In recent years, the frequency and scale of natural disasters have increased, and there is a growing need to deliver supplies quickly and flexibly to affected areas where transportation networks are disrupted. Aerial deployment is one possible solution. However, conventional cardboard boxes and containers have fixed shapes and cannot be easily reconfigured at the site. In addition, parachutes used for deceleration are large, may cause collision risks, and are strongly affected by wind, which makes it difficult to control the landing position. This study aims to develop a modular device that can be reassembled after aerial deployment. This paper focuses on its falling behavior during deployment. Inspired by the flight mechanism of wind-dispersed seeds, we designed a blade structure to induce autorotation and generate lift for stable descent. Free-fall experiments were conducted using a prototype modular device. We evaluated the effects of blade attachment angle, installation inclination, and blade twist on falling time and posture stability. The results show that specific attachment angles and inclination conditions produced stable autorotation, leading to longer falling time and improved stability. These findings provide basic design guidelines for achieving stable descent of modular aerial deployment devices.
Labor shortages driven by aging workforces have increased demand for robotic automation on construction sites, yet existing LLM-based systems represent task plans in formats that field operators cannot readily inspect or verify, and provide no mechanism for targeted modification once execution has begun. This paper presents a human-verifiable execution-time DAG refinement framework in which the active plan is rendered as a color-coded node-edge graph that operators can inspect and confirm at each step. Natural-language corrections are translated into targeted edits of the pending subgraph, with completed subtasks locked and in-progress subtasks paused only when necessary for a validated update. A refinement taxonomy and validation pipeline enforce execution-state consistency before each update is dispatched. Evaluation in simulation, a 12-case error-correction study spanning three complexity levels, and deployment on an edge device confirm practical feasibility. A controlled user study further shows that the DAG-based operator interface achieves 94.2
In this study, we propose a novel method called Entropy Regularized Robust Sparse Fuzzy Clusterwise Regression (ERSFCR). This method is designed to perform clustering and regression simultaneously while maintaining robustness against outliers. By employing entropy-based fuzzy regularization, ERSFCR integratively addresses challenges associated with high dimensionality, sparsity, and the presence of outliers, which are commonly encountered in real-world datasets and practical applications. From the numerical simulation, the results demonstrated that ERSFCR performed particularly well for datasets with a high proportion of outliers, achieving the lowest median absolute residuals among all compared methods. Moreover, by comparing six different methods under various conditions, we were able to identify which approaches are most effective depending on the characteristics of the datasets, thus providing practical guidance for selecting appropriate modeling strategies in diverse scenarios. In the real data analysis, ERSFCR also achieved the lowest RMSE, outperforming all comparison methods. These results indicate that ERSFCR effectively mitigates the influence of outliers while maintaining predictive accuracy. Overall, the proposed approach provides a robust and flexible framework for simultaneous clustering and regression in challenging high-dimensional scenarios.
As humanoid robots become widespread in our daily lives, providing psychological safety against their sudden and unpredictable behavior has become a critical issue. To reduce this psychological barrier, this study proposes a method that projects a shadow before the robot’s activation to preview its motion. In order to evaluate the effectiveness of this method, two experiments were carried out. In Experiment 1, human subject impressions were evaluated using the semantic differential (SD) method. The results showed that shadow projection significantly improved predictability for all evaluated actions. Additionally, the increased predictability led to a substantial increase in the perceived sense of security and reduced surprise. This effect was particularly pronounced for the “Hugging” action, which users anticipated as an invasion of their personal space, confirming that the shadow effectively mitigated threats from approaching actions. In Experiment 2, the optimal timing for the preview was investigated. The optimal delay time was approximately 5 to 6 s on average. Qualitative analysis showed user preferences divided into two groups: “Meaning-Focused” users who needed time to fully understand the robot’s purpose, and “Precursor Detection-Focused” users who preferred immediate reactions. These results suggest that shadow projection is highly effective for promoting psychological safety, but for practical application, it is necessary to develop mechanisms that dynamically control the prediction timing based on individual needs and task content.
Lung CT analysis requires reliable feature representations, while fully supervised training is constrained by expert annotation cost. This study presents the Focal-SE Masked Autoencoder (Focal-SE MAE, where SE denotes Squeeze-and-Excitation), a self-supervised masked reconstruction framework for lung CT representation learning. The method retains the MAE reconstruction objective and modifies the encoder to jointly model multi-scale spatial context and channel responses. The Focal-SE MAE encoder was pre-trained without external pretrained weights. For downstream transfer, the encoder was separately pretrained using only the private patient-level training split, frozen, and evaluated as a feature extractor for four-class 64-slice CT sequence classification. On LUNA16, Focal-SE MAE achieved the best reconstruction among the evaluated MAE encoder variants using the common decoder over five independent runs, with a mean absolute error of 1.325× 10^-3 , a peak signal-to-noise ratio of 46.54 dB, and a structural similarity index measure of 0.9939. In downstream four-class 64-slice CT sequence classification on a private CT dataset, it achieved an accuracy of 92.68
Large language models (LLMs) can support predictive maintenance by reasoning over text-based expert knowledge, but their reliability depends on how such knowledge is structured and retrieved. This work presents an expert-knowledge-grounded LLM inference pipeline that combines Delphi- and FMEA-derived maintenance knowledge, retrieval-augmented prompting, and engineered telemetry summaries. In a multi-robot case study, we evaluate basic knowledge queries, complex diagnostic questions, and telemetry description tasks. Retrieved expert knowledge improves LLM-based diagnostic answers, while a deterministic knowledge-graph reasoner performs best on threshold-driven telemetry questions. The results indicate a complementary design in which graph inference provides traceable rule execution and expert-knowledge-grounded LLMs synthesize diagnostic explanations from selected evidence.
In this study, we propose a framework that incorporates subjective evaluations provided by a Vision-Language Model (VLM) into the fitness evaluation and selection processes of a genetic algorithm. As the target of evolution, we employ virtual soft robots with flexible morphologies and locomotion and present the VLM with sequence images representing the locomotion of two individuals. Selection is performed via pairwise comparisons based on subjective evaluation terms such as adorably and weirdly. The outcomes of these comparisons are used as selection pressure within the genetic algorithm, enabling the simultaneous evolution of morphology and locomotion. Experimental results demonstrate that subjective selection by the VLM accelerates population convergence compared to random selection, while also giving rise to distinctive morphologies and motions corresponding to each evaluation term. An auxiliary experiment with human participants further showed that, although individual pairwise choices only partly agreed with the VLM selections, the resulting morphological and locomotion tendencies were qualitatively similar and repeated human evaluations imposed noticeable fatigue. Moreover, the observation that similar evolutionary outcomes emerged across different evaluation terms suggests that the VLM does not apply these terms in a purely literal manner but instead decomposes them into multiple internal evaluation criteria when making judgments. This work visualizes the evolutionary process through which subjective linguistic expressions are mapped onto embodied phenotypes and provides a foundational framework for analyzing the structure of subjective judgment in VLMs. The proposed approach is expected to contribute to new developments in evolutionary computation and artificial life research based on subjective evaluation.
Monitoring peripheral circulation is crucial for the early detection of peripheral artery disease (PAD) and diabetic foot complications. In this study, we analyzed the phase distribution of pulse wave signals on the dorsum of the foot using remote photoplethysmography (rPPG), a non-contact, non-invasive measurement method. We captured images of the dorsum of the foot under green light using an industrial camera and created phase color maps from the pixel-by-pixel rPPG signals. The results revealed a phenomenon in which the phase reversed by approximately 180 degrees between adjacent regions in the central part of the dorsum and near the ankle. This phase reversal likely reflects the mechanical deformation (compression and expansion) of the dermis associated with subcutaneous arterial dilation, suggesting that the phase difference in rPPG could be used to non-invasively estimate the location of arterial course.
Multi-agent robotic systems (MARS) are well-suited for collective tasks; however, their efficient operation in decentralized settings requires inter-agent communication for coordination. While global broadcasting enables complete information sharing, it incurs substantial transmission power and bandwidth costs. Consequently, recent studies have proposed algorithms that operate under partial communication or limited communication ranges. These approaches typically rely on local planning or learning-based agents to compensate for missing or incomplete information, and evaluate performance degradation primarily as a function of communication range. In MARS, both the frequency of communication and the content of exchanged information significantly affect collective performance; however, these factors are often treated implicitly and rarely studied as explicit parameters. In this work, we introduce a dual-variable impact analysis that explicitly considers both communication range and communication content. Rather than compensating for information loss through complex planning or learning mechanisms, we isolate and examine how communication constraints directly influence collective performance. Using simple rule-based agents that perform cooperative area coverage with explicit knowledge-sharing and reasoning, we systematically evaluate how much performance can be recovered through minimal communication mechanisms. Our results demonstrate that appropriately designed communication content can partially recover information loss and maintain task efficiency even when agents are outside the direct communication range.
Legged robots offer remarkable adaptability in navigating complex environments, but achieving reliable direction control remains a challenge, especially under decentralized controls that avoid high computation, prior training, or detailed modeling. While such controls enable rapid gait formation using only minimal proprioceptive feedback and robot–environment interaction, they often result in unstable heading behavior, such as unintended turns or a random walking direction. In this study, we propose a differential adaptive steering control mechanism that operates in conjunction with decentralized locomotion control to stabilize and guide the robot’s heading using only body orientation in the yaw angle. Our method modulates the leg movement amplitude to achieve target-aligned directional adaptation. We validated our approach on a simulated stick insect-like robot on both flat and rough terrains of varying roughness, including straight and complex paths. Additionally, we demonstrated the proposed control on a physical stick insect-like robot on flat terrain, where the robot achieves stable straight-line locomotion.
This paper proposes an integrated autonomous robotic welding algorithm that combines a deep learning-based vision model with a 3D sensing system. The framework utilizes an AI model trained on 2D images to automatically detect and localize the welding region, which is then mapped onto the corresponding 3D point cloud data. To ensure precision, a convexity-based geometric analysis is performed to extract workpiece edges and evaluate the spatial relationship between these edges and their surrounding surfaces. This process enables accurate detection of gaps along the welding path and the generation of an optimized welding trajectory. The technical feasibility and practicality of the proposed method were experimentally validated using a 6-DoF robotic manipulator equipped with a 3D laser scanning system.
In this paper, a robust control approach for the differential drive wheeled mobile robot (DDWMR) motion, subjected to the skid, slip, and violation of the nonholonomic constraints is presented. The proposed kinematic controller “Perfect Velocity Tracking”, laid on the Lyapunov theory, generates the ideal velocity to ensure the asymptotic convergence of the state vector error. Then, based on the previous output, a new adaptive dynamic power rate sliding-mode control approach is introduced to make the overall system asymptotically stable by generating the desired actuator torques. These two controllers are introduced to promote the tracking motion even in the presence of parameter uncertainties and external disturbances. During the simulation work, an accurate kinematic and dynamic modeling of the differential drive Pioneer 3dx, based on the magic formula’s traction forces, is proposed to simulate the real case motion of the DDWMR.
Peristaltic pumps are widely employed in fields such as medicine, food industries, and biotechnology. In nature, it is known that gastropods such as snails, which utilize peristaltic motion for locomotion, generate propulsion in the direction opposite to the propagation of their muscle contraction waves. Therefore, in this study, we developed a transport device capable of bidirectional transport within a single channel by employing a mechanism that mimics this gastropod peristalsis. The device is composed of a parallel link mechanism and a flexible silicone rubber tube, inducing continuous peristaltic motion in the tube solely through motor rotation. By utilizing eccentric shafts to generate waveform motion and continuously converting the force transmitted to the tube section, smooth wave-like motion was achieved. Experimental results demonstrated that while liquid is transported in the direction of wave propagation, slender objects constrained by the tube walls are transported in the opposite direction. This paper presents the proposal and analysis of this transport mechanism and aims to provide new insights into conventional fluid and object transport technologies.
This paper deals with a two-stage identification of continuous-time Hammerstein systems using a Gaussian process (GP) model. The Hammerstein system is described by the cascade connection of a nonlinear static part followed by a linear dynamic part. In the first stage, the nonlinear static part represented by the GP model is estimated based on multiple sets of data of constant input and corresponding steady-state output. Then, in the second stage, the linear dynamic part is estimated by the linear least-squares method based on the approximated discrete-time estimation model derived by a digital pre-filter. Since the training of the GP model is limited to estimating the nonlinear static part only, the proposed method can be applied to systems whose linear dynamic parts have arbitrary order and can significantly reduce the computational burden. The effectiveness of the proposed method is confirmed through numerical experiments.
3D point clouds detection plays a critical role in autonomous driving, robotic control, intelligent manufacturing et al. Existing methods with single-sampling strategies may focus on long-range context, but lose fine-grained local details during feature aggregation. Moreover, the inherent sparsity of point cloud data also limits effective multi-scale feature learning, leading to degraded feature representations and inaccurate object location, particularly in complex scenarios. Thus, a multi-window parallel voxel transformer with center optimization strategy network is proposed in this paper, which integrates multiple sampling strategies and center optimization strategy into the transformer-based architectures to enhance multi-scale feature representation. Specifically, a multi-scale cross-attention module is introduced to fuse voxel features across different scales through parallel self-attention and multi-window sampling, enabling effective aggregation of both local and global information. Additionally, positional encoding is incorporated to enhance spatial awareness, improving the effectiveness of hybrid-scale feature aggregation. Furthermore, a novel center optimization strategy to mitigate localization errors caused by imprecise feature aggregation. Instead of relying on single voxel-based center, multiple candidate object centers are generated from high-confidence keypoint features and refined using confidence-weighted averaging. This confidence-aware refinement produces a more accurate and robust object center estimation, improving bounding box localization performance in sparse, occluded, or boundary-obscured scenarios.
Urban air mobility (UAM) is a rapidly developing field that proposes a novel transportation service concept, also known as “air taxi”, to alleviate congested transportation with high density in urban areas. This research focuses on developing a flight model for UAM vehicles that incorporates a energy consumption model. This integrated model is designed to evaluate uncertainties inherent in UAM operations, such as the effect of wind, energy consumption, and congestion at arrival ports. This study created a flight plan and conducted simulations using the developed flight model to assess the effectiveness of traffic management from the perspectives of energy consumption and holding time. Specifically, it evaluates the efficacy of traffic management in accurately predicting arrival times and congestion under uncertain conditions, by comparing and assessing three distinct methods: ground holding, airborne holding, and hybrid method.
We investigated cuffless blood pressure estimation from facial videos using remote photoplethysmography (rPPG) with an RGB–NIR camera. We obtained data from 31 participants with a chin rest and five without a chin rest, and rPPG signals from the nasal and cheek regions were processed to extract waveform features. Blood pressure was estimated using a light gradient boosting machine (Light GBM) and evaluated using leave-one-subject-out (LOSO) and within-subject sixfold cross-validation. Compared with prior support vector regression, Light GBM showed improved accuracy (LOSO: systolic blood pressure (SBP), 11.9 mmHg; diastolic blood pressure (DBP), 6.7 mmHg; sixfold: SBP, 8.9 mmHg; DBP, 5.2 mmHg). According to British Hypertension Society (BHS) standards, DBP estimation achieved Grade A, whereas SBP was classified as Grade D, albeit close to Grade C. Shorter acquisition durations of 5 and 10 s, compared with the conventional 60 s recording, as well as chin rest-free measurements, demonstrated that cuffless blood pressure estimation was feasible in the absence of large facial motions that disrupted rPPG signals. These results demonstrate the feasibility of cuffless blood pressure estimation from RGB–NIR facial videos using Light GBM, particularly for DBP estimation.
This paper discusses the potential benefit the use of propellers can provide in terms of balance performance on a flat ground operation of a wheeled inverted pendulum robot. The balance performance is quantify by looking at the response of the robot to a push disturbance. Model Predictive Control (MPC) and Proportional Integrator Derivative (PID) linear corrector use are analyzed in Python simulation. Since the proposed PID-based control framework demonstrated favorable simulation results and offered simplicity of implementation, it was integrated into the real robot, where its effectiveness was confirmed.
This paper presents set-based differential evolution with exponential and binomial crossovers for discrete optimization problems. While differential evolution has been traditionally applied to continuous optimization problems, it can be adopted to discrete optimization problem by extending techniques in continuous space into discrete space by employing set-based representation. In our approach, a candidate solution is defined as a crisp set, and all arithmetic operations in mutation are redefined through novel operations. The mutation operator in our algorithm adds two different solutions selected randomly to the current solution and the new solution is constructed probabilistically. For two crossovers, the number of inherited mutant parameters follows an exponential distribution in exponential crossover and that of inherited mutant parameters follows a binomial distribution in binomial crossover. This study investigates exponential and binomial crossover mechanisms within the framework of set-based differential evolution. Specifically, we examine the influence of crossover rate on solution accuracy in traveling salesman problem, conducted in two phases: broad search and narrow search. To demonstrate the effectiveness of our algorithms, we examine numerical experiments and compare results with existing algorithms.
Acupuncture is increasingly used to improve facial skin conditions. Contact needle technique (CNT) is a traditional Japanese acupuncture method. This pilot study aimed to investigate the acute effect of CNT by evaluating changes in capillaries and improvement in skin moisture content. Fifteen women participated in this study; they received CNT to the right side of their face for 30 min. The primary outcomes were the change in the skin moisture content and the shapes of the capillaries before and after CNT. Skin moisture content significantly increased after treatment. The number of blood vessels increased after treatment. We also observed that the capillaries became thicker, more numerous, and more active after treatment. We propose that the skin condition can be influenced and improved by CNT.
