language model (MLLM) with a dynamically constructed 3D map. The two systems operate via a fast-slow coordination mechanism to enable efficient and robust navigation. This approach pioneers the integration of a spatially augmented dual-system architecture into VLN, effectively combining the strengths of end-to-end policy learning and zero-shot planning, thereby significantly enhancing directional reasoning and generalization. It achieves state-of-the-art performance on the VLN-CE benchmark.

Abstract

Vision-Language Navigation (VLN) approaches have currently followed two primary paradigms: the end-to-end Vision-Language Model (VLM) policy fine-tuned on navigation trajectories to directly predict actions, and the zero-shot modular pipeline integrating pre-trained Multimodal Large Language Model (MLLM) for training-free generalization to unseen environments. However, end-to-end methods struggle with long-horizon navigation and lack dynamic reasoning, whereas zero-shot methods are constrained by limited spatial grounding for reliable planning and also require substantial reasoning time. To bridge this gap, we introduce SEDualVLN, a spatially-enhanced dual-system VLN framework. System 1 is a VLM model enhanced with both global and local spatial awareness, used for action generation. System 2 integrates a general MLLM with a mapping module, wherein the MLLM plans waypoints by leveraging top-down views of the real-time 3D map alongside streams of rendered path images. Both systems leverage different forms of spatial enhancement to cultivate the agent's sense of direction in VLN tasks. Ultimately, they cooperate to complete the navigation task through a fast-slow coordinated approach. SEDualVLN achieves state-of-the-art performance on VLN-CE benchmarks, and further ablation studies demonstrate the effectiveness of each system and module.

Problem

• Vision-Language Navigation
• spatial grounding
• long-horizon navigation
• dynamic reasoning
• zero-shot generalization

Innovation

• Vision-Language Navigation
• Spatial Enhancement
• Dual-System Architecture
• Multimodal Large Language Model
• 3D Map Grounding