The study introduces a domain generalization method for EEG that leverages domain-invariant features and data augmentation to improve cross-subject generalization and noise robustness. The model achieved state-of-the-art accuracy across three public datasets, demonstrating stability and strong noise robustness.
A robotic suction disc with a hybrid adhesion mechanism inspired by the lamprey has been developed, achieving impressive grip strength in both air and water. The device successfully lifted heavy loads and maintained its grip on rough surfaces, with effective adhesion time increased by up to 540% in water.
A novel framework uses multimodal large language models to predict ischemia and reperfusion risk, achieving high sensitivity and specificity in large-scale datasets. The system outperforms baseline models with a 4.8-9.5% relative improvement and maintains consistent performance across diverse subgroups.
A new EEG domain generalization method achieves state-of-the-art accuracy across three public datasets, outperforming baseline methods in noise robustness. The model uses domain-invariant feature and data augmentation to enhance cross-subject generalization, demonstrating robust solutions for BCI technology applications.
The SimTac simulator accurately models biomorphic sensors using particle-based deformation, light-field rendering, and neural network prediction. It achieves high performance in optical and mechanical response accuracy, adaptability to diverse morphologies, and simulation efficiency.
A groundbreaking system repurposes a vehicle's taillight as an LED matrix transmitter to enable data transmission between vehicles in platoons. The approach eliminates the need for roadside units or photodetectors, offering a secure and low-cost solution for V2V communication.
The EMG-driven EVF robot enhances both voluntary motor control and sensory feedback to the wrist and hand muscles, improving sensorimotor integration and restoring balanced motor control. Significant improvements were observed in Fugl-Meyer Assessment scores and Action Research Arm Test scores.
The Marker-GMformer model, a lightweight deep learning approach, accurately predicts lower limb biomechanics with high correlation coefficients (>0.97) and low RMSE values (1.95° for joint angles). It offers real-time inference suitable for applications requiring fast feedback, such as robotic control and monitoring.
The Beijing Institute of Technology Press Co., Ltd is recruiting young editorial board members who will contribute to the publication process and benefit from training and networking opportunities.
A new circular motion paradigm enables continuous decoding of hand motion angles using EEG signals and polar coordinates. The experiment showed excellent performance across six deep learning models, with the best model achieving a mean squared error of 1.012 rad².
The RDP-type actuator enhances bidirectional consistency and motion linearity due to its symmetric driving structure and constant contact force. It demonstrates excellent performance across low and high-frequency ranges, showing minimal speed fluctuation and strong load-bearing capacity.
This study identifies shared genetic variants linking schizophrenia with left- vs right-hemisphere white-matter tracts, revealing hemispheric asymmetry. Hemispheric specificity was observed in the left posterior thalamic radiation and tapetum, with higher proportions of shared loci.
Researchers developed a nanodelivery system that synergizes ferroptosis and STING activation to treat metastatic bladder cancer. The system showed robust tumor control and improved survival rates, with enhanced immune response and reduced metastasis.
The system uses a human-in-the-loop knot planner and three optimization metrics to ensure task reliability. It successfully completed real-world experiments, lifting heavy payloads in unstructured environments with high success rates.
The study investigates regional mechanical variability in porcine pia–arachnoid complex (PAC) using a region-by-region, multi-scale design. Regional differences in mechanical properties are linked to biochemical distributions via two-photon imaging and RNA-seq.
The review explores how multiplayer games, social robots, and virtual agents can advance artificial empathy in therapy and healthcare by embedding interpersonal interaction. It highlights the need for deeper, more seamless closed-loop interaction and emphasizes the importance of emotion recognition, trust, engagement, and rapport.
A new model predicts cerebral blood flow using carotid ultrasound and brain MRI data, enabling non-invasive monitoring of astronauts' cognitive function during long-duration spaceflight. The CatBoost algorithm outperforms other models, providing a practical solution for safeguarding brain health in space.
MMWs inhibit tumor growth by modulating immune proteins, reducing tumor cell membrane synthesis and energy metabolism. The therapy boosts anti-tumor immune responses, increasing infiltration of immune cells and reducing immunosuppressive populations.
Researchers develop an intelligent decision-making algorithm using functional Near-Infrared Spectroscopy (fNIRS) to analyze passengers' physiological states and enhance autonomous vehicle safety. The study shows improved safety, comfort, and convergence speed compared to traditional methods.
Researchers propose a theoretical model to analyze adhesion and escape phenomena during low-velocity impacts between charged dust particles and spacecraft. The study focuses on the interaction between charged particles and spacecraft within a plasma sheath, considering significant size differences.
Space: Science & Technology has been officially indexed in the Web of Science: Science Citation Index Expanded (SCIE) since December 8, 2025. The journal aims to promote innovation and breakthroughs in space science and technology.
Hybrid magneto-acoustic microrobots combine magnetic steering with acoustic propulsion for precise navigation and powerful locomotion. These robots enable targeted drug delivery, minimally invasive surgery, and medical imaging with high accuracy and efficiency.
This study uses MEA-based graph deviation networks to predict autism syndrome signatures in human forebrain organoids exposed to valproic acid. The results show that VPA exposure induces detectable network-level functional alterations, with session-specific models achieving high accuracy in class separation.
A new hybrid approach combines neural perception and acoustic feature learning to detect acoustic targets with high accuracy. The system leverages neuroanatomical priors, confidence-driven fusion, and streaming-mode validation to achieve robust performance.
This innovative module combines vibration-assisted penetration and real-time force sensing to reduce tissue damage and inflammation risks. The device achieves high precision and accuracy, making it suitable for biomedical applications such as neural probe implantation and single-cell puncture.
This study introduces a carrier-free PDR nanoassembly with a tumor-suppressing peptide, pH-sensitive daunorubicin prodrug, and siRNA targeting the LILRB4 gene, achieving targeted therapy of acute myeloid leukemia. The combination treatment significantly enhances cytotoxicity and anti-leukemia effects.
A new platform integrates magnetic actuation with electrophysiological sensing for high-fidelity, multisite monitoring on dynamic ex vivo tissue surfaces. The soft electrode platform successfully recorded stable signals from cardiac tissue and demonstrated excellent motion control and stretchability.
The study introduces BGMix, which separates task-related components from background EEG to generate valid samples. AETF is designed to process the augmented data and capture EEG's spatiotemporal-frequential features using a Transformer-based attention mechanism. This integrated solution improves the practicality of high-speed SSVEP-bas...
This survey reviews cutting-edge hydrogen tank technologies, exploring how to safely store gaseous or liquid hydrogen in extreme conditions. Key advancements promise to transform aviation, with materials like carbon fiber-reinforced polymers offering exceptional strength-to-weight ratios and reducing the weight of hydrogen tanks.
A new monolithic synaptic device mimics human CT afferents for robot emotional interaction, achieving ultralow threshold sensitivity and high energy efficiency. The device enables real-time processing of gentle touch into emotional signals.
Researchers develop comprehensive model to analyze electrostatic and contact interaction between low-velocity lunar dust and spacecraft. The study focuses on the effects of plasma sheaths and electric charges on dust particles, aiming to improve long-term extravehicular activity and permanent station establishment on the lunar surface.
Legged robots face difficulties in hardware design and control due to impact force and kinematic complexity. Single-leg robots offer a simplified alternative for hopping and biomimetic motion, with various structures, modeling, and control strategies being explored.
A new synchronization method for satellite-ground hopping beam communication is proposed, enabling efficient use of frequency resources and power. The method uses a signaling-assisted fast synchronization technique to match business signals between satellite and ground signal stations.
BioCompNet, a dual-channel 2D U-Net framework, enables automated MRI quantification of multiple tissue compartments, including adipose tissue, muscle, and bone. The model achieves high accuracy and efficiency, with a complete pipeline processing each case in under 1 minute.
This research uses D-STATCOM to dynamically balance loads and supply reactive power at charging stations, improving power quality and reducing energy waste. The study showcases superior reactive power management, stabilizing the low-voltage distribution network and enabling more reliable EV infrastructure.
Researchers develop a game-changing magnetic analysis method to authenticate lithium-ion batteries onboard vehicles, ensuring safety and reliability. The breakthrough enables instant detection of counterfeit or low-quality batteries without invasive checks.
Researchers developed a new P2D-coupled non-ideal double-layer capacitor model to analyze lithium-ion batteries under high-frequency periodic signal excitation. The model considers neglected electric double-layer capacitance and its dispersion effects, enabling more accurate mechanism analysis and performance degradation assessment.
Developed a hybrid sampling technique combining dual-mode distal B-scan imaging measurement to actively locate lesion centers and enable programmable boundary segmentation. The system achieved improved shape fitting accuracy, lowered center error, and enhanced real-time performance.
Researchers propose FedM2CT, a federated metadata-constrained method that enables simultaneous reconstruction of multivendor CT images with different imaging geometries and sampling protocols in one framework. The framework outperforms competing methods on objective metrics, achieving higher PSNR/SSIM and lower RMSE.
Researchers developed wearable components for cyborg insects, preserving natural functions and enabling stable neural responses. The design uses multimaterial 3D printing and surface stimulators to control motion and navigation.
Researchers developed advanced imaging strategies using intelligent micro/nanomotors, enabling enhanced detection sensitivity and real-time tracking of subcellular events. The proposed approach integrates MNMs as dynamic contrast agents for multimodal diagnostics, providing a theoretical framework for bioimaging technologies.
A new SnS₂-based in-sensor reservoir computing device introduces tunable multi-timescale optoelectronic dynamics to address high-speed and low-speed movement recognition challenges. The device achieves 100% accuracy in motion recognition tasks using the Weizmann dataset, surpassing conventional networks like LSTM.
Developed a novel EEG transformer model, Augmenting Electroencephalogram Transformer (AETF), to enhance steady-state visually evoked potential-based brain–computer interface (BCI) systems. The AETF model improves decoding efficiency by capturing temporal, spatial, and frequency features of dynamic EEG signals.
This study evaluated the combined effects of epidural electrical stimulation (EES) and physical therapy (PT) on sensory, motor, and autonomic functions in SCI patients. EES+PT significantly enhanced long-term sensory function, muscle strength, spasticity, and urinary function compared to PT alone.
A novel centimeter-scale quadruped piezoelectric robot demonstrates key advancements in miniaturization, actuation, control, communication, and power supply. The robot achieves fast locomotion, exceptional robustness, and real-time image sensing capabilities.
Researchers developed an acoustofluidic device for efficient isolation and biomarker-specific detection of small extracellular vesicles. The device achieves 6-fold signal enhancement for EGFR-positive sEVs in just 20 minutes, offering a portable and low-cost alternative to Western blotting.
A bioinspired soft robotic system combines a steerable, elongating soft arm with a leech-inspired three-finger grasper to overcome limitations of existing ESD robots. The system achieves high precision, force and endurance, but requires improvements in fine force control, haptic feedback and sterilization for clinical translation.
The multimodal limbless crawling soft robot features a foldable, multistable kirigami skin that maintains stable frictional anisotropy under large deformations. It achieves straight crawling, in-place rotation, and lateral turning, resolving the trade-off among large deformation, controllable friction, and agile steering.
A comprehensive study reveals critical vulnerabilities in smart railways, from data breaches to operational sabotage. Key findings propose comprehensive mitigation strategies, emphasizing human factors and staff training in building cyber-resilient systems.
Artificial neural networks offer superior predictive accuracy in predicting biodiesel properties and enable rapid assessment of diverse feedstock options. Hybrid models combining generative and discriminative approaches achieve significant yield improvements and optimize biodiesel production from waste cooking oil.