MerNav: A Highly Generalizable Memory-Execute-Review Framework for Zero-Shot Object Goal Navigation

TL;DR

MerNav introduces a Memory-Execute-Review framework for Visual Language Navigation to address the weak generalization of existing SFT and TF methods in Object Goal Navigation. By integrating a hierarchical memory system with an Execute module for routine decisions and a Review module for anomaly handling, the approach achieves strong performance under both Training-Free and Zero-Shot settings across four datasets, including HM3D_v0.1 and HM3D_OVON, and even surpasses some SFT baselines on MP3D and HM3D_OVON. Ablation studies show complementary contributions from memory, commonsense memory, review strategies, and execution refinements, while case studies demonstrate improved interpretability through structured reasoning and multi-view analysis. The work highlights the practical potential of memory-guided, multi-stage reasoning for robust, generalizable VLN in realistic, open-world settings.

Abstract

Visual Language Navigation (VLN) is one of the fundamental capabilities for embodied intelligence and a critical challenge that urgently needs to be addressed. However, existing methods are still unsatisfactory in terms of both success rate (SR) and generalization: Supervised Fine-Tuning (SFT) approaches typically achieve higher SR, while Training-Free (TF) approaches often generalize better, but it is difficult to obtain both simultaneously. To this end, we propose a Memory-Execute-Review framework. It consists of three parts: a hierarchical memory module for providing information support, an execute module for routine decision-making and actions, and a review module for handling abnormal situations and correcting behavior. We validated the effectiveness of this framework on the Object Goal Navigation task. Across 4 datasets, our average SR achieved absolute improvements of 7% and 5% compared to all baseline methods under TF and Zero-Shot (ZS) settings, respectively. On the most commonly used HM3D_v0.1 and the more challenging open vocabulary dataset HM3D_OVON, the SR improved by 8% and 6%, under ZS settings. Furthermore, on the MP3D and HM3D_OVON datasets, our method not only outperformed all TF methods but also surpassed all SFT methods, achieving comprehensive leadership in both SR (5% and 2%) and generalization.

Figures (5)

• Figure 1: Our Memory-Execute-Review VLN framework, supported by a hierarchical memory structure, can handle regular and exceptional situations using the execute and review modules.
• Figure 2: Overview of Memory-Execute-Review.
• Figure 3: Panorama.
• Figure 4: Jump over obstacles.
• Figure 5: Jump out of floor.