Wi-Fi Sensing

Wi-Fi sensing studies how changes in received signal strength (RSSI), channel state information (CSI), and Doppler patterns reveal people’s presence, motion, location, posture, and concurrent activities, often without cameras or wearables. Common approaches combine signal processing with CNNs, temporal convolutional networks, and Transformers; localization estimates positions or floors, while pose systems map signals to 2D or 3D skeletons. Self-supervised learning, domain adaptation, physics-guided models, and signal-language alignment aim to reduce labeling needs and improve performance across users, devices, and environments, with evaluations including held-out-user and held-out-room tests on datasets such as MM-Fi, Widar 3.0, and WiMANS. Strong results in familiar settings remain difficult to transfer reliably when room layouts, hardware, or Wi-Fi traffic patterns change, especially with few target-domain labels.