In recent years, with the rapid advancement of space development, there has been increasing demand for constructing solar power generation systems on the Moon using modular robots. This study has been proposing a modular construction method aimed at building tall structures on the lunar surface. Their modules adopt a truncated octahedron shape one of a space-filling polyhedron allowing their configuration and placement to be determined geometrically by exploiting its space-filling property. This approach can also be treated as a space quantization and establishes a unified platform in which module development, configuration, and positioning can be derived with comparatively simple calculations. In this paper, we present the results of a field experiment using newly prototyped modules, along with the previously proposed construction method and module design. The results demonstrate the feasibility of the proposed method and highlight the potential for space quantization for robot motions based on space-filling properties.
In self-supervised robot learning, data is acquired through active interaction with the environment, which is costly. Therefore, sample-efficient exploration is vital. To this end, intrinsic motivation (IM) methods such as learning progress (LP) have been adopted in robot learning, with variable success across tasks; whereas in machine learning, active learning (AL) is considered the go-to solution, especially for classification tasks. However, there is no systematic method for fusing both approaches for continuous regression tasks encountered in robot learning. To this end, we propose MUSEL (Model Uncertainty for Sample-Efficient Learning), a novel AL framework tailored for regression tasks in robotics, such as action-effect prediction. MUSEL introduces a novel model uncertainty metric that combines total predictive uncertainty, learning progress, and input diversity to guide experience gathering. We choose Stochastic Variational Deep Kernel Learning (SVDKL) as the base learning model and validate our approach by showing its efficacy in effect prediction tasks where a manipulator robot interacts with objects on a confined tabletop. Experiments comparing MUSEL with strong baselines show that MUSEL improves learning accuracy and sample efficiency. Overall, this study offers MUSEL as an effective online learning model applicable to any robot self-learning task where experience gathering is costly.
With the aim of mitigating the impacts caused by variations in the rainy season that directly affect the production of forage grasses in San Lucas, Michoacán, this work presents the design and evaluation of a direct-solar pumping system for drip irrigation over an area of 0.74 ha. This development considered the meteorological data from the National Renewable Energy Laboratory database, as well as the hydraulic characteristics of the plot and the water requirements of the crop, in order to design both the hydraulic infrastructure and the photovoltaic array. The proposal was evaluated using the specialized software, System Advisor Model, resulting in a system capable of supplying a required flow rate of 3.52 lps during the dry season, ensuring continuous drip irrigation operation, without the use of battery banks, from 10:00 AM. to 4:00 PM. The obtained results support the technical and economic feasibility of the system, therefore, its implementation is recommended as a cost-effective and sustainable alternative for local agricultural production.
Cockpit-style exoskeleton systems are a promising approach for the whole-body teleoperation of humanoid robots, yet they face two major challenges. The first is the difficulty of intuitive torso posture control, arising from the operator’s seated and constrained posture. The second is the lack of transparent feedback for kinematic singularities, which do not manifest as physical forces. This paper proposes two novel methods to address these respective challenges: 1) an indirect torso control interface that converts the operator’s foot- actuated torque into the robot’s torso angular velocity, and 2) a haptic feedback scheme that guides the operator’s arm away from singularities based on the gradient of the arm’s manipulability measure. We implemented these methods into a teleoperation system and successfully demonstrated tasks such as picking up an object from the floor and an operator’s active avoidance of singular postures.
Work-related musculoskeletal disorders (WMSDs) persist in material-handling jobs where lifting, twisting, and carrying induce high, localized muscle demands. This paper presents a sensor-driven framework that (i) detects biomechanical strain from surface electromyography (sEMG) and (ii) quantifies the impact of a passive back-support exoskeleton during industrially relevant tasks. With data from 20 participants performing standardized tasks with and without the device, we introduce a data-driven strain labeling method that replaces ad-hoc thresholds with piecewise linear regression to identify individualized strain onset. A compact deep neural network handles severe class imbalance via SMOTE and decision-threshold optimization, yielding 83.5% overall accuracy and a macro-averaged F1-score of 0.70 for binary strain classification. Muscle-specific analyses reveal significant reductions in biceps and oblique activation (p < 0.001) alongside compensatory increases in erector spinae and lower-limb activity, indicating load redistribution rather than uniform offloading. The result is a scalable, real-time approach that captures both when strain begins and how effort shifts across muscle groups, capabilities that traditional peak-sEMG or subjective assessments miss. By uniting wearable sensing, automated strain onset detection, and imbalance-aware learning, this work advances objective, continuous, and human-centered ergonomic monitoring and provides actionable evidence for the deployment of passive exoskeletons in smart industrial environments.
With the frequency of wildfires increasing, specifically across the Western United States, there exists a need for the reduction in damage to both the community and environment. Many urbanized areas and communities have been significantly affected or demolished by the rapid and devastating spread of recent wildfires. As the impact of global warming increases and the weather patterns begin to vary, the ability to contain a wildfire quickly begins to pose a challenge. While current solutions exist on the market, they are simply not designed for the increasingly faster spread because of ever-changing weather patterns. Additionally, the need for firefighters to have specialized experience has caused a rift between the number of those who can contain fires and the number of fires to be contained. The purpose of this project is the development of a machine learning wildfire UAV swam algorithm to allow the rapid and real time update of containment line calculations/visualizations to allow first-responders to effectively and quickly contain wildfires. By measuring major heat signatures, the UAV system can identify wildfires, and through TensorFlow analysis of wind patterns and environmental markers, make feedback decisions of where to search for spot fires. The combination of wildfires and the spot fires that may arise, result in a containment line visualization with coordinates for first responders to utilize. The goal of the IPCA is a functional autonomous system that can provide the prediction and aid for first responders to have another line of defense against the spread of wildfires into residential and urbanized areas.
Causal Bayesian Networks (CBNs) in robotics are often learned in simulation due to the considerable amount of data required for training. However, discrepancies between simulation and the physical world can cause the learned causal relations to fail in real-world scenarios. Thus, the sim-to-real evaluation is a critical step to deploy a simulation-learned CBN in the real-world. The main challenges in this process are the lack of real-robot evaluation datasets that capture the complexity, noise, and variability of physical environments, which are missing in simulation. In this paper, we propose a set of task-agnostic guidelines for real-robot data collection to evaluate Causal Bayesian Networks (CBNs). The guidelines are generalizable and can be applied to collect real-robot datasets across different robot tasks and platforms. To demonstrate this, we apply them to a robotic platform performing one concrete task, e.g., the robot TIAGo performing a two-cube stacking task, and we collect the real-robot dataset from 100 trials. As a case study, we demonstrate how the dataset can be used to evaluate a simulation-trained CBN on real-robot executions, reporting 10% accuracy drop from sim-to-real transfer. We present this as a first step towards standardized and quantifiable sim-to-real evaluation for CBNs.
In recent years, general-purpose robots have been introduced into domains requiring delicate manipulation, such as materials experimentation. While advances have been made in automating specific processes, generalized robotic pick-and-place operations still pose a challenge due to the diversity of target objects and the need for appropriate force control. This study proposes a novel approach for zero-shot estimation of target grasping force by utilizing the prior knowledge about human motions embedded in Vision-Language Model (VLM). The key idea is to convert robot manipulation images into human-action images using a style transfer approach based on a fine-tuned Variational Auto-Encoder (VAE), enabling the VLM to better infer grasping force requirements. The VLM, specifically GPT-4o, is prompted to estimate target grasping force in discrete categories (no grasp, light grip, firm grip).Experimental results demonstrate that converting robot images into human representations improves the accuracy not only of target grasping force estimation but also of understanding the target objects. Furthermore, the inclusion of target object information in the prompt improves estimation accuracy across all input image types. These findings highlight the effectiveness of utilizing human-knowledge-trained VLM for robotic force control and open new avenues for general-purpose, cost-efficient manipulation without relying on large-scale robot force datasets.
3D printed parts are widely used, which often require high strength and stiffness. However, 3D printed parts face various challenges, properties for printing materials, continuous fiber orientation, moisture absorption, and printing accuracy. Quantitative evaluation of each of these factors will determine appropriate design methods. Therefore, some studies are currently underway to clarify their properties. One example is the insertion of continuous fibers into 3D printed parts, which has been shown to improve strength and stiffness. However, there has been little quantitative evaluation of the insertion orientation and content of continuous fiber. Especially, while the insertion position is theoretically critical to strength and stiffness, there has been no prior approach to evaluate the effect of CF layer insertion position. In this study, we quantitatively evaluated the degree of strength and stiffness improvement by inserting continuous carbon fiber into 3D printed parts during printing. To evaluate the insertion orientation, content and position, we calculated theoretical equations and discussed the discrepancy between theoretical values and experimental results. The results showed that the degree of stiffness improvement achievable in 3D printed parts was clarified in comparison with theoretical values.
To detect obstacles in narrow spaces, such as under-floor area, herein, we proposed a method that estimates the location of obstacles using a distributed model of swarm robots. First, we constructed a swarm robot model, in which a robot moves by jumping at regular intervals in a field enclosed by walls. We confirmed that, when no obstacles were placed in the field, the spread of the swarm robots followed a Gaussian distribution over a certain period. We then assumed that, even when obstacles were present, the distribution of the robots in the field would follow the Gaussian distribution except in the neighborhood of obstacles. Under this assumption, we proposed a method to clearly estimate the positions of obstacles by taking the difference between the approximated Gaussian distribution based on an average gross distribution of robots over a certain period and the actual average gross distribution of robots in each subdivided area of the field.
This study proposes a novel fault-tolerant control strategy for a quadcopter by integrating dynamic equilibrium analysis with nonlinear model predictive control (NMPC). First, we formulate an optimization problem based on dynamic equilibrium analysis. The feasibility of maintaining a constant altitude under a fault condition is determined by solving this problem. This analysis yields an optimal target that minimizes yaw angular velocity, while satisfying physical limitations. The target then serves as an online-updated reference for NMPC. This integrated approach enables a single controller to manage various rotor faults with a single fixed set of controller weights. Furthermore, the controller explicitly handles angular velocity constraints within the sensor limits, enhancing its reliability in a real quadcopter. The effectiveness of this strategy is verified through numerical simulations using a model based on an actual quadcopter, which enables visualization of operational bounds for various fault conditions. Simulations also demonstrate that stable flight can be achieved, while satisfying constraints even after a severe rotor fault.
As robots are deployed into new domains, teleoperation with a human-in-the-loop remains an important method for training new skills. Direct control of a remote robot over a network demands high concentration of the human teleoperator and remains challenging due to unavoidable network delays and congestion that can degrade performance. In this work, we demonstrate a shared control solution using well established virtual fixtures (VFs) as a teleoperator assistance tool and introduce a novel VF generation pipeline leveraging mate constraints in available CAD data of the manipulated objects. This approach simplifies the definition of VFs by taking parameters from relationships already defined within the CAD data. Using an industry-sponsored task board and a bilateral leader-follower hand-guided robot scenario, we demonstrate how a set of VFs can be constructed and activated in a series to support teleoperators with a multi-step manipulation task including the pressing of buttons, peg-in-hole with the picking and inserting of a Multimeter Probe Plug, and a novel VF task of opening of a hinged door. We present data from a pilot user study with eight teleoperators and 67 trial attempts with two scenarios (with and without VFs enabled) across three test conditions of round-trip network delays of 0 ms, 100 ms, and 250 ms. We found teleoperators had an increased task success rate, lowered the total travel distance, and overall reduced task execution time, in the best case of 0 ms delay by 26 seconds, or 21%, when VFs were enabled versus when they were not. Performance was maintained or improved at higher network delays.
The performance of robotic imitation learning (IL) largely depends on the quality of human demonstrations. To address this challenge, we present AIREC-Basic, a leader–follower teleoperation system equipped with a dual-arm mobile manipulator that enables efficient data collection. The system employs 8-DoF redundant arms to realize diverse task postures, but such redundancy can reduce the consistency of demonstrations. To overcome this issue, we propose a novel control strategy, Soft Homing Control (SHC), which mitigates redundancy while preserving intuitive operator control, thereby improving dataset consistency. We validate our approach on three household tasks using state-of-the-art IL algorithms (ACT, Diffusion Policy, and HSARNN). Experimental results show that SHC significantly reduces joint trajectory variance and improves task success rates, particularly in scenarios with strong trajectory constraints and frequent contacts.
Radial pulse measurement is an important physiological assessment method in healthcare and wellness monitoring. While most existing pulse measuring devices are dedicated instruments requiring precise placement, this work explores a human-like approach using a general-purpose humanoid robot. In this work, we present a proof-of-concept study on radial pulse measurement using the AIREC humanoid robot equipped with a 3-axis 8×6 matrix uSkin tactile sensor embedded in its palm. A dual-arm motion strategy allows one hand to support the subject’s wrist while the other applies gentle pressure for stable human-robot contact. Contact stability is evaluated by both normal and shear measurements. Then, the tactile signals from the array are scanned to identify the optimal sensing location, and bandpass filtering is applied to extract the pulse waveform. Preliminary results show accurate pulse rate estimation compared with a commercial pulse oximeter, demonstrating the feasibility of a human-like bio-information measurement in humanoid robots.
This paper presents the design, fabrication, and evaluation of a textile-based strain sensor integrated into a soft robotic joint for real-time monitoring of bending angles. The proposed sensor consists of a conductive fabric encapsulated in a flexible silicone substrate, created using a custom co-casting process that enables seamless integration without compromising structural compliance. Electromechanical characterization under tensile and compressive loading demonstrated a strong piezoresistive response, with a gauge factor (GF) of 77.76 under tensile strain and 9.24 under compression, within working ranges of 6–12% and 3–14%, respectively. The sensor was embedded into a soft joint with asymmetric geometry to assess directional sensitivity and real-time performance. Experimental results confirmed reliable bidirectional strain detection and an over 500% increase in resistance under tensile bending. A second-order polynomial model accurately mapped the sensor’s resistance changes to joint bending angles (10°–25°), achieving R2 values above 0.67 in both tensile and compressive configurations. These findings validate the sensor’s potential for real-time posture estimation in wearable and soft robotic systems.
The state estimation is important issues in system theory and control systems. Especially, estimation of the state variables of general nonlinear systems is essential but it is still difficult at the same time. This paper proposes design methods for an observer of a nonlinear system described by Takagi-Sugeno(T-S) fuzzy system. Because of the capability of representation of T-S fuzzy system, our observers are designed for a quite large class of nonlinear systems, which cover most systems in various engineering fields. Our observer design employs multiple Lyapunov matrix methods, which result from Lyapunov function candidate with the multiple integrals of the membership functions. Our resulting observers are a nonparallel distributed observer (PDO). This method drastically reduces the conservativeness of observer design conditions. In order to provide the usefulness of our proposed design approach, an illustrative example is provided. Finally, we end with concluding remarks.
This paper proposes a control method for robotic hammering that can attenuate the reaction torques on the robot’s joints while maintaining a sufficient impact on the environment. A key challenge in hammering tasks is this trade-off between delivering a sufficient impact and preventing damage to the manipulator. The proposed method addresses this conflict by generating the entire pre-collision trajectory, including the required initial posture, by calculating backward in time from a desired state at the moment of impact. The trajectory consists of a preparatory backswing and a striking motion, both tracked using a PD controller. To further attenuate reaction torques, a feedforward torque pulse is applied to the final joint just before impact. After impact, a force-controlled lift-up mode is activated to raise the hammer and return the robot to its initial posture. The effectiveness of the proposed method is validated through experiments, demonstrating an attenuation in joint torque.
This study evaluates a soft robotic wearable device, equipped with a Hyper-extension Pneumatic Actuator (HPA), to prevent deep vein thrombosis (DVT). We investigate the hemodynamic efficacy of a novel "Combined" exercise protocol, which strategically integrates the high-load advantage of resistance exercise with the wide range of motion from induced exercise, all delivered by a single soft actuator system. The effects of four distinct exercise modalities on key hemodynamic parameters, including time-averaged maximum blood flow velocity (TAMAX), were systematically evaluated in healthy participants. Results demonstrated that all active exercise modalities showed a strong tendency to augment blood flow velocity compared to rest (effect size r>0.8). Notably, the "Combined" protocol yielded a statistically significant increase in TAMAX compared to exercise without the device (p < .05). This enhancement is attributed to the protocol's ability to elicit high-quality, high-tension muscle contractions by balancing exercise load with a sufficient range of motion, thereby optimizing the muscle pump mechanism. These findings provide engineering guidelines for developing effective DVT prevention strategies, demonstrating that the intelligent integration of a control protocol—not just the hardware alone—is key to maximizing hemodynamic efficacy.
In this paper, we propose spectral-to-spatial distillation, a novel denoising framework for real-time anomalous sound detection. While anomalous sound detection is crucial for industrial applications, its reliability is often compromised by background noise, which can lead to false positives. Our proposed method addresses the issue of background noise by distilling knowledge from a general-purpose spectral filtering network into an environment-specific spatial filtering network. Specifically, we generate distillation targets, which are audio signals with reduced noise, using a pre-trained foundation model. A spatial filtering network is then trained using these targets. A key feature of our distillation process is its ability to automatically generate these targets using only one-shot, brief, noise-free reference signal of the target sound. Furthermore, we introduce a new quality metric for these distillation targets, called Semantic Clarity improvement (SCi). By leveraging the semantic audio embedding capabilities of a foundation model, SCi measures the improvement in semantic similarity between the distillation target and the reference signal. This SCi allows for effective distillation by weighing the loss function based on the quality of the targets. Experimental results demonstrate that our method achieves the best denoising and anomaly detection performance while maintaining real-time processing capabilities, making it a practical solution for noisy industrial environments.
Because of labor shortages, robots are expected to provide work assistance in a variety of settings, including the home environment. At home we often deal with flexible objects, but flexible objects are characterized by their tendency to change position and shape. Because of this nature, data dealing with flexible objects involves uncertainty. Although deep learning has been used to perform a variety of complex tasks, the deterministic nature of conventional RNN makes it difficult to handle data with a probabilistic structure. In this study, we propose a method based on deep predictive learning that enables real-time motion generation and predicts the variance of joint angles, which facilitates learning of probabilistic structures and can handle dynamic changes. Experimental results show that the robot is able to generate motions that are adaptive to flexible objects with dynamic position changes.
