VISTA: Scale-Aware Visual Navigation via Action History Conditioning

2026-06-15

AI Summary

This work addresses trajectory distortion caused by action normalization and poor robustness in visually repetitive environments in existing visual navigation models. The authors propose an end-to-end navigation policy that, for the first time, incorporates normalized action history as a conditional input to explicitly model the mapping between predicted actions and actual displacements. By integrating the DINOv3 visual encoder, the method enhances spatial and geometric awareness, effectively mitigating scale sensitivity and significantly improving navigation performance in texture-poor environments. Evaluated in real-world zero-shot scenarios—including outdoor, forest, and office settings—the approach achieves 100% target prediction accuracy and an average checkpoint success rate of 95%, demonstrating exceptional cross-domain path-following capability.

Abstract

Vision Navigation Foundation Models (VNMs) promise end-to-end learned navigation policies capable of zero-shot deployment across diverse embodiments and environments. To maintain generality, many vision-based navigation models predict normalized actions. However, this normalization introduces a critical deployment vulnerability: applying different scaling factors to the same normalized trajectory alters its physical geometry, which degrades navigation performance and increases collision risks. We address this vulnerability by conditioning the model on normalized action histories alongside image observations, providing explicit context on the relationship between the model's predictions and the robot's actual physical displacement. Furthermore, current VNMs often struggle in visually repetitive environments that lack distinct features. To resolve this issue, we integrate a DINOv3 encoder, whose richer representations enable our model to capture both spatial and geometric dimensions between observations. VISTA generalizes robustly to out-of-distribution environments, achieving 100% goal prediction accuracy in zero-shot, real-world deployment in Outdoor, Forest and Office settings, and an average of 95% checkpoints crossed, demonstrating consistent path following in unseen environments.

Problem

• visual navigation
• action normalization
• scale sensitivity
• visually repetitive environments
• zero-shot deployment

Innovation

• action history conditioning
• scale-aware navigation
• vision navigation foundation models
• DINOv3 encoder
• zero-shot generalization