Xinyu Zhang visits us from UC San Diego. He will give a talk on "RF Digital Twins: A Physics-Grounded Paradigm for RF-Based Spatial Intelligence". This is part of our Excellence Cluster work on Spatial AI. The talk is also accessible on zoom: https://lu-se.zoom.us/j/66506922051

Abstract:

As a foundational input for spatial AI, radio frequency (RF) signals offer unique advantages over traditional optical and acoustic modalities. They can enable novel sensing applications such as non-line-of-sight localization, mapping, object characterization, and material identification. However, the efficacy of RF sensing is fundamentally limited by the underlying propagation models. While traditional physics-based models offer interpretability and efficiency, they often fail to capture site-specific geometric and material complexities. Conversely, black-box machine learning models can capture complex interactions but often lack the interpretability and generalizability across diverse environments.

In this talk, I will present RF Digital Twin (RFDT), a new paradigm to bridge this gap through differentiable ray tracing. Unlike static simulators, RFDT functions as a "differentiable world model" that incorporates physics-based simulation directly into a closed optimization loop.  This architecture allows the RF sensing task to be formulated as a joint optimization problem: by minimizing the discrepancy between simulated and measured signals, the system can "sense" and reconstruct environment geometry, material properties, and sensor states simultaneously. I will demonstrate the effectiveness of RFDT in high-fidelity field modeling and complex sensing tasks, supported by extensive testbed implementations and experimental validation across diverse real-world scenarios.  In addition, I will discuss the potential of using RFDT as a fundamental, differentiable module within larger machine-learning pipelines for multi-modal sensing and spatial AI systems.

Biography: 
Xinyu Zhang is the Ericsson Endowed Chair Professor in the Department of ECE at UC San Diego, and also serves as the Director for the Center for Wireless Communications. He received his Ph.D. in Computer Science and Engineering from the University of Michigan in 2012. His research interest lies in wireless networking and ubiquitous sensing systems. He is the recipient of two ACM MobiCom Best Paper Awards (2011 and 2020), SenSys Best paper Award (2023), Communications of the ACM Research Highlight (2018, 2023), ACM SIGMOBILE Research Highlight (2018), NSF CAREER Award (2014), Google Research Award (2017, 2018, 2020), and ACM SIGMOBILE RockStar Award in 2023.  He served as the TPC chair for ACM MobiCom 2019, IEEE SECON 2017, co-chair of the NSF millimeter-wave research coordination network, and Associate Editor for IEEE Transactions on Mobile Computing from 2017 to 2020.