HybridSim: A Physics-Learning Hybrid Digital Twin for mmWave Human Sensing

ECCV 2026

Synthesizing realistic mmWave radar signals from dynamic human meshes and fixed-room indoor scenes.

Weitao Xiong^1,2 Tianyu Liu² Peng Li² Kok Chung Chua¹
Toa Chean Khim¹ Pu Wang³ Hongfei Xue^3†

¹Xiamen University ²The Hong Kong University of Science and Technology ³University of North Carolina at Charlotte
^†Corresponding Author

Abstract

High-fidelity simulation of mmWave radar signals for dynamic human motion is valuable for developing radar-based human sensing models, but collecting accurately labeled measurements for each deployment site remains expensive.

HybridSim synthesizes mmWave radar signals from dynamic human meshes under a fixed indoor room configuration by decoupling propagation into direct inverse-rendering paths and indirect 3DGS-based paths. This hybrid design improves agreement with physically based references while keeping site-specific data augmentation practical.