AirGroundBench: Probing Spatial Intelligence in Multimodal Large Models under Heterogeneous Multi-View Embodied Collaboration

2026-06-26

AI Summary

Existing benchmarks lack a systematic evaluation of multimodal large models’ ability to achieve geometrically consistent spatial understanding across heterogeneous aerial-ground collaborative scenarios. To address this gap, this work proposes AirGroundBench—the first multiview spatial intelligence diagnostic benchmark tailored for aerial-ground collaboration. Built upon high-fidelity simulation, it generates synchronized aerial-ground observation pairs and establishes a four-tier progressive evaluation framework encompassing spatial perception, cross-view alignment, spatial reasoning, and embodied decision-making, supported by structured 2D/3D annotations and cross-view object identity alignment. Evaluation across 13 state-of-the-art multimodal large models reveals that while models exhibit moderate performance in basic spatial perception, they face significant bottlenecks in cross-view alignment and geometric transformation reasoning, with performance remaining substantially below human levels even when provided with dual-view inputs.

Abstract

In recent years, multimodal large language models (MLLMs) have shown strong potential for embodied intelligence, yet their ability to maintain geometrically consistent spatial understanding across heterogeneous views remains under-evaluated. Existing benchmarks largely focus on single-agent, single-view perception, leaving a gap in the systematic assessment of collaborative air-ground settings, where multi-scale observations are complementary but introduce scale mismatch, asymmetric occlusion, and reference-frame inconsistencies. We present AirGroundBench, a diagnostic benchmark for evaluating multi-view spatial intelligence in heterogeneous UAV-UGV collaboration. AirGroundBench is built from 11 high-fidelity simulated environments with 1,021 synchronized air-ground observation pairs, yielding approximately 62,000 dual-view, four-option single-choice visual question answering instances and 115 closed-loop vision-language navigation episodes. It covers 10 task types organized into four progressively demanding capability dimensions: spatial perception, cross-view alignment, spatial transformation and reasoning, and embodied decision-making. To support geometry-grounded evaluation and analysis, we provide structured spatial annotations, including cross-view object identities and metric 2D and 3D bounding boxes. Evaluations of 13 representative MLLMs under UAV-only, UGV-only, and dual-view input settings reveal consistent bottlenecks: models perform relatively well on spatial perception but struggle with cross-view alignment and transformation-intensive reasoning, and these deficits propagate to sequential decision-making in vision-language navigation. Although dual-view inputs provide measurable gains over single-view variants, a persistent gap from human performance remains, highlighting geometric consistency as a key limitation of current embodied MLLMs.

Problem

• spatial intelligence
• multimodal large language models
• heterogeneous multi-view
• embodied collaboration
• geometric consistency

Innovation

• multimodal large language models
• spatial intelligence
• heterogeneous collaboration
• cross-view alignment
• embodied AI