Spatial Channel Modeling
How to move from stochastic channel descriptions to geometry-conditioned models.
Spatial Intelligence for Wireless Communications and Sensing
Methods to Applications
From wireless radiation fields to world models: a half-day tutorial by HKUST iComAI Lab on reusable spatial knowledge for 6G communications, sensing, and embodied agents.
Tutorial Motivation
Why Spatial Intelligence?
Move wireless systems from link-level optimization to spatial intelligence.
Next-generation networks are becoming AI-native, but a deeper bottleneck remains: we still understand channels as short-window, per-link samples, while the physical world is a continuous, shared, and reusable 3D space.
Massive MIMO, ultra-reliable low latency, mixed bands, and dense deployment all ask the same question: when pilots are scarce, CSI is imperfect, and environments change rapidly, how do communications and sensing upgrade from estimating one link to understanding one space?
Spatial intelligence answers this by treating wireless channels as shared projections of the physical environment, building reusable spatial priors through wireless radiation fields and world models—so networks can not only transmit, but also reason, predict, and plan in space.
Learning Outcomes
After This Tutorial, Attendees Will Understand
3D-GS For Wireless
How 3D Gaussian Splatting can support wireless radiance field reconstruction.
Spatial Intelligence Generalization
How datasets, few-shot adaptation, and cross-scene transfer shape spatial intelligence.
6G Applications
How spatial intelligence enables positioning, planning, UAV networking, and embodied AI.
Half-Day Program
Tutorial Outline
Aligned with the slide deck Spatial Intelligence Tutorial v1 (PDF).
Vision for Wireless Spatial Intelligence
- Why is reusable spatial knowledge needed?
- What is the wireless spatial field?
- How should wireless space be represented?
Radio Map & Spatial Intelligence Generalization
- Few/Zero-shot generalization in a single scene — URF-GS
- Generalization across scenes — Point2Radio
- Indoor positioning
Spatial-Intelligent Enabled Applications
- Intelligent Agents
- Beamforming
- AP Deployment and Low-altitude Spatial Intelligence
Conclusions & Outlook
- Conclusions
- Outlook toward wireless world models
Tutorial Team
Speakers
Jun Zhang 张军
Professor, HKUST. IEEE Fellow.
Wireless communications, networking, edge AI, and cooperative AI.
Jiajia Guo 郭佳佳
Research Assistant Professor, HKUST.
AI-native air interface, massive MIMO, ISAC, and large AI models.
Yumeng Zhang 张宇萌
Post-Doctoral Fellow, HKUST.
AI for wireless, sensing, ISAC, massive MIMO, and channel estimation.
Chenghong Bian 卞程泓
Post-Doctoral Fellow, HKUST.
Wireless communications, machine learning, and sensing.
Special Thanks
Featured contributor to the tutorial materials and core demos.
Chaozheng Wen 温朝政
PhD Student, HKUST iComAI Lab.
3D Gaussian Splatting for wireless radiation field reconstruction and radio map learning (WRF-GS, URF-GS, Point2Radio).
Acknowledgments
For significant contributions to the tutorial materials, we also thank:
- Mr. Yijie Bian 卞艺杰 For beamforming applications ybianaf@connect.ust.hk
- Dr. Ziqi Ke 柯子琦 For UAV applications zqke24@m.fudan.edu.cn
Conference Versions
Accepted Tutorials
IEEE ICCC 2026
From wireless radiation fields to world models — a half-day path through vision, 3D-GS methods, generalization, and applications.
IEEE GLOBECOM 2026
Conference-specific time, venue, registration link, and released materials will be added here.
Resources
Materials
Slides
Tutorial PDF (v1)
Reference Papers
Key papers below
Demo Notebooks
Coming soon
Code / Datasets
Coming soon
Key Papers (Parts II–IV)
- WRF-GS — Wireless Radiation Field Reconstruction with 3D Gaussian Splatting (INFOCOM 2025; arXiv)
- WRF-GS+ — Neural Representation for Wireless Radiation Field Reconstruction: A 3D Gaussian Splatting Approach (IEEE TWC 2025; arXiv)
- GeoGS-CE — Learning Delay–Beam Channel Priors with 3D Gaussians for High-Mobility Scenarios (arXiv)
- URF-GS — Bridging Visual and Wireless Sensing via a Unified Radiation Field for 3D Radio Map Construction (arXiv)
- Point2Radio — A Foundation Model for Cross-Scene Radio Fields from Material-Aware Point Clouds (arXiv)
- GSBF — Gaussian Splatting for Environment-Aware Beamforming (arXiv)