Elamur Research⌕
☰
home / papers / 2026 / CityCube: Benchmarking Cross-view Spatial Reasoning on Vision-Language Models in Urban Environments

CityCube: Benchmarking Cross-view Spatial Reasoning on Vision-Language Models in Urban Environments

Paper · 2026 · Haotian Xu, Yue Hu, Zhengqiu Zhu et al.

Summary

Vision-language models increasingly act as the reasoning core of embodied agents such as drones, robots and self-driving cars. Yet spatial benchmarks mostly cover indoor rooms or street scenes and ignore open urban space, with its dense repeated objects, intricate geometry and shifts between ground-level and overhead views. The authors build CityCube to measure cross-view spatial intelligence (CvSI) in cities. It combines real imagery from 74 cities with renders from the EmbodiedCity and MatrixCity simulators, about 18.1K images in total. The images are grouped by camera-movement patterns and by viewing platform, from pedestrian and vehicle cameras to drones and satellites. On top of them sit 5,022 human-verified multiple-choice questions. Testing 33 VLMs showed that none passes roughly 54% accuracy, far below human scores. Small Qwen3-VL models fine-tuned on the training split with LoRA beat that ceiling. Text-only ablations confirm that the questions need the images to be answered. The benchmark exposes concrete gaps in mental rotation, cross-view matching and motion understanding.

Key points

  • Coverage: the 5,022 QA pairs span 59 subtasks organized into five cognitive categories, namely reasoning about how objects relate in space (SR), adopting another viewpoint (PT), rebuilding a scene in the mind (MR), applying world knowledge (WK), and integrated multi-step reasoning (CR). Questions use three spatial-relation framings (egocentric, exocentric, allocentric), and images are arranged by four viewpoint-dynamics primitives.
  • Model performance: across 33 evaluated VLMs, the best stays at or below 54.1% accuracy, which is 34.2 points short of humans. The CityBot-2B/4B/8B models (Qwen3-VL with LoRA fine-tuning) score above 60%.
  • Human vs. model difficulty: per-subtask human accuracy barely correlates with average VLM accuracy (Pearson r = 0.098, p = 0.46). Model accuracy across subtasks spreads widely, from 19.2% to 91.8%.
  • Visual necessity: with images masked, Qwen3-VL-8B falls from 43.1% to 27.5% and GLM-4.1V-9B-Base from 42.6% to 22.5%, close to the 22.8% random baseline. Only some commonsense WK tasks stay above chance.
  • Stated limitations: Sim-to-Real effects are not evaluated, model internals are not probed, and post-training methods such as spatial Chain-of-Thought are left to future work.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksComputer Vision and Pattern Recognition · 2,869 worksEngineering · 133 worksAutomotive Engineering · 6 works

🧭 Topics 2Research subjects and languages the work is about.Reasoning · 1,426 worksVision & language · 657 works

Cite

Cite as Haotian Xu, Yue Hu, Zhengqiu Zhu, Chen Gao, Ziyou Wang, Junreng Rao, Wenhao Lu, Weishi Li, Quanjun Yin and Yong Li. 2026. CityCube: Benchmarking Cross-view Spatial Reasoning on Vision-Language Models in Urban Environments. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.416

@inproceedings{xu2026citycube,
  title = {{CityCube: Benchmarking Cross-view Spatial Reasoning on Vision-Language Models in Urban Environments}},
  author = {Haotian Xu, Yue Hu, Zhengqiu Zhu, Chen Gao, Ziyou Wang, Junreng Rao, Wenhao Lu, Weishi Li, Quanjun Yin and Yong Li},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.416},
  url = {https://doi.org/10.18653/v1/2026.acl-long.416}
}

References

53 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Panos Achlioptas, Ahmed Abdelreheem, Fei Xia, Mohamed Elhoseiny, and Leonidas Guibas. 2020. Referit3d: Neural listeners for fine-grained 3d object identification in real-world scenes. In European conference on computer vision, pages 422–440. Springer. find
  2. Daichi Azuma, Taiki Miyanishi, Shuhei Kurita, and Motoaki Kawanabe. 2022. Scanqa: 3d question answering for spatial scene understanding. In proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 19129–19139. find
  3. Shuai Bai, Yuxuan Cai, and Ke Zhu. 2025. Qwen3-vl technical report. Preprint, arXiv:2511.21631.
  4. Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh Vora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, and Oscar Beijbom. 2020. 9198 9 nuscenes: A multimodal dataset for autonomous driving. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition,… find
  5. Wenxiao Cai, Iaroslav Ponomarenko, Jianhao Yuan, Xiaoqi Li, Wankou Yang, Hao Dong, and Bo Zhao. 2025a. Spatialbot: Precise spatial understanding with vision language models. In 2025 IEEE International Conference on Robotics and Automation (ICRA), pages 9490–9498. IEEE. find
  6. Zhongang Cai, Ruisi Wang, and Lei Yang. 2025b. Scaling spatial intelligence with multimodal foundation models. arXiv preprint arXiv:2511.13719. — [arXiv:2511.13719](https://arxiv.org/abs/2511.13719) find
  7. Zhongang Cai, Yubo Wang, Qingping Sun, Ruisi Wang, Chenyang Gu, Wanqi Yin, Zhiqian Lin, Zhitao Yang, Chen Wei, Xuanke Shi, and 1 others. 2025c. Has gpt-5 achieved spatial intelligence? an empirical study. arXiv preprint arXiv:2508.13142, 3. — [arXiv:2508.13142](https://arxiv.org/abs/2508.13142) find
  8. Angel Chang, Angela Dai, Thomas Funkhouser, Maciej Halber, Matthias Niessner, Manolis Savva, Shuran Song, Andy Zeng, and Yinda Zhang. 2017. Matterport3d: Learning from rgb-d data in indoor environments. arXiv preprint arXiv:1709.06158. — [arXiv:1709.06158](https://arxiv.org/abs/1709.06158) find
  9. Boyuan Chen, Zhuo Xu, Sean Kirmani, Brain Ichter, Dorsa Sadigh, Leonidas Guibas, and Fei Xia. 2024. Spatialvlm: Endowing vision-language models with spatial reasoning capabilities. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14455–14465. find
  10. An-Chieh Cheng, Hongxu Yin, Yang Fu, Qiushan Guo, Ruihan Yang, Jan Kautz, Xiaolong Wang, and Sifei Liu. 2024. Spatialrgpt: Grounded spatial reasoning in vision-language models. Advances in Neural Information Processing Systems, 37:135062–135093. find
  11. Meng Chu, Zhedong Zheng, Wei Ji, Tingyu Wang, and Tat-Seng Chua. 2024. Towards natural language-guided drones: Geotext-1652 benchmark with spatial relation matching. In European Conference on Computer Vision, pages 213–231. Springer. find
  12. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, and 1 others. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arX…
  13. Angela Dai, Angel X Chang, Manolis Savva, Maciej Halber, Thomas Funkhouser, and Matthias Nießner. 2017. Scannet: Richly-annotated 3d reconstructions of indoor scenes. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5828–5839. find
  14. Jingtao Ding, Yunke Zhang, Yu Shang, Yuheng Zhang, Zefang Zong, Jie Feng, Yuan Yuan, Hongyuan Su, Nian Li, Nicholas Sukiennik, and 1 others. 2025. find
  15. Understanding world or predicting future? a comprehensive survey of world models. ACM Computing Surveys, 58(3):1–38. find
  16. Mengfei Du, Binhao Wu, Zejun Li, Xuan-Jing Huang, and Zhongyu Wei. 2024. Embspatial-bench: Benchmarking spatial understanding for embodied tasks with large vision-language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pag… find
  17. SM Ali Eslami, Danilo Jimenez Rezende, Frederic Besse, Fabio Viola, Ari S Morcos, Marta Garnelo, Avraham Ruderman, Andrei A Rusu, Ivo Danihelka, Karol Gregor, and 1 others. 2018. Neural scene representation and rendering. Science, 360(6394):1204– 1210. find
  18. Jie Feng, Tianhui Liu, Yuwei Du, Siqi Guo, Yuming Lin, and Yong Li. 2025a. Citygpt: Empowering urban spatial cognition of large language models. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2, pages 591–602. find
  19. Jie Feng, Shengyuan Wang, Tianhui Liu, Yanxin Xi, and Yong Li. 2025b. Urbanllava: A multi-modal large language model for urban intelligence with spatial reasoning and understanding. arXiv preprint arXiv:2506.23219. — [arXiv:2506.23219](https://arxiv.org/abs/2506.23219) find
  20. Jie Feng, Jun Zhang, Junbo Yan, Xin Zhang, Tianjian Ouyang, Tianhui Liu, Yuwei Du, Siqi Guo, and Yong Li. 2024. Citybench: Evaluating the capabilities of large language model as world model. arXiv e-prints, pages arXiv–2406. find
  21. Rao Fu, Jingyu Liu, Xilun Chen, Yixin Nie, and Wenhan Xiong. 2024. Scene-llm: Extending language model for 3d visual understanding and reasoning. arXiv preprint arXiv:2403.11401. — [arXiv:2403.11401](https://arxiv.org/abs/2403.11401) find
  22. Chen Gao, Baining Zhao, Weichen Zhang, Jinzhu Mao, Jun Zhang, Zhiheng Zheng, Fanhang Man, Jianjie Fang, Zile Zhou, Jinqiang Cui, and 1 others. 2024. Embodiedcity: A benchmark platform for embodied agent in real-world city environment. arXiv preprint arXiv:2410.09604. — [arXiv:2410.09604](https://arxiv.org/abs/2410.09604) find
  23. Mohsen Gholami, Ahmad Rezaei, Zhou Weimin, Sitong Mao, Shunbo Zhou, Yong Zhang, and Mohammad Akbari. 2025. Spatial reasoning with vision-language models in ego-centric multi-view scenes. arXiv preprint arXiv:2509.06266. — [arXiv:2509.06266](https://arxiv.org/abs/2509.06266) find
  24. Dong Guo, Faming Wu, and Zuquan Song. 2025. Seed1.5-vl technical report. Preprint, arXiv:2505.07062.
  25. Zoey Guo, Yiwen Tang, Ray Zhang, Dong Wang, Zhigang Wang, Bin Zhao, and Xuelong Li. 2023. Viewrefer: Grasp the multi-view knowledge for 3d visual grounding. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 15372–15383. 9199 10 find
  26. Xixuan Hao, Wei Chen, Yibo Yan, Siru Zhong, Kun Wang, Qingsong Wen, and Yuxuan Liang. 2024. Urbanvlp: A multi-granularity vision-language pre-trained foundation model for urban indicator prediction. CoRR. find
  27. Jun He, Yi Lin, Zilong Huang, Jiacong Yin, Junyan Ye, Yuchuan Zhou, Weijia Li, and Xiang Zhang. 2025. Urbanfeel: A comprehensive benchmark for temporal and perceptual understanding of city scenes through human perspective. arXiv preprint arXiv:2509.22228. — [arXiv:2509.22228](https://arxiv.org/abs/2509.22228) find
  28. Yining Hong, Haoyu Zhen, Peihao Chen, Shuhong Zheng, Yilun Du, Zhenfang Chen, and Chuang Gan. 2023. 3d-llm: Injecting the 3d world into large language models. Advances in Neural Information Processing Systems, 36:20482–20494. find
  29. Jiangyong Huang, Silong Yong, Xiaojian Ma, Xiongkun Linghu, Puhao Li, Yan Wang, Qing Li, Song-Chun Zhu, Baoxiong Jia, and Siyuan Huang. 2024. An embodied generalist agent in 3d world. In Proceedings of the 41st International Conference on Machine Learning, pages 20413–20451. find
  30. Mengdi Jia, Zekun Qi, Shaochen Zhang, Wenyao Zhang, Xinqiang Yu, Jiawei He, He Wang, and Li Yi. 2025. Omnispatial: Towards comprehensive spatial reasoning benchmark for vision language models. arXiv preprint arXiv:2506.03135. — [arXiv:2506.03135](https://arxiv.org/abs/2506.03135) find
  31. Silwal, Paul Mcvay, Oleksandr Maksymets, Sergio Arnaud, and 1 others. 2024. Openeqa: Embodied question answering in the era of foundation models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16488–16498. find
  32. OpenAI. 2025. Gpt-5 system card. Accessed: 2026-01-03. find
  33. Jean Piaget. 2013. Child’s conception of space: Selected works vol 4. Routledge. find
  34. Zekun Qi, Runpei Dong, Shaochen Zhang, Haoran Geng, Chunrui Han, Zheng Ge, Li Yi, and Kaisheng Ma. 2024. Shapellm: Universal 3d object understanding for embodied interaction. In European Conference on Computer Vision, pages 214–238. Springer. find
  35. Chan Hee Song, Valts Blukis, Jonathan Tremblay, Stephen Tyree, Yu Su, and Stan Birchfield. 2025. Robospatial: Teaching spatial understanding to 2d and 3d vision-language models for robotics. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 15768–15780. find
  36. Peiwen Sun, Shiqiang Lang, and Xiangyu Yue. 2025. Spacevista: All-scale visual spatial reasoning from mm to km. arXiv preprint arXiv:2510.09606. — [arXiv:2510.09606](https://arxiv.org/abs/2510.09606) find
  37. Dingming Li, Hongxing Li, Zixuan Wang, Yuchen Yan, Hang Zhang, Siqi Chen, Guiyang Hou, Shengpei Jiang, Wenqi Zhang, Yongliang Shen, and 1 others. 2025. Viewspatial-bench: Evaluating multi-perspective spatial localization in vision-language models. arXiv preprint arXiv:2505.21500. — [arXiv:2505.21500](https://arxiv.org/abs/2505.21500) find
  38. Barbara Tversky. 2019. Mind in motion: How action shapes thought. Basic Books. find
  39. Yixuan Li, Lihan Jiang, Linning Xu, Yuanbo Xiangli, Zhenzhi Wang, Dahua Lin, and Bo Dai. 2023. Matrixcity: A large-scale city dataset for city-scale neural rendering and beyond. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3205–3215. find
  40. Jihan Yang, Shusheng Yang, Anjali W Gupta, Rilyn Han, Li Fei-Fei, and Saining Xie. 2025a. Thinking in space: How multimodal large language models see, remember, and recall spaces. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 10632–10643. find
  41. Jingli Lin, Runsen Xu, and Jiangmiao Pang. 2025. Mmsi-video-bench: A holistic benchmark for video-based spatial intelligence. Preprint, arXiv:2512.10863. — [arXiv:2512.10863](https://arxiv.org/abs/2512.10863) find
  42. Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. 2024. Depth anything: Unleashing the power of large-scale unlabeled data. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10371–10381. find
  43. Tianhui Liu, Jie Feng, Hetian Pang, Xin Zhang, Tian-jian Ouyang, Zhiyuan Zhang, and Yong Li. 2025a. Citylens: Benchmarking large language-vision models for urban socioeconomic sensing. arXiv preprint arXiv:2506.00530. — [arXiv:2506.00530](https://arxiv.org/abs/2506.00530) find
  44. Yuhong Liu, Beichen Zhang, Yuhang Zang, Yuhang Cao, Long Xing, Xiaoyi Dong, Haodong Duan, Dahua Lin, and Jiaqi Wang. 2025b. Spatial-ssrl: Enhancing spatial understanding via self-supervised reinforcement learning. arXiv preprint arXiv:2510.27606. — [arXiv:2510.27606](https://arxiv.org/abs/2510.27606) find
  45. Arjun Majumdar, Anurag Ajay, Xiaohan Zhang, Pranav Putta, Sriram Yenamandra, Mikael Henaff, Sneha Tianwen Wei, Liang Zhao, and Yahui Zhou. 2023. Skywork: A more open bilingual foundation model. Preprint, arXiv:2310.19341.
  46. Sihan Yang, Runsen Xu, Yiman Xie, Sizhe Yang, Mo Li, Jingli Lin, Chenming Zhu, Xiaochen Chen, Haodong Duan, Xiangyu Yue, and 1 others. 2025b. Mmsi-bench: A benchmark for multi-image spatial intelligence. arXiv preprint arXiv:2505.23764. — [arXiv:2505.23764](https://arxiv.org/abs/2505.23764) find
  47. Chun-Hsiao Yeh, Chenyu Wang, Shengbang Tong, Ta-Ying Cheng, Ruoyu Wang, Tianzhe Chu, Yuexiang Zhai, Yubei Chen, Shenghua Gao, and Yi Ma. 2025. Seeing from another perspective: Evaluating multi-view understanding in mllms. arXiv preprint arXiv:2504.15280. 9200 11 — [arXiv:2504.15280](https://arxiv.org/abs/2504.15280) find
  48. Baiqiao Yin, Qineng Wang, Pingyue Zhang, Jianshu Zhang, Kangrui Wang, Zihan Wang, Jieyu Zhang, Keshigeyan Chandrasegaran, Han Liu, Ranjay Krishna, and 1 others. 2025. Spatial mental modeling from limited views. In Structural Priors for Vision Workshop at ICCV’25. find
  49. Sha Zhang, Di Huang, Jiajun Deng, Shixiang Tang, Wanli Ouyang, Tong He, and Yanyong Zhang. 2024. Agent3d-zero: An agent for zero-shot 3d understanding. In European Conference on Computer Vision, pages 186–202. Springer. find
  50. Weichen Zhang, Zile Zhou, Zhiheng Zheng, Chen Gao, Jinqiang Cui, Yong Li, Xinlei Chen, and Xiao-Ping Zhang. 2025. Open3dvqa: A benchmark for comprehensive spatial reasoning with multimodal large language model in open space. arXiv preprint arXiv:2503.11094. — [arXiv:2503.11094](https://arxiv.org/abs/2503.11094) find
  51. Baining Zhao, Jianjie Fang, Zichao Dai, Ziyou Wang, Jirong Zha, Weichen Zhang, Chen Gao, Yue Wang, Jinqiang Cui, Xinlei Chen, and 1 others. 2025. Urbanvideo-bench: Benchmarking vision-language models on embodied intelligence with video data in urban spaces. arXiv preprint arXiv:2503.06157. find
  52. Baichuan Zhou, Haote Yang, Dairong Chen, Junyan Ye, Tianyi Bai, Jinhua Yu, Songyang Zhang, Dahua Lin, Conghui He, and Weijia Li. 2025. Urbench: A comprehensive benchmark for evaluating large multimodal models in multi-view urban scenarios. In Proceedings of the AAAI Conference on Artificial Intellig… find
  53. Chenming Zhu, Tai Wang, Wenwei Zhang, Jiangmiao Pang, and Xihui Liu. 2024. Llava-3d: A simple yet effective pathway to empowering lmms with 3d-awareness. arXiv preprint arXiv:2409.18125. 9201 12 — [arXiv:2409.18125](https://arxiv.org/abs/2409.18125) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xu-2026-citycube-benchmarking-cross-view-spatial/