Kyutacar : Kyushu Institute of Technology Academic Repository
Not a member yet
9746 research outputs found
Sort by
Extraction and Summarization from Visiting Nurse Transcriptions Using Improved Prompt Techniques
Get PDFIn this paper, we propose a modular framework that integrates fewshot and Generated Knowledge Prompting (FS-GKP) for health information extraction and summarization from nurse-elderly conversation transcripts. This tasks is essential for monitoring elderly patients and assisting nurses in completing the visiting nurse form. FS-GKP generates additional domain-specific knowledge from transcription data, which serves as the basis for more accurate extraction and summarization. FS-GKP uses a structured chain of prompts that allows each step to build on the previous step, thus improving interpretability and precision. Experiments reveal that the GKP using few-shot technique significantly enhances extraction performance with average accuracy across all health categories is 78.57%, outperforming individual methods like zero-shot (52.49%) and few-shot (45.24%). FS-GKP also provides the best results for the summarization task compared to the other five techniques (zero-shot, few-shot, Chain-of-Thouth (CoT), Self-consistency, Few-shot CoT) with ROUGE-1: 0.43, ROUGE-2: 0.22, ROUGE-L: 0.32, BLEU: 0.28, BERTScore Precision: 0.75, Recall: 0.72, F1: 0.73, and SBERT Cosine Similarity: 0.83. These results highlight the potential of FS-GKP, to improve the accuracy of health information extraction and streamline the summarization process, effectively aligning it with categories in visiting nurse forms.journal articl
Material-Inspired Echo State Network for In-materio Physical Reservoir Simulation
Get PDFconference pape
アクティブサスペンションを有する四輪型惑星ローバーの耐故障性と地形適応型ロコモーション
Get PDF九州工業大学九州工業大学博士学位論文（要旨）学位記番号:工博甲第608号 学位授与年月日:令和7年9月25日thesi
Fundamental study on the impact of herbaceous plant root systems on slope surface stability
Get PDF植生工はのり面に植物を生育させることで表層地盤を補強する工法である。既往研究により，植物根系が斜面表層に侵入して地盤を緊縛することで，土のせん断抵抗が増加し，その結果として，土の強度定数である粘着力 c や内部摩擦角 φ，またはその両方が上昇することが明らかとなっている。本研究では草本植物を生育させた供試体を作製し，3段階の鉛直応力条件下で定圧一面せん断試験を実施して，植物根系によって補強された土の強度定数を求めた。さらに，得られた土の強度定数を基に無限長斜面を想定した安定解析を行い，植生工が斜面表層の安定性を大きく向上させることを確認した。Vegetation works involve reinforcing surface soil by growing vegetation on slopes. Previous studies have shown that plant root systems, invading the slope surface and binding the soil, increase the shear resistance of the soil, consequently increasing the soil strength parameters, i.e., cohesion (c), internal friction angle (Φ), or both. This study prepared soil specimens with herbaceous plants growing on them and conducted constant pressure box shear tests under three different vertical stress conditions to determine the strength parameters of the soil reinforced by plant root systems. Furthermore, as part of this study, based on the obtained strength parameters of the soil, a stability analysis was conducted assuming an infinite length slope, and it was confirmed that vegetation works significantly improved slope surface stability.journal articl
Detecting Bacteria from Gram Stained Smears Images by the Family of YOLOs
No full textIn this paper, we focus on 14 types of bacteria, that is, 4 types of Gram positive cocci (Enterococcus faecalis, Staphylococcus aureus, Streptococcus pneumoniae and Group B Streptococcus), 1 type of Gram negative cocci (Branhamella catarrhalis), 3 types of Gram positive bacilli (Clostridium perfringens, Corynebacterium and Mycobacterium) and 6 types of Gram negative bacilli (Pseudomonas aeruginosa, Campylobacter, Escherichia coli, Helicobacter pylori, Haemophilus influenzae and Klebsiella pneumoniae). Then, we detect their bacteria from Gram stained smear images, by annotating them and by applying the family of YOLOs, that is, YOLOv5, YOLOv7 and YOLOv8 as object detectors in deep learning to them. In particular, we investigate the effect of the variation of YOLOv8x with shifting convolution layers to detect bacteria.journal articl
Energy Detection Based Carrier Sense in Sub-GHz Band LPWANs and Its Characteristics
Get PDFIoT（Internet of Things）システムの無線通信インフラとして期待されている技術が，Sub-GHz帯（我が国では920 MHz帯）を用いるLPWAN（Low Power Wide Area Network）である．著者らはこれまでに，Sub-GHz帯LPWANの特性改善を目的として，キャリアセンス期間中の平均電力を用いる電力検出キャリアセンスの特性を理論的・実験的の両面から性能評価・解析してきた．その結果，ノイズフロア以下の受信信号についても検出可能であり，従来用いられていたキャリアセンス期間中のピーク電力を用いるピーク検出キャリアセンスと比較してLPWANの特性を大きく改善できることを示してきた．本稿では，著者らによるこれまでの解析結果を元に，電力検出キャリアセンスの特性やこれを用いたSub-GHz帯LPWANの特性，誤警報確率などの信号検出パラメータが電力検出キャリアセンスに与える影響など解説する．Sub-GHz band low-power wide-area networks (LPWANs) are expected as wireless communication infrastructure for Internet of Things (IoT) systems. Both theoretically and experimentally, we have analyzed and evaluated the characteristics of the energy detection based carrier sense that can achieve low carrier sense levels for the improvement in sub-GHz band LPWANs. As a result, it has been shown that energy detection for carrier sense detects interference packets with power below the noise floor and significantly improves the characteristics of sub-GHz band LPWANs compared to with conventional peak detection for carrier sense. In this paper, we describe the fundamental characteristics of the energy detection based-carrier sense in sub-GHz band LPWANs using our previous analysis results, which are characteristics of the energy detection based carrier sense, characteristics of sub-GHz band LPWANs with the energy detection based carrier sense, and effects of signal detection parameters, such as target false alarm probability, on the performance of sub-GHz band LPWANs with the energy detection based carrier sense.journal articl
Realizing Human-Robot Cooperative Rope-Spinning with Central Pattern Generator-Based Control Using Visual Information
Get PDFAchieving coordinated motion through flexible objects remains a significant challenge in Human-Robot Interaction (HRI). This study investigates a novel application of Central Pattern Generator (CPG) control, previously used in handshake robots, to a rope-spinning task involving human-robot cooperation. A real-time motion feedback system was developed using Azure Kinect, enabling a robot to synchronize its movements with human input by dynamically adjusting CPG outputs. We evaluated the system’s performance by varying rope lengths (250--400 cm) and analyzing spatial trajectories and Euclidean distances between the human and robot end-effectors. Results showed that while high coordination was achieved under shorter rope conditions, longer ropes introduced increased slack and tension variability, which reduced the robot's tracking stability. Frequency analysis also revealed weaker synchronization on the robot side, particularly in the vertical (Z) direction. These findings indicate that vision-based feedback alone is insufficient for robust adaptation to the dynamic characteristics of flexible objects. The vision-based method demonstrated lower amplitude fidelity and synchronization precision than our previous force-feedback approach. Future work will focus on integrating multimodal feedback, combining visual and force sensing, to improve coordination and robustness in flexible-object-mediated HRI.journal articl
A lightweight deep learning-free gesture recognition system by a local-connected reservoir computing model for service robots
Get PDFGesture recognition plays a crucial role in enabling intuitive and efficient human–robot interaction. Various deep learning-based methods have been recently proposed for implementing gesture recognition in robotic systems. However, the high computational cost of deep learning renders it less suitable for deployment in service robots with limited computational resources. This study aims to address this issue of computational cost by developing a lightweight, deep learning-free gesture recognition system based on reservoir computing (RC). Conventional RC models have the limitation that the spatial structure of the input data is often disrupted owing to random coupling during video processing. This study overcomes this limitation by proposing an RC model that incorporates spatial constraints into network coupling, thereby preserving the positional relationships of inputs during video processing. The experimental results reveal that the proposed method requires less processing time for both training and inference compared with an existing conventional RC-based method.conference pape
Prediction of QoL in healthy older adults using non-motor information from smart devices
Get PDFAlthough global interest in well-being and QoL is increasing, continuous awareness of one’s QoL in daily life remains challenging due to the need for repeated questionnaire responses. In this study, we evaluate the performance of a prediction model for QoL in healthy older adults using a Garmin Venu 3S fitness tracker and a FonLog data collection application to collect non-motor information and QoL data, and predict QoL using a support vector machine (SVM). The results of the prediction using a SVM showed that the Accuracy was approximately 0.96 and the F1-Score for each class was approximately 0.88 or higher. These results suggest the effectiveness of the QoL prediction model using nonmotor information. In the future, we plan to improve the processing and prediction in real time, and to evaluate the accessibility, usability, and effectiveness of the system for a wider range of users through experiments with non-motor subjects.journal articl
Fault-Tolerant and Terrain Adaptive Locomotion for a Four-Wheeled Planetary Exploration Rover with Active Suspension
Get PDF九州工業大学博士（工学）1 Introduction| 2 Rover Prototype Design| 3 Fault Tolerance| 4 Terrain Property Estimation| 5 Adaptive Resilient Locomotion| 6 ConclusionThis thesis presents the development and validation of a compact, resource-constrained planetary rover designed for lunar surface exploration and infrastructure construction. It focuses on achieving fault-tolerant, terrain-adaptive, and resilient locomotion using minimal hardware and intelligent actuator feedback processing. The proposed system integrates active suspension, one-way clutch mechanisms, real-time terrain estimation using resistive force theory (RFT), and realtime anomaly detection. Through a combination of experimental validation, simulation, and analytical modeling, the research provides a comprehensive framework for enabling autonomous, robust mobility under the challenging conditions of planetary terrains.
The Moon’s surface poses unique challenges to robotic mobility due to its loosely compacted regolith, harsh terrain features, and the absence of gravity. Traditional rover designs typically rely on passive suspensions and hardware redundancy to manage faults and irregular surfaces. However, such systems become impractical when strict constraints on mass, power, and volume are imposed, such as in infrastructure construction missions or pathfinder scouting operations. This work addresses this gap by proposing a novel rover system developed under the Japanese Moonshot Research and Development Program, Goal 3, which emphasizes open design principles, low resource utilization, and the use of integrated sensor feedback.
The rover platform developed in this study features a four-legged configuration, with one wheel at the end of each leg, resulting in a cross-shaped layout. Each wheel is actuated independently with three degrees of freedom: drive, steering, and suspension, for a total of 12 actuators. The modularity and symmetry of this system enable differential active suspension control and advanced locomotion modes. Each actuator is a HEBI X8-16 with built-in encoders and IMUs, which provide feedback on torque, velocity, current, position, and other parameters. The use of integrated feedback eliminates the need for external sensors, reducing system complexity and power consumption.
The first contribution of this thesis is the implementation of a fault-tolerant framework that combines passive mechanical resilience with active anomaly detection and reconfiguration. To address wheel actuator failure scenarios, a oneway clutch mechanism was integrated into each wheel. This passive system permits free forward rotation while restricting backward motion. In the event of actuator failure, the affected wheel transitions to a free rolling mode, conserving energy and preserving operational capability. This mechanism, while unidirectional, introduces no additional power demand, making it ideal for long-duration missions.
On the software side, fault detection is realized through an autoencoder-based anomaly detection system, trained using real actuator feedback data under both nominal and faulty conditions. The system monitors velocity, current, torque, and posture signals from all wheels and dynamically identifies anomalies in real time. A dynamic thresholding approach improves detection reliability by adapting to mission phases and terrain conditions. Experimental results show that the system can detect both terrain-induced and internal faults with high sensitivity, making it suitable for autonomous anomaly detection on future lunar rovers.
The second core contribution is the development of an in-situ terrain property estimation method using resistive force theory (RFT). RFT divides the contact surface of a wheel into elements and estimates the total resistive force by summing contributions from each component. By analyzing actuator feedback—torque, velocity, slip ratio—during controlled experiments, terrain parameters are derived in situ. The derived parameters enable real-time estimation of soil data, eliminating the need for vision or external instruments. This is especially advantageous in lunar settings where dust and lighting variations compromise optical sensors.
To validate this method, experiments were conducted using lunar regolith simulants. The feedback data were compared against RFT-based model outputs, showing a good correlation. This terrain estimation framework supports environment-aware decision-making, improving both mobility performance and safety margins during exploration tasks.
The third significant contribution lies in the development of adaptive locomotion strategies based on terrain feedback and mechanical configuration. A series of inching locomotion techniques were devised to handle high-sinkage or lowtraction environments. Unlike traditional forward-only inching, the proposed system supports omnidirectional inching by coordinating steering angles and suspension postures. This enables lateral movement and extraction maneuvers without requiring full-body rotation. Experiments demonstrated that the rover could maintain mobility even with partial actuator failures, and the theoretical
displacement per cycle closely matched experimental values, indicating robust kinematic predictability.
The rover’s active suspension system supports three height configurations—low, nominal, and high—which were leveraged to optimize ground clearance, stability, and traction. Differential active suspension control was evaluated by commanding asymmetric suspension postures and comparing mobility performance over uneven terrain. Results demonstrated that the rover could redistribute loads and dynamically adapt its body posture, reducing slip and enhancing obstacle negotiation. This capability, when integrated with the terrain estimation framework, provides a good technique for resilient mobility.
Collectively, the experimental campaigns across Chapters 2 to 4 confirm that the rover can autonomously detect faults, adapt its configuration to terrain properties, and execute resilient locomotion strategies even in failure cases. The use of minimal hardware—limited to actuators and standard computing modules such as the Jetson Orin Nano—ensures feasibility for actual space deployment where power and payload budgets are tight. The software architecture, built on ROS2, allows modular integration, real-time visualization, and teleoperation, while supporting future extensions to machine learning-based decision frameworks.
While the thesis makes significant strides in rover design and autonomy, some limitations remain. All experiments were conducted under Earth gravity using lunar regolith simulants. Although the simulants emulate particle size and cohesion, gravitational effects on sinkage and traction may vary in the actual lunar environment. Furthermore, the autoencoder anomaly detection system currently requires pre-training on representative data, which may limit adaptability to entirely new terrains or failure modes.
Future work should explore deployment in better lunar analog environments or participate in field campaigns. Integrating onboard machine learning, realtime terrain classification, and cooperative behavior among multiple rovers could also enhance exploration efficiency. Lightweight AI strategies that balance performance with embedded system constraints are a promising avenue to pursue. The mechanical configuration can be further optimized for navigating complex terrain.
In summary, this thesis presents a unified framework for fault-tolerant, terrainaware, and adaptive mobility in planetary rovers. Through a combination of mechanical simplicity, intelligent use of actuator feedback, and modeling, it demonstrates that high performance and resilience can be achieved even with limited resources. The outcomes contribute to the broader goal of enabling sustainable, autonomous, and low-cost robotic operations on the Moon, laying foundational principles for future lunar infrastructure development and Mars preparation missions.九州工業大学博士学位論文 学位記番号:工博甲第608号 学位授与年月日:令和7年9月25日令和7年度doctoral thesi