Efficiently localizing the source of odor in complex environments remains a critical challenge, especially in environments with obstacles or walls. We study diffusion state perception as an unsupervised classification problem that can inform downstream search behaviors. We built a dodecahedral gas-sensing device equipped with 11 ethanol sensors and introduced a novel diffusion state classification system designed to capture the odor diffusion state via multi-directional gas measurements. The proposed system employs a manifold learning method known as Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction and K-means++ for clustering, enabling unsupervised classification of odor diffusion states. Experimental results demonstrated that the proposed method achieves clearer diffusion state classification compared to conventional two-directional sensor systems. Furthermore, the system was tested using data from unseen environments, suggesting its generalization capability. These findings are expected to contribute to the development of adaptive search algorithms and enhance the efficiency of odor source localization.
This study proposes and evaluates a novel, proxy method for assessing the performance of back support exoskeletons (BSEs), to address some limitations of electromyography (EMG) and metabolic cost (MC) measurements in workplace and field environments. We introduce the use of interface force between the user and exoskeleton as an alternative evaluation metric. To enable this, a compact load cell was integrated into the exoskeleton's thigh cuff, providing reliable force measurements without noticeably altering the system's kinematics or inertia. While interface force measurements can be affected by sweat and other human-related factors like EMG and MC, interface force can still represent the assistive force, allowing for real-time assessment of exoskeleton assistance in both laboratory and real-world settings. Experimental results demonstrated a statistically significant, strong correlation between mean interface force and lower-back muscle EMG activity during the upright phase of a squat lifting task with load. Whereas for MC, though not statistically significant, there was a strong correlation between mean interface force and MC reduction during a repetitive squat lifting task. These findings indicate that interface force sensing offers a promising proxy alternative to traditional EMG and MC evaluations, enabling more feasible, field-ready assessment of BSE performance in real-world workplace environments.
As social robots increasingly engage in collaborative activities, they must express intentions through interpretable nonverbal cues. In human-robot interaction, intention inference may involve both identifying the target directly communicated by a robot and inferring the robot's underlying intention behind that communicative behavior. This study examined how humans infer a robot's intention when gaze direction and torso orientation are inconsistent. In a collaborative decoration task, participants interacted with a social robot under one of four experimental groups based on three gaze-torso consistency conditions: consistency, head-gaze inconsistency, and eye-gaze inconsistency. Results suggested that target inference remained relatively stable across conditions, whereas inference of underlying intention tended to be higher under inconsistency, especially when the mismatch was produced through eye-gaze. At the session level, eye-gaze inconsistency also increased negative mental state attribution without reducing social acceptance or likeability. These findings support a two-layer view of robot intention inference and suggest that eye-gaze and head-gaze play different roles in robot intention expression.
MRI-guided needle insertion is considered as a radiation-free, minimally invasive treatment. However, constraints such as the confined space within the gantry and strong magnetic field environment necessitate the use of robotic technology to assist the operator. Many existing MRI-compatible robots face challenges of increased size and complexity owing to the need for multiple degree-of-freedom (DOF). Therefore, we developed a compact 5-DOF manipulator system using a spherical gear-based pneumatic motor. This motor generates pitch and yaw DOF within a single module, thereby enabling a simple joint structure without a rotational center offset. This paper details the manipulator's mechanical design and evaluates its puncture accuracy using phantoms, as well as its compatibility within a 1.5-T MRI environment. The experimental results showed no reduction in the image SNR owing to robotic actuation, achieving an average error of 4.5 mm with this prototype. This accuracy is sufficient for accessing clinically significant lesions, thereby demonstrating the effectiveness of the proposed method. Furthermore, the system was confirmed to operate normally even under remote actuation conditions via a 6.5-m-long pneumatic transmission line. These results demonstrate the fundamental utility of the proposed non-magnetic pneumatic drive mechanism, suggesting the potential for future application in MRI-guided punctures.
Environmental perception systems for mobile robots face significant challenges in dynamic and weak-texture indoor environments. Inspired by biological vision's "active observation," this paper presents a perception method using a bionic eye. By simulating human fixation behavior through its six-degree-of-freedom gaze control, the method enables active acquisition of key environmental information. The contributions are: (1) a pedestrian motion perception and dynamic interference elimination method based on YOLOv8-Pose; (2) a real-time texture quality evaluation model combining LBP analysis and feature point stability; (3) a gaze point optimization model integrating texture distribution and pedestrian motion trends, guiding the bionic eye towards high-value areas. Experimental validation within a visual SLAM system shows that, compared to fixed-view ORB-SLAM2, our method reduces absolute trajectory error by 27.5% in dynamic scenes and 48.8% in low-texture corridors, while effectively preventing feature loss and tracking failure. This work offers new insights for achieving human-like active perception in mobile robots.
Based on vision and prior experience, humans can make rough physical predictions and adjust their manipulation strategies. This paper aims to endow robots with a similar ability. To collect paired data of vision and forces, we use a rigid-body simulator commonly adopted in robotics. However, unlike simulators that output noisy point forces, humans are able to make consistent predictions even in unfamiliar situations. Based on this observation, we hypothesize that predicting smooth force distributions rather than raw point forces can improve both force prediction itself and downstream task performance. To validate this hypothesis, we construct a model that predicts three-dimensional force distributions from a single RGB image of piled daily objects. The target distribution is generated by applying statistical smoothing to point forces obtained from the simulator. Moreover, by incorporating object geometry into the smoothing process, we aim to account for variations in contact states and achieve more consistent vision-based predictions. We conduct extensive evaluations in both simulation and real environments. Results show that our approach improves prediction accuracy, enhances downstream task performance through smoothing, and further benefits from geometry-guided smoothing. Remarkably, the trained model generalizes effectively to real-world scenes despite being trained solely in simulation.
In this work, we investigate swarm self-clustering, where robots autonomously organize into spatially coherent groups using only local sensing and decision-making, without relying on external commands, global positioning, or inter-robot communication. This fully decentralized approach enables each robot to form and maintain stable clusters by responding solely to distances from nearby neighbors, as detected through onboard range sensors with limited fields of view. Motivated by real-world scenarios such as autonomous underwater robot retrieval and human gathering during emergency response, the proposed method is designed for GPS-denied and communication-constrained environments. Unlike conventional approaches, it requires no prior knowledge of cluster parameters such as size, number, or member identity, making it highly adaptable to unpredictable settings. A mechanism that enables each robot to adaptively alternate between consensus-based and random goal assignment based on local neighborhood size, ensuring robust, scalable, and untraceable clustering independent of initial configuration. Through extensive simulations and real-robot experiments, we demonstrate the method's scalability, empirical convergence, and robustness under varying initial conditions and dynamic robot additions. The proposed communication-free approach outperforms local-only baselines across standard cluster quality metrics, producing clusters that exhibit untraceability even under identical initial conditions.
This paper presents a novel exosuit designed to augment elbow capabilities using a hybrid actuator that integrates chain mail fabrics with a fiber-reinforced actuator. The actuator provides a bending moment to support elbow movement, while the chain mail enables tunable stiffness through a self-stiffening approach. Unlike conventional jamming-based exosuits, ours does not rely on an external power source to induce the jamming transition, making the exosuit lighter, more efficient, and more comfortable than traditional designs. Rotational stiffness experiments were performed to evaluate the mechanical characteristics and validate the jamming transition within the proposed actuator. Experimental results (at 120 kPa) showed that the hybrid actuator achieved a maximum increase in rotational stiffness of a factor of 13 compared to the fiber-reinforced actuator alone. To define the design space, additional experiments were conducted to evaluate the bending deformation angle, radius of curvature, and output force. Further tests were performed to evaluate the efficacy of the exosuit in enhancing human performance during load-lifting activity. The results demonstrated the ability of the exosuit to support elbow function effectively.
This paper presents a novel method for accelerating path planning tasks in unknown scenes with obstacles by utilizing Wasserstein Generative Adversarial Networks (WGANs) with Gradient Penalty (GP) to learn the distribution of the collision-free configuration space under given conditions. Our proposed approach involves conditioning the WGAN-GP with a Variational Auto-Encoder in a continuous latent space to handle multimodal datasets. However, training a Variational Auto-Encoder with WGAN-GP can be challenging for image-to-configuration-space problems, as the Kullback-Leibler loss function often converges to a random distribution. To overcome this issue, we simplify the configuration space as a set of Gaussian distributions and divide the dataset into several local models. This enables us to not only learn the model but also speed up its convergence. Our experiments show promising results for accelerating path planning tasks in unknown scenes while generating quasi-optimal paths with our WGAN-GP. The source code is openly available (https://bitbucket.org/joro3001/multiwgangp/).
We propose an occlusion-aware framework for human pose estimation based on temporal point-cloud sequences. Training data are generated via simulation and augmented with synthetic occlusions using Perlin-noise masks. The network combines PointNet++ for spatial features extraction, a Transformer for temporal encoding, and a graph convolutional network with inverse DCT for skeletal reconstruction. We evaluate the method against an RGB-based baseline (MediaPipe) under real robot-induced occlusions using 128 annotated frames of right-hand reaching. The proposed method achieves significantly lower errors than the baseline at the shoulder and elbow. An ablation study shows that occlusion augmentation significantly improves performance under occlusion. Visibility analysis further indicates that, after multiple-comparison correction, error-visibility correlations remain for the baseline but not for the proposed method, suggesting reduced sensitivity to occlusion. These results demonstrate the potential of simulation-to-real training for robust single-sensor pose estimation in assistive robotics.
Emergent communication in multi-agent systems poses scalability challenges, as conventional models are typically limited to dyadic interactions. This study introduces the Variational Bayes Naming Game (VBNG), a decentralized Bayesian inference framework for modeling symbol emergence among three or more agents. Based on the Collective Predictive Coding (CPC) hypothesis, VBNG enables agents to infer shared signs by minimizing free energy using their individual observations. In our experiment, we evaluated the proposed method against a conventional sampling-based approach (RMHNG) and a centralized topline model (VB) using both synthetic and patched MNIST image datasets. VBNG achieved accuracy comparable to the topline model. Although the conventional method failed under conditions where individual agents' observations were insufficient for identification, VBNG demonstrated high robustness (ARI > 0.6). The proposed method demonstrated significantly improved computational efficiency, converging 2-4 times faster than the conventional method in experiments with up to 10 agents. This performance was achieved while attaining high classification accuracy and high inter-agent agreement (kappa = 0.94) on the MNIST task. This study presents a robust, scalable, and theoretically coherent framework for symbol emergence research, demonstrating variational inference as an effective mechanism for consensus formation in decentralized multi-agent systems.
As social robots enter everyday environments as ubiquitous companions, they provide opportunities such as personalized support and enhanced safety, but also introduce challenges for heightened concerns over privacy and surveillance. Addressing these socio-technical factors requires empirical evidence on how technical features of robots elicit psychological responses. This study investigates the influence of robot eye features on public self-awareness in observation-only contexts through a between-participants experiment (N = 650). A follow-up study (N = 201) examined the moderating role of individual differences in social anxiety. Results show that visually complex eyes and blinking behavior indirectly increase public self-awareness through enhanced perceived agency. Moreover, under high-agency conditions, participants with high social anxiety reported significantly greater public self-awareness than those with low social anxiety. Based on the findings, we propose IMROSF, a structural equation model that explains the effect of the appearance of peripheral robots on public self-awareness. These findings provides empirical evidence on the effects of robot eye design and underscores the social implications of treating robotic eyes as adaptive interfaces for managing users' awareness of being evaluated.
With the rapid aging of the global population, there is an increasing need for innovative solutions to mitigate the physical burden on caregivers and improve safety for older people in medical and nursing-care settings. Falls represent a major threat to older people, significantly impacting their mobility and functional independence. In the present study, we introduced and evaluated a suspension-type fall-impact-mitigation robot designed to reduce fall speed and injury risk in daily living environments. The robot, which operates without the need for an external power source, was tested in a living lab simulating indoor and outdoor home spaces. High-performance, markerless motion-capture technology was employed to measure the velocity of falls across five scenarios, including forward/backward falls, sliding from a chair, and tripping on slopes. The results showed that wearing the suspension-type fall-impact-mitigation robot substantially reduced the velocity of falls, with hip velocities upon contact with the ground remaining below the biomechanical reference threshold of 2.0 m/s in 98.9% of measurements across all scenarios. A usability survey conducted among healthcare professionals further indicated high scores for safety and ease of use in care settings. These findings suggest that the fall-impact-mitigation robot is a promising solution for reducing fall-related injuries among older adults.
