搜索结果：npj Robotics

Accurate indoor positioning is vital for applications such as augmented reality and autonomous robotics. Channel state information (CSI)-based methods, particularly when combined with beamforming, massive multiple input multiple output (mMIMO) techniques, and artificial intelligence (AI) algorithms, offer enhanced indoor user equipment (UE) positioning accuracy and robustness in complex indoor environments. In this paper, we present an AI-driven CSI-based indoor positioning method for mMIMO systems, where channel impulse, channel frequency, and angular response domain features are extracted from the CSI data and combined to form both uni-domain and multi-domain feature sets. We introduce a deep attention network (DAN), an AI algorithm that leverages attention mechanisms to effectively integrate and process multi-domain CSI data, thereby enhancing UE positioning performance. We evaluate DAN using a publicly available mMIMO dataset and compare its performance against the baseline and multi-domain convolutional neural network (CNN) models. Our results show that multi-domain DAN outperforms CNN approaches in positioning accuracy, though at the cost of increased inference complexity-highlighting a trade-off between performance and computational overhead. These findings demonstrate the potential of attention mechanisms and multi-domain CSI features for accurate indoor UE positioning systems.

The Nobel Turing Challenge (NTC) proposes AI systems capable of autonomous, Nobel-level scientific discovery. The IJNTC2025 workshop convened researchers from India and Japan to advance this goal in health and biomedicine. This perspective synthesizes four themes: knowledge extraction, laboratory automation, hypothesis generation, and equitable healthcare applications. It identifies how Japan's robotics and precision AI expertise and India's large-scale data infrastructure and frugal innovation offer complementary strengths for collaborative progress toward the NTC.

Synthetic data generation across domains can bridge gaps between visual training, skill development, and personalized surgical planning, ultimately transforming how surgeons and artificial intelligence (AI) systems prepare for the complexities of the operating room. In this Perspective, we explore applications of synthetic data to advance surgical education and AI across three key areas: visual data synthesis for training surgeons and AI systems, surgical simulation for skill development and robotics, and digital twins for patient-specific surgical planning and guidance. These domains have largely remained siloed, but their integration has the potential to transform surgical training and AI development across the entire surgical workflow. To fully realize this potential, synthetic data must extend beyond routine surgical events to model atypical anatomy and intraoperative complications-the high-stakes clinical scenarios where enhanced training and AI support are most critical.

Human cognitive impairment associated with sleep loss, circadian misalignment and work overload is a major concern in any high stress occupation but has potentially catastrophic consequences during spaceflight human robotic interactions. Two safe, wake-promoting countermeasures, caffeine and blue-enriched white light have been studied on Earth and are available on the International Space Station. We therefore conducted a randomized, placebo-controlled, cross-over trial examining the impact of regularly timed low-dose caffeine (0.3 mg per kg per h) and moderate illuminance blue-enriched white light (~90 lux, ~88 melEDI lux, 6300 K) as countermeasures, separately and combined, in a multi-night simulation of sleep-wake shifts experienced during spaceflight among 16 participants (7 F, ages 26-55). We find that chronic administration of low-dose caffeine improves subjective and objective correlates of alertness and performance during an overnight work schedule involving chronic sleep loss and circadian misalignment, although we also find that caffeine disrupts subsequent sleep. We further find that 90 lux of blue-enriched light moderately reduces electroencephalogram (EEG) power in the theta and delta regions, which are associated with sleepiness. These findings support the use of low-dose caffeine and potentially blue-enriched white light to enhance alertness and performance among astronauts and shiftworking populations.

The electrical stimulation of the nervous system has shown great clinical potential in injury and pathology, yet experimentally driven practice makes it challenging to identify effective design choices and personalized stimulation protocols. This review outlines emerging model-based optimization frameworks that address these challenges by leveraging biophysical digital twins of neural interfaces. Enabling acceleration strategies and complementary data-driven approaches are also highlighted, along with key factors that currently limit clinical translation.

Gripping to smooth and wavy substrates, such as naturally occurring ice, presents a challenge for climbing robots in the field. Existing ice anchoring solutions require either substantial initial surface compression force (drilling; at least 50 Newtons) or require large energy expenditure (thermal picks; almost 1000 Joules). We present an anchoring mechanism capable of attaching to ice with lower initial surface compression force and lower energy consumption compared to drill-based or melt-based methods. The system leverages surface fracture caused by dynamic impacts with axes - inspired by mountaineers - to create indentations for grasping. A model describes the indentation depth, recoil energy, and surface compression force required for anchoring success, each as a function of impact energy. An integrated dual-ax gripper system successfully generates usable indents with as low as 8.3 Newtons of initial surface compression force and 8 Joules of combined mechanical potential energy on -14∘ C freshwater ice - a result consistent with first-principle model predictions. The gripper then successfully holds its own weight on steep glacier slopes in the field. These results indicate fracture-based grasping approaches are promising for climbing systems on ice. This concept can also apply to other surfaces such as wood, rock, and packed soil.

Although muscle mass loss is an emerging public health concern, its prevalence, associated factors, and clinical significance in Parkinson's disease (PD) remain unclear. This matched case-control study aimed to investigate the prevalence of low muscle mass (LMM) and to examine its association with orthostatic hypotension (OH) and orthostatic symptoms in 409 PD patients with Hoehn and Yahr stage ≤3, compared with 2045 age-, sex-, and height-matched controls from a nationwide database. OH was defined according to the international consensus. LMM was more prevalent in PD patients than in controls, particularly among men and those aged ≥70 years. Among PD patients, the prevalence of OH did not differ between those with and without LMM. Although LMM was linked to greater orthostatic blood pressure reductions at 30 s after standing, there were no differences in the frequency or severity of orthostatic symptoms according to LMM status. These findings suggest that although mild to moderate PD is associated with an increased risk of LMM, its impact on OH and related symptoms appears to be modest. Further longitudinal studies are needed to clarify the clinical implications of LMM in PD.

Coordinated trajectory planning is essential for multi-robot applications, ranging from factory automation to entertainment. The main challenge is providing long-term coordination guarantees, such as freedom from collisions, deadlocks, and livelocks, as well as kinodynamic agility, especially in densely populated environments. Although continuous optimization provides agility, it is computationally expensive. In contrast, discrete search is scalable but lacks physical realism for robot execution. This study introduces concrete planning, a hybrid approach that captures real-world continuous dynamics while maintaining scalable guaranteed planning via discrete search. We integrate recent advances in robot dynamics learning, optimal control, and anytime complete planning into a modular framework. The framework is deployed with 40 robots, including 20 aerial, 8 ground, and 12 obstacle robots, operating in a compact laboratory space. Despite the dense and time-varying setup, the robots achieve consecutive navigation missions on-demand, while executing aggressive maneuvers that substantially reduce task completion time.

Neuroinflammation contributes to the progression of many neurological diseases. Here, we explore whether ultrasound can reduce microglia-mediated inflammation in vitro and in vivo. We tested a broad range of ultrasound parameters in a BV2 microglial cell line, treated with lipopolysaccharide (LPS) to induce an inflammatory response. We found that specific combinations of centre frequency, acoustic pressure and treatment duration can significantly lower the levels of pro-inflammatory cytokines, including tumor necrosis factor (TNF)-α, interleukin (IL)-1β and IL-6. These effects lasted up to 72 h and were associated with the downregulation of the nuclear factor κB (NF-κB), suggesting a mechanistic link between ultrasound and inflammation. Further investigation in vivo, in LPS-treated mice, revealed a reduction in TNF-α expression in the hippocampus following ultrasound. Overall, our findings showcase the potential of ultrasound as a non-invasive therapeutic strategy to reduce neuroinflammation and restore brain homeostasis.

We aimed to establish a robust vision-language model ("Glio-LLaMA-Vision") for molecular status prediction and radiology report generation (RRG) in adult-type diffuse gliomas. Multiparametric MRI data and paired radiology reports from 1001 patients with adult-type diffuse gliomas were included in the institutional training set. A vision-language model, Glio-LLaMA-Vision, was developed from LLaMA 3.1 pre-trained on 2.79 million biomedical image-text pairs from PubMed Central and further fine-tuned from the institutional training set. The performance was validated in 100 patients and 75 patients with paired MRI-radiology reports from an institutional validation set and another tertiary institution (AMC), and in 170 and 477 patients with MRI from TCGA and UCSF datasets, respectively. In terms of IDH mutation status prediction, Glio-LLaMA-Vision showed AUCs ranging from 0.85-0.95 in the internal validation and external datasets. In terms of RRG, the BLEU-1 and ROUGE-L scores were 0.50 and 0.49 in the internal validation, respectively, and 0.32 and 0.36 on the AMC dataset, respectively. Overall, 37.8% of generated reports were considered superior or equal to the original reports, while 91.0% of generated reports were considered clinically acceptable by neuroradiologists. In conclusion, Glio-LLaMA-Vision demonstrates promising performance in molecular status prediction and RRG in adult-type diffuse gliomas, showing potential for clinical assistance.

Artificial intelligence (AI) is increasingly used in gastrointestinal endoscopy for polyp detection and classification. However, most AI models are trained on images from multiple video processors, whereas clinical environments typically rely on a single processor. We developed EndoStyle, a StarGANv2-based style transfer system trained to mimic the visual characteristics of five different endoscopic processors. On a validation dataset, Fréchet Inception Distance and Learned Perceptual Image Patch Similarity indicated high visual fidelity and perceptual similarity across processors. Semantic similarity analysis using three foundation models confirmed that converted images were equally consistent with both content and style inputs. In a multicenter study, endoscopists considered real and converted images realistic at comparable rates. When used to augment polyp detection model training, synthetic images significantly improved precision and specificity, reducing false positives by over 40% on two distinct evaluation datasets. EndoStyle thus offers a practical solution for processor-specific AI generalization.

Distinguishing epileptic seizures from parasomnias is challenging due to overlapping motor features. This study evaluated a SlowFast deep learning model using video recordings of 167 individuals to classify Sleep-Related Hypermotor Epilepsy, Disorders of Arousal, and REM Sleep Behavior Disorder. The model achieved a mean accuracy of 83.3% across three data splits. This work represents an initial step toward developing automated tools to support clinicians in assessing sleep-related motor events.

Vision is an essential part of attitude control for many flying animals, some of which have no dedicated sense of gravity. Flying robots, on the other hand, typically depend heavily on accelerometers and gyroscopes for attitude stabilization. In this work, we present the first vision-only approach to flight control for use in generic environments. We show that a quadrotor drone equipped with a downward-facing event camera can estimate its attitude and rotation rate from just the event stream, enabling flight control without inertial sensors. Our approach uses a small recurrent convolutional neural network trained through supervised learning. Real-world flight tests demonstrate that our combination of event camera and low-latency neural network is capable of replacing the inertial measurement unit in a traditional flight control loop. Furthermore, we investigate the network's generalization across different environments, and the impact of memory and different fields of view. While networks with memory and access to horizon-like visual cues achieve best performance, variants with a narrower field of view achieve better relative generalization. Our work showcases vision-only flight control as a promising candidate for enabling autonomous, insect-scale flying robots.

Accurate classification of renal masses before treatment is crucial for therapeutic decision-making and patient outcome. This study developed and validated Multi-Phase Attention Network (MPANet), a multimodal deep learning model integrating multiphase contrast-enhanced CT and clinical information, which can utilize both complete-phase and missing-phase CT data for multiclass classification of four common and easily confusable renal tumors-clear cell renal cell carcinoma (ccRCC), papillary renal cell carcinoma (pRCC), oncocytic neoplasms (including chromophobe renal cell carcinoma (chRCC) and renal oncocytoma (RO)), and fat-poor angiomyolipoma (fpAML). A total of 1688 multi-center cases were enrolled. Across all test sets, MPANet consistently outperformed single-phase models. In the internal test set, MPANet achieved a macro-average AUC of 0.850, a micro-average AUC of 0.865, and an accuracy of 73.3%. These results compared favorably to assessments by four radiologists based on CT (accuracies 43.6-62.4%) and two radiologists using MRI with clear cell likelihood score (ccLS) system (accuracies 52.5% and 49.5%). The net improvement rate of MPANet over radiologist assessment ranged from 10.9% to 29.7%. In the two external test sets, macro-average AUCs were 0.811 and 0.813, and micro-average AUCs were 0.867 and 0.909, respectively. MPANet shows potential as a clinical decision-support tool for personalized renal tumor diagnosis.

Food security faces growing challenges due to population growth, resource limitations, economic pressures, and industrialization-induced lifestyle changes. Traditional food systems struggle to adapt, necessitating innovative solutions and sustainable practices to meet future food demands. This review article explores emerging food system models and alternative food sources, including edible insects, seaweeds, plant-based and lab-cultured meats, underutilized crops, hydroponics, and next-generation fish farming. It highlights the role of food processing technologies such as blockchain, biotechnology, and robotics in enhancing sustainability, reducing waste, and improving food system efficiency. Consumer acceptance of engineered and fortified foods emerges as a critical factor in driving these innovations. The review also emphasizes the need for a transformative approach to food production, incorporating innovative technologies and sustainable practices to ensure food security by 2050. A coordinated effort to integrate alternate food sources and advanced processing methods will be vital for achieving a secure and sustainable global food future.

The therapeutic targeting of kinase signaling pathways represents a pivotal strategy in gastric cancer, yet the rational design of single agents capable of dual-kinase inhibition remains a challenge in precision oncology. Here, we develop the DuoKinaseNet, a dual-task spectral graph neural network that integrates global topological information from a heterogeneous biomedical graph to enable structure-preserving prediction of drug-kinase interactions. The core innovation of our model is the Structure-Preserving Spectral Expansion (SPSE) module, which injects global graph topology from a biomedical knowledge graph into the learning process via spectral coordinates and diffusion-distance biased attention. Evaluated on a comprehensive dataset curated from DrugBank, DuoKinaseNet achieves state-of-the-art performance, particularly on the challenging "unseen protein" benchmark, with an AUC-ROC of 0.903 for HER2 and 0.895 for FGFR2b. It significantly outperforms a wide range of baseline models, including 3D-aware methods and single-task variants, empirically validating the synergistic benefits of the dual-task learning and SPSE frameworks.

Deep learning techniques have significantly enhanced the convenience and precision of ultrasound image diagnosis, particularly in the crucial step of lesion segmentation. However, recent studies reveal that both train-from-scratch models and pre-trained models often exhibit performance disparities across sex and age attributes, leading to biased diagnoses for different subgroups. In this paper, we propose APPLE, a novel approach designed to mitigate unfairness without altering the parameters of the base model. APPLE achieves this by learning fair perturbations in the latent space through a generative adversarial network. Extensive experiments on both a publicly available dataset and an in-house ultrasound image dataset demonstrate that our method improves segmentation and diagnostic fairness across all sensitive attributes and various backbone architectures compared to the base models. Through this study, we aim to highlight the critical importance of fairness in medical segmentation and contribute to the development of a more equitable healthcare system.

Understanding surgical data in real-time will lead to improved feedback, learning, and performance for surgeons. This is important as data-driven systems offer safer, more standardized surgery, and faster training times. Artificial intelligence shows great promise in filling the gap where humans and classical computing algorithms cannot process information in an efficient manner. Defining a true application and development of robust artificial intelligence models mandates that they be explainable and transparent in how they make decisions. In this work, we meet this need by creating two models for surgical task and skill classification, respectively, to predict and provide an explanation of how surgical decisions are made. We further investigate how models based on a liquid time constant can be effectively utilized to develop better models under constraints and explain how the model makes internal decisions.

Computational research tools have reached a level of maturity that enables efficient simulation of neural activity across diverse scales. Concurrently, experimental neuroscience is experiencing an unprecedented scale of data generation. Despite these advancements, our understanding of the precise mechanistic relationship between neural recordings and key aspects of neural activity remains insufficient, including which specific features of electrophysiological population dynamics (i.e., putative biomarkers) best reflect properties of the underlying microcircuit configuration. We present ncpi, an open-source Python toolbox that serves as an all-in-one solution, effectively integrating well-established methods for both forward and inverse modeling of extracellular signals based on single-neuron network model simulations. Our tool serves as a benchmarking resource for model-driven interpretation of electrophysiological data and the evaluation of candidate biomarkers that plausibly index changes in neural circuit parameters. Using mouse LFP data and human EEG recordings, we demonstrate the potential of ncpi to uncover imbalances in neural circuit parameters during brain development and in Alzheimer's Disease.