Sensors

Passive radio frequency identification (RFID) localization is attractive for low-cost indoor positioning, but mixed line-of-sight (LOS) and non-line-of-sight (NLOS) propagation causes both RSSI fluctuations and carrier-phase ambiguity. This study proposes a coarse-to-fine PRTFC–SD-PDOA framework in which an RSSI fingerprint likelihood defines the admissible decision domain, candidate-wise phase ambiguity is resolved inside that domain, and normalized RSSI and phase costs are jointly minimized. Dominant-reflection compensation is used when the reflecting surface is known, and temporal filtering is applied to repeated observations. In independently repeated simulations, the joint estimator achieved RMSEs of 0.098 m in LOS and 0.250 m in the mixed scene. In a nine-position R420 experiment, the particle-filtered estimator achieved mean point-wise RMSEs of 0.122 m in LOS and 0.457 m in the mixed scene; the corresponding Kalman-filtered results were 0.118 and 0.424 m. Thus, both filters reduced the variability of the unfiltered joint estimates, while the Kalman filter was slightly more accurate for the stationary, approximately Gaussian measurement sequences considered here.

The dynamic development of non-invasive medical diagnostics and continuous health monitoring using biomedical signals in recent years has been one of the main research directions in biomedical electronics. This paper presents a developed multimodal, multi-site, and synchronous measurement system designed for continuous cardiovascular signal acquisition. The system integrates reflective and transmissive photoplethysmographic (PPG) sensors and active microwave (MW) sensors operating at a frequency of 4.9 GHz. One of the reflective PPG sensors enables multispectral (MWPPG) recording for six different light wavelengths, ranging from blue (465 nm) to infrared (940 nm). The system was evaluated in a trial involving a group of 32 volunteers. During our research, the PPG sensors were placed on the fingers of the left hand, wrist, and forehead, while the MW sensors were placed on the thumb, wrist, biceps, and chest. In addition, ECG signals and accelerometer signals are recorded synchronously. The averaged pulses of the individual sensors were appointed to compare their shape and quality depending on the technology and location of the measurement. The PPG signals exhibited higher signal quality, with mean SNR values ranging from 8.66 ± 2.17 to 12.61 ± 3.98 dB, compared with 5.91 ± 0.98 to 7.95 ± 1.26 dB for the MW signals. Similarly, the pulse-detection F1-score ranged from 0.34 ± 0.22 to 0.74 ± 0.17 for PPG and from 0.33 ± 0.07 to 0.43 ± 0.15 for MW sensors. Significant differences in the shape of the pulses from the MW sensors were also demonstrated between individual participants. Furthermore, the paper presents examples of recorded signals and discusses the advantages and limitations of various sensor technologies and their locations, while also indicating potential directions for further research.

Leveraging discrete sequence prediction for manipulation skill learning has demonstrated immense potential for complex long-horizon tasks. However, guaranteeing precise and robust execution under complex scenarios remains a core challenge for efficient autonomous robotic operation. This demands execution policies capable of generating precise and efficient action sequences based on perceptual feedback. To address the loss of fine-grained action details induced by action discretization and the difficulty in preserving long-horizon temporal dependencies in long-horizon tasks, this work proposed a hierarchical manipulation skill learning method based on residual finite scalar quantization (RFSQ). At the action representation level, a robust RFSQ with an adaptive linear scaling mechanism was introduced. It effectively mitigated residual decay in multi-level action quantization and realized accurate discrete reconstruction of continuous action sequences. At the action sequence generation level, a hybrid hierarchical architecture integrating Mamba2 and Transformer was constructed. Mamba2 efficiently captured temporal dependencies among actions, while Transformer precisely modeled fine-grained residual dependencies across action quantization hierarchies. A multimodal perception fusion module was further incorporated to strengthen instruction-semantic alignment on language-instructed tasks. A series of experiments were conducted on the LIBERO and Meta-World simulation environments. Experimental results demonstrated that the proposed method achieved outstanding performance in multi-task and long-horizon task learning.