aiqbench analysis
MultihopSpatial introduces a critical benchmark addressing the gap in multi-hop compositional spatial reasoning for Vision-Language Models, which is essential for advancing VLA agents in physical environments. By providing a large-scale training corpus and a new metric that combines answer selection with precise visual grounding, this work directly tackles the limitations of current models in handling complex, chained spatial queries. The findings confirm that while reinforcement learning post-training improves embodied manipulation performance, robust compositional reasoning remains a significant hurdle for next-generation AI systems.
MultihopSpatial: Multi-hop Compositional Spatial Reasoning Benchmark for Vision-Language Model
arXiv:2603.18892v2 Announce Type: replace-cross Abstract: Spatial reasoning is foundational for Vision-Language Models (VLMs), particularly when deployed as Vision-Language-Action (VLA) agents in physical environments. However, existing benchmarks predominantly focus on elementary, single-hop relations, neglecting the multi-hop compositional reasoning and precise visual grounding essential for real-world scenarios. To address this, we introduce MultihopSpatial, offering three key contributions: (1) A comprehensive benchmark designed for multi-hop and compositional spatial reasoning, featuring 1- to 3-hop complex queries across diverse spatial perspectives. (2) Acc@50IoU, a complementary metric that simultaneously evaluates reasoning and visual grounding by requiring both answer selection and precise bounding box prediction - capabilities vital for robust VLA deployment. (3) MultihopSpatial-Train, a dedicated large-scale training corpus to foster spatial intelligence. Extensive evaluation of 37 state-of-the-art VLMs yields eight key insights, revealing that compositional spatial reasoning remains a formidable challenge. Finally, we demonstrate that reinforcement learning post-training on our corpus enhances both intrinsic VLM spatial reasoning and downstream embodied manipulation performance.
Abstract:Spatial reasoning is foundational for Vision-Language Models (VLMs), particularly when deployed as Vision-Language-Action (VLA) agents in physical environments. However, existing benchmarks predominantly focus on elementary, single-hop relations, neglecting the multi-hop compositional reasoning and precise visual grounding essential for real-world scenarios. To address this, we introduce MultihopSpatial, offering three key contributions: (1) A comprehensive benchmark designed for multi-hop and compositional spatial reasoning, featuring 1- to 3-hop complex queries across diverse spatial perspectives. (2) Acc@50IoU, a complementary metric that simultaneously evaluates reasoning and visual grounding by requiring both answer selection and precise bounding box prediction - capabilities vital for robust VLA deployment. (3) MultihopSpatial-Train, a dedicated large-scale training corpus to foster spatial intelligence. Extensive evaluation of 37 state-of-the-art VLMs yields eight key insights, revealing that compositional spatial reasoning remains a formidable challenge. Finally, we demonstrate that reinforcement learning post-training on our corpus enhances both intrinsic VLM spatial reasoning and downstream embodied manipulation performance.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2603.18892 [cs.CV]
(or arXiv:2603.18892v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2603.18892 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 249 words
Related coverage
SceneBench: A Hierarchical Benchmark for Vision-Language Understanding of 3D Scenes
arXiv · Sep 16, 2026
Perceive, Interact, Reason: Building Tool-Augmented Visual Agents for Spatial Reasoning
arXiv · Jun 12, 2026
MVVBench: Benchmarking 4D Reasoning in Vision-Language Models
arXiv · Sep 28, 2026
Small yet Assistive: Spatially-Aware Post-Training for Low Vision
arXiv · Sep 25, 2026