Embodied mmWave Spotlight: How Stripped-Down mmWave Radars Finally Made Egocentric 3D Human Mesh Reconstruction Practical

Imagine you're at home playing motion-sensing games or perfecting your fitness movements. You want real-time 3D feedback on your posture-whether to correct your form, track progress, or create a lifelike avatar for the metaverse. Yet today's solutions leave you with an unsatisfying choice: point a camera (e.g., RGB, depth, or thermal) at yourself and sacrifice privacy while struggling with varying light conditions and occlusions, or rely on bulky, expensive motion-capture (MoCap) systems that no ordinary person would install at home. Researchers have explored camera-free alternatives based on RF sensing technologies [1-5]—for example, an mmWave radar positioned in front of the user. However, this frontal-view approach suffers from limited sensing range, self-occlusion, and severe multipath interference from walls and furniture-performance collapses the moment the user turns or moves a few steps away.