No Training, Better Flights: Test-Time Scaled VLMs for UAV Navigation

2026-07-21

AI Summary

This work addresses the challenge of suboptimal or unsafe trajectory generation in vision-language navigation for unmanned aerial vehicles operating in complex environments, a limitation often stemming from existing methods’ reliance on single-pass inference without error correction. To overcome this, the authors propose a training-free, test-time scaling approach that leverages a frozen vision-language model to generate multiple candidate trajectories in parallel. These candidates are iteratively refined through a self-correction mechanism guided by a multi-criteria scoring function that jointly evaluates safety, goal alignment, and forward progress. The proposed method significantly enhances both navigation accuracy and flight safety, achieving state-of-the-art performance on standard benchmarks without requiring additional training.

Abstract

Test-time scaling offers a promising method to improve the inference performance of Vision-Language Models (VLMs) without additional training. Existing approaches to vision-language navigation (VLN) for Unmanned Aerial Vehicle (UAV) typically relies on a single inference pass, which can falter in complex environments by producing suboptimal or unsafe trajectories. In this paper, we explore a simple and effective approach to apply test-time scaling to VLN for UAV. We enhance navigation reasoning through an iterative refinement process that requires no extra model training, guiding the model to re-evaluate its initial navigation plan for better accuracy and safety. Our method first prompts the model to generate multiple parallel candidates and then performs a self-correction step, achieving deeper and more robust planning without changing the underlying model. To further strengthen decision-making, we design a multi-criteria scoring function to evaluate the refined candidates based on safety, goal alignment, and forward-progress. This simple yet powerful combination enables a frozen UAV navigation VLMs to self-correct and generate more accurate and reliable flight plans, achieving SOTA performance in this task.

Problem

• Vision-Language Navigation
• UAV
• Test-time Scaling
• Trajectory Safety
• Navigation Accuracy

Innovation

• test-time scaling
• vision-language navigation
• UAV
• self-correction
• multi-criteria scoring