AeroGround: A Comprehensive Benchmark for Aerial-Ground Collaborative Reasoning

2026-08-12

AI Summary

This study addresses the critical lack of evaluation benchmarks for air-ground collaborative reasoning in Vision-Language Models (VLMs) by constructing a comprehensive dataset comprising 29,000 observations and 2,250 question-answer pairs. Generated within a high-fidelity simulation environment, this benchmark fills a significant gap in cross-view multimodal assessment. We systematically evaluated sixteen models on tasks involving cross-view correspondence and spatial reasoning. Results indicate that the state-of-the-art model achieves only 54.4% accuracy, substantially underperforming human baselines at 93.3%. These findings expose substantial performance limitations in current VLMs regarding air-ground collaboration. Consequently, this work establishes a foundational evaluation framework and identifies key research directions to advance air-ground collaborative embodied intelligence.

Abstract

Vision-language models (VLMs) have been widely employed in understanding and reasoning tasks for unmanned aerial vehicles (UAVs). Existing UAV benchmarks primarily focus on aerial-view scenarios. However, whether current VLMs can perform well on understanding and reasoning tasks in aerial-ground collaborative scenarios which are practical in real-world applications like rescue and infrastructure inspection remains underexplored. To address this gap, we introduce AeroGround, a comprehensive benchmark for evaluating VLMs in aerial-ground collaborative reasoning. AeroGround is built upon a simulated aerial-ground dataset containing approximately 29,000 multimodal observation groups from diverse open environments, and provides 2,250 high-quality question-answering instances covering cross-view correspondence, spatial understanding, and reasoning. Experiments on 16 pretrained VLMs, together with two domain-adapted variants, reveal a substantial gap between current models and human performance: the best model achieves an average accuracy of 54.4%, whereas humans reach 93.3%. By systematically revealing the strengths and limitations of existing models in aerial-ground collaborative reasoning, AeroGround provides a foundation for developing more capable aerial-ground collaborative embodied intelligence systems.

Problem

• Aerial-Ground Collaborative Reasoning
• Vision-Language Models
• Benchmark Evaluation
• Cross-View Understanding

Innovation

• Aerial-Ground Collaborative Reasoning
• Vision-Language Models Benchmark
• Cross-View Correspondence
• Embodied Intelligence
• Multimodal Dataset