SpatialUAV: Benchmarking Spatial Intelligence for Low-Altitude UAV Perception, Collaboration, and Motion

2026-06-26

AI Summary

This work addresses critical limitations in existing low-altitude UAV benchmarks, which inadequately support 3D spatial reasoning, multi-view collaboration, dynamic modeling, and task diversity. To bridge this gap, we introduce SpatialUAV, the first unified evaluation benchmark tailored to real-world low-altitude scenarios. It encompasses 14 fine-grained tasks structured as visual input–question–answer triples, accommodating multimodal inputs and heterogeneous outputs. Data quality is ensured through detector-assisted region extraction, depth supervision, rule-based generation, human annotation, and blind screening, while task-specific metrics are carefully designed for precise evaluation. Comprehensive assessments across three categories of vision-language models reveal substantial performance gaps compared to human capabilities—particularly in cross-view association, structured localization, geometric reasoning, and temporal understanding—highlighting key bottlenecks in spatial intelligence for UAV systems.

Abstract

Spatial intelligence is essential for low-altitude unmanned aerial vehicle (UAV) perception, collaboration, and navigation. However, existing UAV benchmarks often emphasize image-level recognition, single-view understanding, or narrow answer formats, leaving 3D spatial inference, multi-view collaboration, scene dynamics, and diverse task formulations insufficiently evaluated. To address these gaps, we introduce SpatialUAV, a real low-altitude UAV benchmark comprising 4,331 curated instances across 14 fine-grained task types, covering semantic discrimination, spatial relation, aerial--aerial collaboration, aerial--ground collaboration, and motion understanding. SpatialUAV organizes all samples into a unified visual-input--question--answer schema, while supporting seven input configurations and nine answer formats, including option labels, region identifiers, geometric values, cross-view correspondences, and free-form motion descriptions. To ensure reliable and grounded evaluation, our data construction pipeline integrates detector-assisted regions, depth supervision, metadata-derived rules, extensive manual annotation, blind filtering, and multi-turn human validation, together with task-specific metrics for heterogeneous outputs. Evaluating representative vision-language models across three categories, we show that current models remain far from human-level performance, with pronounced bottlenecks in cross-view association, structured grounding, geometric reasoning, and temporal viewpoint understanding. These results offer empirical guidance for advancing low-altitude UAV spatial intelligence. Code and data are available at https://github.com/Hyu-Zhang/SpatialUAV.

Problem

• spatial intelligence
• low-altitude UAV
• 3D spatial inference
• multi-view collaboration
• scene dynamics

Innovation

• spatial intelligence
• low-altitude UAV
• multi-view collaboration
• 3D spatial reasoning
• vision-language benchmark