Is your VLM Sky-Ready? A Comprehensive Spatial Intelligence Benchmark for UAV Navigation

2025-11-17

AI Summary

Current vision-language models (VLMs) lack rigorous spatial intelligence evaluation for drone navigation and exhibit limited capabilities in dynamic environment understanding and navigation decision-making. To address this, we introduce SpatialSky-Bench—the first aerial-scene-oriented spatial intelligence benchmark—comprising 13 fine-grained tasks across two categories: environmental perception and scene understanding. Concurrently, we release SpatialSky-Dataset, a large-scale, multi-scenario dataset containing one million high-quality annotated samples. Methodologically, our approach integrates multi-granularity spatial reasoning, modular task modeling (e.g., distance estimation, altitude analysis, landing safety assessment), and end-to-end vision-language joint training. Experimental results reveal that mainstream VLMs underperform significantly on this benchmark; in contrast, Sky-VLM—a model specifically trained on SpatialSky-Bench—achieves state-of-the-art performance across all tasks, substantially enhancing spatial comprehension and autonomous navigation capabilities in complex airspace.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Vision-Language Models (VLMs), leveraging their powerful visual perception and reasoning capabilities, have been widely applied in Unmanned Aerial Vehicle (UAV) tasks. However, the spatial intelligence capabilities of existing VLMs in UAV scenarios remain largely unexplored, raising concerns about their effectiveness in navigating and interpreting dynamic environments. To bridge this gap, we introduce SpatialSky-Bench, a comprehensive benchmark specifically designed to evaluate the spatial intelligence capabilities of VLMs in UAV navigation. Our benchmark comprises two categories-Environmental Perception and Scene Understanding-divided into 13 subcategories, including bounding boxes, color, distance, height, and landing safety analysis, among others. Extensive evaluations of various mainstream open-source and closed-source VLMs reveal unsatisfactory performance in complex UAV navigation scenarios, highlighting significant gaps in their spatial capabilities. To address this challenge, we developed the SpatialSky-Dataset, a comprehensive dataset containing 1M samples with diverse annotations across various scenarios. Leveraging this dataset, we introduce Sky-VLM, a specialized VLM designed for UAV spatial reasoning across multiple granularities and contexts. Extensive experimental results demonstrate that Sky-VLM achieves state-of-the-art performance across all benchmark tasks, paving the way for the development of VLMs suitable for UAV scenarios. The source code is available at https://github.com/linglingxiansen/SpatialSKy.

Problem

• Evaluating spatial intelligence capabilities of Vision-Language Models for UAV navigation
• Addressing performance gaps in VLMs for complex aerial navigation scenarios
• Developing specialized models for spatial reasoning in unmanned aerial vehicles

Innovation

• SpatialSky-Bench evaluates VLM spatial intelligence for UAVs
• SpatialSky-Dataset provides 1M annotated samples for training
• Sky-VLM model achieves state-of-the-art UAV navigation performance