Human Computer Research Articles

Personalized interactive urban maps for autism: an innovative approach involving persons with Level 1 autism spectrum disorder to user-centered design

Purpose Because of social interaction difficulties, cognitive characteristics and sensory sensitivity, people with autism spectrum disorder (ASD) may struggle with independence and active participation in urban contexts. Through understanding how people with ASDs experience and perceive urban spaces, the Personalized Interactive Urban Maps for Autism (PIUMA) project’s purpose is to create a useful platform to support persons with ASD in their movements in the city, helping them in managing their daily lives, promoting their autonomy and active participation in urban contexts, taking in to account their direct point of view. This paper aim to present a user-centered approach involving autistic people. Design/methodology/approach The authors involved researchers with different backgrounds applying mixed research techniques such as semi-structured qualitative interviews, participatory design, exploiting user-adapted systems techniques, human–computer interaction techniques and interviews. The authors involved people with autism (PWA) from the beginning of the project to gain user requirements, as well as to assess the acceptability and effectiveness of our solution in the final stage. In specific phases of the project, the authors involved a group of parents of PWA and a group of neurotypical individuals as a control group. Findings The authors collected various data and created an App taking into account PWA needs, suggestions and points of view. Research limitations/implications Choosing different methodologies to allow participants to express their point of view, their needs and suggestions, the authors limited the quantitative data collected. The male-to-female ratio in the study, along with the sample size and participants’ age, can also be considered limitations. Practical implications The authors collected information about how autistic people move in urban contexts, represent the space of their cities, move independently and use technological devices. The data the author collected also provides suggestions about how to improve computer systems designed for them. Social implications The results suggest how to improve computer systems designed for autistic people and how to encourage their full participation in community life according to their cognitive and sensory characteristics. Originality/value The PIUMA project takes into account the direct input of PWA in all the different stages from conception and design to deployment of the device in a human-centered design approach.

Enhancing vector control: AI-based identification and counting of Aedes albopictus (Diptera: Culicidae) mosquito eggs

BackgroundDengue fever poses a significant global public health concern, necessitating the monitoring of Aedes mosquito population density. These mosquitoes serve as the disease vectors, making their surveillance crucial for dengue prevention. The objective of this study was to address the difficulty associated with identifying and counting mosquito eggs of wild strains during the monitoring of Aedes albopictus (Diptera: Culicidae) density via ovitraps in field surveys.MethodsWe constructed a dataset comprising 1729 images of Ae. albopictus mosquito eggs from wild strains and employed the Segment Anything Model to enhance the applicability of the detection model in complex environments. A two-stage Faster Region-based Convolutional Neural Network model was used to establish a detection model for Ae. albopictus mosquito eggs. The identification and counting process involved applying the tile overlapping method, while morphological filtering was employed to remove impurities. The model’s performance was evaluated in terms of precision, recall, and F1 score, and counting accuracy was assessed using R-squared and root mean square error (RMSE).ResultsThe experimental results revealed the model’s remarkable identification capabilities, achieving precision of 0.977, recall of 0.978, and an F1 score of 0.977. The R-squared value between the actual and identified egg counts was 0.997, with an RMSE of 1.742. The average detection time for a single tile was 0.48 s, which was more than 10 times as fast as the human–computer interaction method in counting an entire image.ConclusionsThe model demonstrated excellent performance in recognizing and counting Ae. albopictus mosquito eggs, indicating great application potential. This study offers novel technological support for enhancing vector control effectiveness and public health standards.Graphical

Computer Vision-Based Gait Recognition on the Edge: A Survey on Feature Representations, Models, and Architectures

Computer vision-based gait recognition (CVGR) is a technology that has gained considerable attention in recent years due to its non-invasive, unobtrusive, and difficult-to-conceal nature. Beyond its applications in biometrics, CVGR holds significant potential for healthcare and human–computer interaction. Current CVGR systems often transmit collected data to a cloud server for machine learning-based gait pattern recognition. While effective, this cloud-centric approach can result in increased system response times. Alternatively, the emerging paradigm of edge computing, which involves moving computational processes to local devices, offers the potential to reduce latency, enable real-time surveillance, and eliminate reliance on internet connectivity. Furthermore, recent advancements in low-cost, compact microcomputers capable of handling complex inference tasks (e.g., Jetson Nano Orin, Jetson Xavier NX, and Khadas VIM4) have created exciting opportunities for deploying CVGR systems at the edge. This paper reports the state of the art in gait data acquisition modalities, feature representations, models, and architectures for CVGR systems suitable for edge computing. Additionally, this paper addresses the general limitations and highlights new avenues for future research in the promising intersection of CVGR and edge computing.

A Meta-Analysis of the Effects of Interaction on Value Co-Creation in Online Collaborative Innovation Communities Based on the Service Ecosystem Framework

Interaction is typically at the core of the value co-creation process through operant resource exchange in online collaborative innovation communities (OCICs). While some studies emphasize the facilitating effect of interaction on value co-creation, others have drawn opposite conclusions, such as more peer interaction leads to less idea generation. Thus, the purpose of this paper is to utilize the service ecosystem framework to clarify the overall relationship between interaction and value co-creation and to explore the moderating factors that may have contributed to the divergence and inconsistency of previous studies. We conducted a meta-analysis of 65 effect sizes obtained from 63 articles with a cumulative sample size of 25,185 between 2004 and 2023, using a random effects model. The results indicate that interaction has a significantly positive impact on user value co-creation within OCICs (r = 0.453, 95%CI [0.405, 0.499]), and the heterogeneity among studies was significant (Q = 1409.29, p < 0.001). The strength of this correlation was moderated by the types of interaction (human–computer or human–human interactions), the types of OCICs (business-sponsored or socially constructed online communities), and the number of involved OCICs (one or multiple online communities), but not by the cultural background. These findings support the service ecosystem perspective rather than resource scarcity theory by resolving the mixed findings regarding the relationship between interaction and user value co-creation. Furthermore, this study systematically examined the contingent factors separately across three levels, micro (types of actor interactions), meso (types and number of OCICs), and macro (cultural background), combining the whole and the part insights, and empirically integrating service ecosystems as the foundational paradigm and unit of analysis for value co-creation research for the first time. This research contributes to theoretical frameworks in service ecosystems and offers actionable insights for management practices in business and marketing.

Automated landmark-based cat facial analysis and its applications

Facial landmarks, widely studied in human affective computing, are beginning to gain interest in the animal domain. Specifically, landmark-based geometric morphometric methods have been used to objectively assess facial expressions in cats, focusing on pain recognition and the impact of breed-specific morphology on facial signaling. These methods employed a 48-landmark scheme grounded in cat facial anatomy. Manually annotating these landmarks, however, is a labor-intensive process, deeming it impractical for generating sufficiently large amounts of data for machine learning purposes and for use in applied real-time contexts with cats. Our previous work introduced an AI pipeline for automated landmark detection, which showed good performance in standard machine learning metrics. Nonetheless, the effectiveness of fully automated, end-to-end landmark-based systems for practical cat facial analysis tasks remained underexplored. In this paper we develop AI pipelines for three benchmark tasks using two previously collected datasets of cat faces. The tasks include automated cat breed recognition, cephalic type recognition and pain recognition. Our fully automated end-to-end pipelines reached accuracy of 75% and 66% in cephalic type and pain recognition respectively, suggesting that landmark-based approaches hold promise for automated pain assessment and morphological explorations.

Room‐Temperature‐Integrated Flexible Diffusive Oxide Memristors for Artificial Nociceptive Systems

AbstractNovel areas of human–computer interaction and intelligent robotics are rapidly boosted by flexible electronic devices, among which artificial nociceptors for pain perception are considered essential sensory devices. Recently, flexible diffusive memristors with volatile behavior are extensively studied to emulate biological synapses and neurons because of their rich threshold–switching dynamics, which are crucial for nociceptive functions. Herein, flexible diffusive memristors with an Ag/TaO1.8/TaO2.7/TaO1.8/Ag (ATTTA) structure are fabricated at room temperature, demonstrating a high switching ratio of 107, an ultra‐fast turn‐on speed of 40 ns, a long endurance up to 104 sweep cycles, low positive and negative threshold variation rates (6.7%/7.8%) for 50 devices, a good thermal stability at an 85 °C measurement temperature, and a high flexibility under an 8 mm bending radius. These high performances are attributed to the oxygen‐rich TaO2.7 layer, which can prevent the excessive migration of Ag atoms. Moreover, multiple biological nociceptor functions, including threshold firing, no adaptation, and sensitization, are successfully emulated in the flexible ATTTA diffusive memristor. Accordingly, the human–skin nociceptive system is successfully emulated by integrating ATTTA diffusive memristor array with force–sensing resistors (FSRs), which enables stereoscopic injury perception of inhomogeneous stimuli.

Drosophila Visual System Inspired Ambipolar OFET for Motion Detection

AbstractDrosophila can rapidly and precisely detect changes in light in their surroundings and achieve acute perception of motion information with high energy efficiency and adaptivity owing to the cooperation of “ON” channel and the “OFF” channel in its visual system. Optical controlled bidirectional synaptic behavior of neuromorphic device is important for modeling parallel processing channels of Drosophila's visual system. In this study, an ambipolar transistor utilizing a bilayer architecture composed of p‐type pentacene and n‐type C60 as semiconductors is developed, with near‐infrared (NIR) PbS quantum dots serving as the charge‐trapping layer. This design enables a gate‐tunable positive and negative photoresponse, driven by photogating and photovoltaic effects at visible and NIR wavelengths. When regulated by a negative gate voltage, the device exhibits a suppressed photocurrent relaxation time exceeding 1000 s, demonstrating stable long‐term inhibitory characteristics. Consequently, high‐contrast excitatory and inhibitory synapses facilitate orientation and motion detection. Identification accuracies of up to 94.8% for motion direction and 98.1% for dynamic gestures are achieved. Practical applications such as intelligent monitoring and human–computer interaction stand to benefit significantly from these findings.

Examining the use of multiple cognitive load measures in evaluating online shopping convenience: an EEG study

PurposeIn the past decade, the use of neurophysiological measures as a complementary source of information has contributed to our understanding of human–computer interaction. However, less attention has been given to their capability in providing measures with high temporal resolution. Two studies are designed to address the challenge of measuring users’ cognitive load in an online shopping environment and investigate how it is related to task difficulty, task uncertainty and shopping convenience.Design/methodology/approachTwo experiments using behavioral and neurophysiological measures are conducted to investigate how various types of the cognitive load construct can be measured and used in an online shopping context.FindingsResults of the first study suggest that although all cognitive load measures are influenced by task difficulty, only accumulated load (i.e. total cognitive load experienced during a task) is sensitive to task uncertainty. Results of the second study show that convenience negatively influences accumulated load, and the latter negatively influences user satisfaction.Practical implicationsOur research offers practical value by providing designers with a validated method to measure users’ cognitive load, enabling the identification of usability issues and design improvement.Originality/valueThis study contributes to the literature by developing a rich and temporally high-resolution measurement of the cognitive load construct and examining how it can inform us about users’ cognitive state in an online shopping environment.

Developing Innovative Feature Extraction Techniques from the Emotion Recognition Field on Motor Imagery Using Brain–Computer Interface EEG Signals

Research on brain–computer interfaces (BCIs) advances the way scientists understand how the human brain functions. The BCI system, which is based on the use of electroencephalography (EEG) signals to detect motor imagery (MI) tasks, enables opportunities for various applications in stroke rehabilitation, neuroprosthetic devices, and communication tools. BCIs can also be used in emotion recognition (ER) research to depict the sophistication of human emotions by improving mental health monitoring, human–computer interactions, and neuromarketing. To address the low accuracy of MI-BCI, which is a key issue faced by researchers, this study employs a new approach that has been proven to have the potential to enhance motor imagery classification accuracy. The basic idea behind the approach is to apply feature extraction methods from the field of emotion recognition to the field of motor imagery. Six feature sets and four classifiers were explored using four MI classes (left and right hands, both feet, and tongue) from the BCI Competition IV 2a dataset. Statistical, wavelet analysis, Hjorth parameters, higher-order spectra, fractal dimensions (Katz, Higuchi, and Petrosian), and a five-dimensional combination of all five feature sets were implemented. GSVM, CART, LinearSVM, and SVM with polynomial kernel classifiers were considered. Our findings show that 3D fractal dimensions predominantly outperform all other feature sets, specifically during LinearSVM classification, accomplishing nearly 79.1% mean accuracy, superior to the state-of-the-art results obtained from the referenced MI paper, where CSP reached 73.7% and Riemannian methods reached 75.5%. It even performs as well as the latest TWSB method, which also reached approximately 79.1%. These outcomes emphasize that the new hybrid approach in the motor imagery/emotion recognition field improves classification accuracy when applied to motor imagery EEG signals, thus enhancing MI-BCI performance.

An Overview of the Empirical Evaluation of Explainable AI (XAI): A Comprehensive Guideline for User-Centered Evaluation in XAI

Recent advances in technology have propelled Artificial Intelligence (AI) into a crucial role in everyday life, enhancing human performance through sophisticated models and algorithms. However, the focus on predictive accuracy has often resulted in opaque black-box models that lack transparency in decision-making. To address this issue, significant efforts have been made to develop explainable AI (XAI) systems that make outcomes comprehensible to users. Various approaches, including new concepts, models, and user interfaces, aim to improve explainability, build user trust, enhance satisfaction, and increase task performance. Evaluation research has emerged to define and measure the quality of these explanations, differentiating between formal evaluation methods and empirical approaches that utilize techniques from psychology and human–computer interaction. Despite the importance of empirical studies, evaluations remain underutilized, with literature reviews indicating a lack of rigorous evaluations from the user perspective. This review aims to guide researchers and practitioners in conducting effective empirical user-centered evaluations by analyzing several studies; categorizing their objectives, scope, and evaluation metrics; and offering an orientation map for research design and metric measurement.

Enabling Perspective-Aware Ai with Contextual Scene Graph Generation

This paper advances contextual image understanding within perspective-aware Ai (PAi), an emerging paradigm in human–computer interaction that enables users to perceive and interact through each other’s perspectives. While PAi relies on multimodal data—such as text, audio, and images—challenges in data collection, alignment, and privacy have led us to focus on enabling the contextual understanding of images. To achieve this, we developed perspective-aware scene graph generation with LLM post-processing (PASGG-LM). This framework extends traditional scene graph generation (SGG) by incorporating large language models (LLMs) to enhance contextual understanding. PASGG-LM integrates classical scene graph outputs with LLM post-processing to infer richer contextual information, such as emotions, activities, and social contexts. To test PASGG-LM, we introduce the context-aware scene graph generation task, where the goal is to generate a context-aware situation graph describing the input image. We evaluated PASGG-LM pipelines using state-of-the-art SGG models, including Motifs, Motifs-TDE, and RelTR, and showed that fine-tuning LLMs, particularly GPT-4o-mini and Llama-3.1-8B, improves performance in terms of R@K, mR@K, and mAP. Our method is capable of generating scene graphs that capture complex contextual aspects, advancing human–machine interaction by enhancing the representation of diverse perspectives. Future directions include refining contextual scene graph models and expanding multi-modal data integration for PAi applications in domains such as healthcare, education, and social robotics.

PortraitEmotion3D: A Novel Dataset and 3D Emotion Estimation Method for Artistic Portraiture Analysis

Facial Expression Recognition (FER) has been widely explored in realistic settings; however, its application to artistic portraiture presents unique challenges due to the stylistic interpretations of artists and the complex interplay of emotions conveyed by both the artist and the subject. This study addresses these challenges through three key contributions. First, we introduce the PortraitEmotion3D (PE3D) dataset, designed explicitly for FER tasks in artistic portraits. This dataset provides a robust foundation for advancing emotion recognition in visual art. Second, we propose an innovative 3D emotion estimation method that leverages three-dimensional labeling to capture the nuanced emotional spectrum depicted in artistic works. This approach surpasses traditional two-dimensional methods by enabling a more comprehensive understanding of the subtle and layered emotions often in artistic representations. Third, we enhance the feature learning phase by integrating a self-attention module, significantly improving facial feature representation and emotion recognition accuracy in artistic portraits. This advancement addresses this domain’s stylistic variations and complexity, setting a new benchmark for FER in artistic works. Evaluation of the PE3D dataset demonstrates our method’s high accuracy and robustness compared to existing state-of-the-art FER techniques. The integration of our module yields an average accuracy improvement of over 1% in recent FER systems. Additionally, combining our method with ESR-9 achieves a comparable accuracy of 88.3% on the FER+ dataset, demonstrating its generalizability to other FER benchmarks. This research deepens our understanding of emotional expression in art and facilitates potential applications in diverse fields, including human–computer interaction, security, healthcare diagnostics, and the entertainment industry.

Icon similarity model based on cognition and deep learning

Human-computer cooperation guided by natural interaction, intelligent interaction, and human–computer integration is gradually becoming a new trend in human–computer interfaces. An icon is an indispensable pictographic symbol in an interface that can convey pivotal semantics between humans and computers. Research on similar icons’ cognition in humans and the discrimination of computers can reduce misunderstandings and facilitate transparent cooperation. Therefore, this research focuses on images of icons, extracted contours, and four features, including the curvature, proportion, orientation, and line of the contour, step by step. By manipulating the feature value change to obtain 360 similar icons, a cognitive experiment was conducted with 25 participants to explore the boundary values of the feature dimensions that cause different levels of similarity. Its boundary values were applied to deep learning to train a discrimination algorithm model that included 1500 similar icons. This dataset was used to train a Siamese neural network using a 16-layer network branch of a visual geometry group. The training process used stochastic gradient descent. This method of combining human cognition and deep learning technology is meaningful for establishing a consensus on icon semantics, including content and emotions, by outputting similarity levels and values. Taking icon similarity discrimination as an example, this study explored the analysis and simulation methods of computer vision for human visual cognition. The accuracy evaluated is 90.82%. The precision was evaluated as 90% for high, 80.65% for medium, and 97.30% for low. Recall was evaluated as 100% for high, 89.29% for medium, and 83.72% for low. It has been verified that it can compensate for fuzzy cognition in humans and enable computers to cooperate efficiently.

CCW-YOLO: A Modified YOLOv5s Network for Pedestrian Detection in Complex Traffic Scenes

In traffic scenes, pedestrian target detection faces significant issues of misdetection and omission due to factors such as crowd density and obstacle occlusion. To address these challenges and enhance detection accuracy, we propose an improved CCW-YOLO algorithm. The algorithm first introduces a lightweight convolutional layer using GhostConv and incorporates an enhanced C2f module to improve the network’s detection performance. Additionally, it integrates the Coordinate Attention module to better capture key points of the targets. Next, the bounding box loss function CIoU loss at the output of YOLOv5 is replaced with WiseIoU loss to enhance adaptability to various detection scenarios, thereby further improving accuracy. Finally, we develop a pedestrian count detection system using PyQt5 to enhance human–computer interaction. Experimental results on the INRIA public dataset showed that our algorithm achieved a detection accuracy of 98.4%, representing a 10.1% improvement over the original YOLOv5s algorithm. This advancement significantly enhances the detection of small objects in images and effectively addresses misdetection and omission issues in complex environments. These findings have important practical implications for ensuring traffic safety and optimizing traffic flow.

Embodied participation: Dance knowledge informing workplace training

Research into physical simulation practices across industries (aviation, transport and healthcare), provides insights into embodied knowledge. Embodied knowledge informed by dance knowledge can positively expand the capacity of workplace training. Through a detailed examination of a variety of technologically advanced industry environments where embodied knowledge is applied, this article identifies the ways in which dance knowledge can be transformed into new and enhanced forms of embodied participation. The question of how dance knowledge is transferred across industries is addressed through theories of embodied cognition paired with emergent communication strategies from human–computer interaction and human factors. Key insights extend concepts of somatic practice across arts and industry for mutual benefit.

Low light recognition of traffic police gestures based on lightweight extraction of skeleton features

With the increasing demand for information interaction between intelligent vehicles and traffic managers, understanding traffic police commands has become an essential element in achieving human–computer interaction. The aim of this study is to investigate the low light traffic police gesture recognition based on lightweight extraction of skeleton features. A two-stage recognition framework for eight gesture commands is proposed. The first stage utilizes convolutional neural networks to acquire the human skeleton under low light. The second stage utilizes a recurrent neural network to construct a lightweight feature extraction model for gesture recognition. HRnet acquires spatial features of the action, high-resolution and multi-scale fusion circumvents the effect of low-light intensity. Self-attention (SA) mechanism and MobileNet lightweight feature extraction have addressed the time-consuming issue in the feature extraction phase. The improved GRU network is used to extract time domain features, the fusion of spatio-temporal features improves the recognition accuracy. For training and validation, a gesture dataset (TPGD) containing 3391 gesture instances is constructed. The proposed method obtains 94.75% accuracy in the TPGD test set, demonstrating its effectiveness. This study holds significant importance for enhancing the level of intelligence in traffic management, promoting the development of intelligent transportation systems, and ensuring road traffic safety.

Ultra‐Sensitive and Broad‐Range Iontronic Pressure Sensor Based on ZnO‐Reinforced Sodium Alginate With CaCl2

AbstractFlexible pressure sensors are unprecedentedly developed on monitoring human physical activities and human–computer interaction. Simultaneously, improving the sensitivity and pressure range of flexible pressure sensors is a great challenge. Despite the introduction of microstructures on the dielectric layer, the deformability is still limited due to the size constraints of the microstructures, and the sensitivity of iontronic capacitive pressure sensor (ICPS) is still difficult to significantly improve. Here, an innovative strategy of combining nano‐ZnO and sodium alginate (SA) ionogels with two‐stage enhancement (TSE) structures, which can significantly enhance the toughness and sensitivity of ICPS for ultrasensitive pressure range, is proposed. Compared with other previously reported pressure sensors, the sensor developed in this work exhibits an ultra‐high sensitivity (Smax>17000 kPa−1, Smin>360 kPa−1) in a broad sensing range of more than 900 kPa and maintains stable long‐term working durability for 4000 cycles. The sensor demonstrates its versatility potential in practical applications in physical activity monitoring and human–computer interaction. This work provides a general strategy to achieve an ICPS with broader pressure‐response range and higher sensitivity, offering a stable and effective way for a low‐cost, scalable pressure sensor.

Development of highly sensitive and stable patterned PDMS flexible strain sensors for motion monitoring via laser direct writing

Flexible strain sensors have received widespread attention for their potential applications in wearables, human–computer interaction, and healthcare. However, achieving a good balance between sensitivity, stretchability, and stability remains challenging. Here, we report a cost-effective and scalable fabrication strategy that combines laser direct writing (LDW) with 3D printing (3DP) to prepare various patterned Polydimethylsiloxane (P-PDMS) flexible strain sensors. By varying the laser parameters and processing paths, different microstructured patterns can be obtained, which significantly influence the sensor’s performance. By introducing patterned composite microstructures, the sensitivity of the strain sensor was increased by 339 % compared to the sensor without surface structures. Additionally, the strain sensors exhibit high stability and durability, a fast response time (140 ms), low hysteresis (0.009), and an ultra-low detection limit (0.0125 % strain). Besides, the sensors demonstrate excellent electrical performance and thermal stability. Based on their superior performance, we demonstrated their capability for real-time monitoring of human physiological signals. These findings successfully illustrate the potential of laser processing in fabricating complex microstructures, enabling the development of high-sensitivity flexible strain sensors for applications such as wearable health monitoring and human–computer interaction.

Dancing robots: aesthetic engagement is shaped by stimulus and knowledge cues to human animacy.

Artificial intelligence (AI) and robots are increasingly shaping the aesthetic preferences of art consumers, influencing how they perceive and engage with artistic works. This development raises various questions: do cues to the humanness of the origin of an artwork or artist influence our aesthetic preferences?. Across two experiments, we investigated how the perception and appreciation of dance is influenced by cues to human animacy. We manipulated Agent Form (human-like or robot-like dancer), Belief about Movement Source (human motion capture or computer animation), Source of Choreography (human- or computer-generated), and Belief about Choreography Source (believed to be human- or computer-generated). Results pointed toward agent congruence: In Experiment 1, robot agents were preferred when the movement source was believed to be computer animation. In Experiment 2, robot agents were preferred when the choreography was believed to be computer-generated, while choreographies believed to be human-generated were generally preferred. Participants could not accurately identify the actual source of choreography. These results persisted beyond the effects of age, dance expertise, technological expertise, attitudes toward AI, and perceived familiarity, complexity, evocativeness, technical competence, or reproducibility of the dance. Dance expertise, technological expertise, and attitudes toward AI independently impacted aesthetic judgments. These findings provide insights into the design of robotic dance, highlighting features of dance choreography and audience characteristics that influence aesthetic engagement. To enhance AI-driven creative productions, shaping perceptions will be crucial for better audience reception and engagement.

Design and application of a new high-performance flexible six-axis force/torque sensor for massage therapy

Precise sensing of force magnitude and direction in massage therapy can significantly enhance treatment effectiveness. Despite advancements in flexible multidimensional force sensors, achieving comprehensive spatial force sensing with soft materials remains challenging. This study presents a novel flexible six-axis force/torque sensor, calibrated using a deep neural network. The calibrated sensor exhibits a maximum class I error of 0.603%F.S. and a maximum class II error of 0.751%F.S., indicating excellent measurement performance. Demonstrated in massage physiotherapy, the sensor effectively distinguishes between various techniques and accurately detects subtle force variations within the same technique. The present work introduces a novel strategy for designing flexible six-axis force/torque sensors, thereby laying the groundwork for advancements in intelligent robotics, human–computer interfaces, as well as in the fields of rehabilitation and medicine.