The prevailing single-source paradigm in wireless sensing produces specialized models that are unscalable and generalize poorly to new tasks. Multi-source pre-training offers a path toward a generalist backbone but poses challenges including task heterogeneity, data redundancy, structural incompatibility, and the lack of a general-purpose pre-training objective. To address these issues, we propose WiSwiss, a comprehensive self-supervised multi-source pre-training framework that learns a general-purpose backbone for each modality. WiSwiss integrates semantic deduplication for dataset curation and a transformation-invariant pre-training objective. Experiments show that WiSwiss outperforms models trained from scratch, improving WiFi and mmWave performance by 4.5% and 10.3%, respectively, while reducing fine-tuning data requirements by 22.2% and 28.6%. We also present a qualitative study of scaling laws, showing that gains are task-dependent and that larger models require sufficiently large and diverse pre-training corpora to achieve substantial improvements.

Tracking-based social behavior analysis of lab animals, especially mice, is crucial for research in biology, medicine, and psychology. However, existing visual tracking systems struggle to maintain long-term, accurate tracking due to frequent identity association errors, which lead to extensive manual correction and limit research scalability. This paper introduces mTrack, an RFID–vision hybrid mouse tracking system that leverages the precise identification capability of UHF RFID to assist the visual tracker, enabling long-term, self-correcting, and high-accuracy mouse tracking. Our experiments demonstrate that mTrack can track up to ten mice simultaneously with over 99.23% identification accuracy and a 0.4 cm 99th-percentile localization error, reducing error rate by more than 40x compared with visual tracking systems. Our field studies indicate that mTrack can sustain this performance for over two hours and be seamlessly integrated into existing animal behavior research workflows. The code and dataset are open-sourced at https://github.com/SOAR-PKU/mTrack.

Following the broad prospect of Radio Frequency Identification (RFID) technology is the security concern of unauthorized tag identification, which poses threats to the privacy of both objects and users. In this paper, we propose RF-Rock, the first RFID unauthorized identification attack that operates without tag activation, thereby evading almost all existing defenses. This attack exposes the vulnerabilities of current RFID networks in identification legitimacy and privacy. It is based on the intermodulation effect originating from intrinsic nonlinearity within tag circuits. To this end, we explore the distinctness and consistency of the intermodulation-based physical layer fingerprint of RFID tags with theoretical analysis and empirical validation, and optimize the attack accuracy and efficiency with delicate excitation plan. Real-world experiments show that RF-Rock achieves an attack success rate of 93.2% on average under various conditions. The entropy of our proposed fingerprint is 15.5 bits and implies sufficient capacity in practical attacks.

Generative models have gained significant attention for their ability to produce realistic synthetic data that supplements the quantity of real-world datasets. While recent studies show performance improvements in wireless sensing tasks by incorporating all synthetic data into training sets, the quality of synthetic data remains unpredictable and the resulting performance gains are not guaranteed. To address this gap, we propose tractable and generalizable metrics to quantify quality attributes of synthetic data—affinity and diversity. Our assessment reveals prevalent affinity limitation in current wireless synthetic data, leading to mislabeled data and degraded task performance. We attribute the quality limitation to generative models' lack of awareness of untrained conditions and domain-specific processing. To mitigate these issues, we introduce SynCheck, a quality-guided synthetic data utilization scheme that refines synthetic data quality during task model training. Our evaluation demonstrates that SynCheck consistently outperforms quality-oblivious utilization of synthetic data, and achieves 4.3% performance improvement even when the previous utilization degrades performance by 13.4%.

We introduce ErLang Sight, a novel multimodal system designed for liquid identification, integrating vision-based object detection, millimeter-wave (mmWave) Synthetic Aperture Radar (SAR) imaging, and large language model (LLM)-based contextual reasoning. Initially, the system leverages a visual detection pipeline to identify potential liquid containers within the environment, subsequently directing a mmWave sensor to perform targeted SAR imaging of these identified regions, and the permittivity values of the liquids are estimated using reflection coefficient analysis techniques. These physical measurements, combined with visual context and environmental indicators (such as whether the scenario is a kitchen, laboratory, or bar), are then input into a pretrained LLM. The LLM employs advanced semantic and situational reasoning to accurately determine the most likely type of liquid by integrating physics-based data with contextual knowledge. Experimental evaluations demonstrate that ErLang Sight significantly enhances the accuracy of distinguishing visually ambiguous liquids and exhibits robust generalization to previously unseen environments.

As embodied AI gradually transitions into practical applications, enhancing the fidelity of how embodied agents perceive the physical world has become a critical challenge. Current perception methods typically rely on computer vision-based fiducial marker systems, which suffer from limitations such as insufficient reading distance, poor localization accuracy, and high susceptibility to environmental lighting conditions. Currently, SPAD sensor-based LiDAR technology is emerging in commercial mobile devices due to its compact size, high precision, and low power consumption. This paper presents the design of the RetroLiDAR system, which chimes with the concept of backscatter in wireless technology, to create a liquid-crystal fiducial marker system that can be directly read by LiDAR. On the marker side, we use retroreflective materials to reflect the LiDAR's emitted light back and employ a liquid crystal modulator to adjust the intensity of the light signal. On the LiDAR end, we design a signal processing pipeline to demodulate the marker's modulation message using the temporal received signal strength. Experimental results from our prototype demonstrate that compared to visual fiducial markers, RetroLiDAR extends the reading distance by 2.6x compared to QR codes and by 44% compared to AprilTags, while reducing the median ranging error by 85%. We also present a low-power marker circuit design, a link budget analysis, and two proof-of-concept applications to validate the system's efficacy and practicality.

The burgeoning field of smart agriculture is increasingly leveraging unmanned aerial vehicles (UAVs) for data collection. However, inadequate visual features and plant occlusion can hamper visual-based simultaneous localization and mapping (SLAM) of UAVs. As a potential solution, RFID can work as an efficient reference infrastructure, enabling a connection between aerial imagery and real-world contexts. Despite this promise, hurdles remain in attaining high-accuracy, high-throughput, and long-range RFID localization, as well as practical deployment of RFID tags and reader implementation on UAVs. Overcoming these challenges holds significant potential, particularly considering their impact on numerous applications, such as large-scale agricultural management and plant stand reduction detection.

In a modern logistics network, high-performance automation of inventory tracking and package management calls for a reliable, high-throughput and long range RFID localization system. We present RF-Chord, the first RFID localization system that simultaneously meets all these requirements. RF-Chord features a one-shot multisine-constructed wideband design that can process the RF signal with a 200 MHz bandwidth in real-time to facilitate one-shot localization at scale. In addition, multiple SINR enhancement techniques are designed for range extension. Finally, we propose a kernel-layer-based near-field localization and a multipath-suppression algorithm that reduces the 99% long-tail errors.

Visible Light Backscatter Communication (VLBC) presents an emerging ultra-low-power IoT connectivity solution with high spatial-spectral efficiency and intrinsic human-perceivable privacy advantages. However, research progress on enhanced data rate and sophisticated device coordination of state-of-the-art VLBC systems still cannot meet the low-latency requirement (sub-second level for an IoT network) for massive connections.