Keywords: Robot Vision and Sensing, Off-road mobile robots, search and rescue robots
Abstract: Disaster images are often corrupted by motion blur due to rapid capture during emergency escapes, which can mislead AI models and result in incorrect decisions. To address this challenge, we propose BRIDGE, a blur-adaptive spatial transformation module based on Spatial Transformer Networks (STN), designed to enhance pretrained image classifiers (e.g., SwinV2) without requiring image restoration or structural modification. BRIDGE dynamically transforms blurred inputs to improve classification robustness while preserving modularity. Experiments demonstrate that BRIDGE consistently outperforms existing baselines on both corrupted and restored disaster images across four tasks in the Crisis Image Benchmark dataset. Furthermore, given the growing deployment of rescue robots and UAVs in disaster response, our modular and cost-effective framework provides practical potential for integration into real-time robotic perception systems, since most on-board vision stacks rely on lightweight, plug-and-play modules.

Keywords: Off-road mobile robots, search and rescue robots, Robot Vision and Sensing, Autonomous Vehicles
Abstract: Retrieving tracked vehicles from dangerous settings after deployment poses major problems and risks to personnel. Remotely operated transport vehicles are utilized for retrieving tracked vehicles, which increases operational safety. However, latency in camera feedback can reduce operator accuracy during retrieval. This study presents a visual servo control approach to mitigate such delays by enabling the tracked vehicle to autonomously align with the rear of the transport vehicle. By integrating detection through ArUco markers affixed to the transport vehicle ramp, the tracked vehicle adjusts its positioning during its operation.

This study developed an alternative multi-robot unmanned ground vehicle (UGV) system consisting a large transport vehicle and a compact tracked vehicle for exploration and data gathering application in dangerous settings. A remotely controlled door-to-ramp system for easy deployment and retrieval of the tracked vehicle was developed. Semi-autonomous control strategy to improve the tracked vehicle's performance was developed. An actual retrieval procedure was done focusing on tracked vehicle visual servo control, with an average error distance of 75 mm and a maximum error of 210 mm which was within the allowable margin of 282.7 mm , indicated the effectiveness of the system. The tracked vehicle successfully implemented the visual servo control and adjusted its position within the transport vehicle.

Keywords: Robot Vision and Sensing, Socio-economic Impacts, Machine Learning and Robot Learning
Abstract: Flooding remains one of the most destructive and expensive natural hazards, yet many existing models lack the high-quality, occlusion-free data required for accurately estimating key parameters necessary for flood risk modeling. In this paper, we propose a comprehensive pipeline that combines a mobile data collection platform with a multi-view diffusion-based inpainting algorithm to capture and reconstruct suburban unobstructed building facade features. By fusing LiDAR and camera data, our approach effectively handles vegetation and other obstructing objects, generating compact, occlusion-free images. Experimental results show that our multi-view fusion and consistency loss substantially outperform single-view baselines and GAN-based inpainting in terms of fidelity and cross-view alignment. We demonstrate notable gains in feature detection availability (doors and stoops) using the reconstructed occlusion-free images.

Keywords: Robot Vision and Sensing, Traceability and Safety, Off-road mobile robots, search and rescue robots
Abstract: Technologies such as robotics, Artificial Intelligence (AI), and Computer Vision (CV) can be applied to crime scene analysis (CSA) to help protect lives, facilitate justice, and deter crime, but an overview of the tasks that can be automated has been lacking. Here we follow a speculative prototyping approach: First, the STAIR tool is used to rapidly review the literature and identify tasks that seem to have not received much attention, like accessing crime scenes through a window, mapping/gathering evidence, and analyzing blood smears. Secondly, we present a prototype of a small drone that implements these three tasks with 75%, 85%, and 80% performance, to perform a minimal analysis of an indoor crime scene. Lessons learned are reported, toward guiding next work.

Keywords: Intelligent robots for transportation
Abstract: This paper addresses the critical logistical challenges in disaster-struck urban environments, where rapid response is essential for clearing debris and rubble and delivering supplies and medical aid. We introduce a novel framework for Obstacle-Aware Continuous Location-Routing in DisasterStruck Areas. In the continuous location-routing problem, facilities can be located anywhere in the environment, which may include areas obstructed by obstacles. Our approach integrates motion planning techniques to navigate obstacles effectively and identify collision-free areas for facility placement. The solution unfolds in three phases: First, we use sampling-based motion planning to construct a probabilistic roadmap, enabling collision-free trajectories between pickups. Second, we address the capacitated clustering problem through a capacitated agglomerative clustering algorithm to form pickup clusters. For each cluster, we then solve the continuous facility location problem using a reinforcement learning-based method, a grid map approach, and a modified center of gravity method. Finally, we tackle the vehicle routing problem as a multi-pickup and delivery problem with time windows, utilizing OPTIC as our planning solver. The experimental setup employs a second-order vehicle model within an obstacle-rich environment. We evaluate the performance of our methods by analyzing various routes and costs under different parameters, such as the alpha value of the reinforcement algorithm and the maximum item load. The results are measured in terms of runtime and average travel distance, demonstrating the effectiveness of our approach.

Keywords: The role of AI in the implementation of cognitive robots
Abstract: The increasing inefficiencies in conventional waste management systems, including suboptimal recycling rates and environmental degradation, necessitate innovative solutions. This paper discusses the development of an Internet of Things (IoT)-enabled waste tracking and recycling optimization system designed to address these challenges and contribute to sustainable waste management practices. The primary focus is on automating the waste classification process and enhancing recycling efficiency through real-time monitoring and data-driven analysis. The methodology integrates IoT technology and machine learning to tackle waste classification and collection inefficiencies. A Convolutional Neural Network (CNN) trained on a dataset of aluminium cans and plastic bottles is deployed for waste identification. Real-time monitoring is enabled by IoT sensors and machine vision algorithms, facilitating precise detection of waste levels and material types. Advanced data preprocessing, such as augmentation and normalization, ensures robust model training, while optimized algorithms guide waste sorting based on classification results. Findings demonstrate that the system achieves over 90% accuracy in classifying recyclable materials. Real-time data logging enables analysis of waste composition, container utilization, and operational patterns, enhancing efficiency and reducing overflow incidents. Data visualization highlights the system’s potential for providing actionable insights to improve recycling practices. In conclusion, this project validates the feasibility of integrating IoT and machine learning to optimize waste management. The system reduces environmental impact and promotes sustainability, offering a scalable framework for addressing global waste challenges.

Keywords: Robot Vision and Sensing, Machine Learning and Robot Learning, Industrial robotics
Abstract: The growing global waste situation, especially involving recyclable materials like aluminum cans, transparent bottles, and milk cartons, emphasizes the need for efficient waste sorting systems. However, the lack of diverse, high-quality datasets limits machine learning models' performance in real-world recycling scenarios. This study addresses this challenge by generating a synthetic dataset using Blender, simulating diverse environmental conditions such as lighting, backgrounds, and object orientations. The generated dataset also automatically outputs the labels corresponding to the bounding boxes of the location of each object in each image, eliminating the need for manual labeling. We used the YOLOv11 object detection model on our generated dataset. The model demonstrated strong performance in detecting and classifying recyclable objects of varying sizes and complexities. This approach showcases the potential of synthetic data to train robust models, reducing reliance on labor-intensive data collection and improving recycling efficiency to advance sustainable waste management techniques.

Keywords: Robotics for manufacturing, Industrial robotics, Socio-economic Impacts
Abstract: Ship recycling, which involves dismantling of ships, contributes to the circular economy by recycling materials from decommissioned ships. However, despite the existence of international regulations, ships are still dismantled on beaches under rudimentary conditions. Workers manually cut the ship's metal plates with oxy-fuel torches, often resulting in accidents, injuries, and occupational diseases due to unsafe working conditions and exposure to toxic substances. Coastal ecosystems and local communities are exposed to toxic spills and pollution due to lack of containment. We believe that a sustainable ship dismantling process that prioritizes worker safety and environmental responsibility can be achieved through robotic cutting of metal blocks. As a first step toward realizing this robotic cutting concept, this paper examines three cutting methods (oxy-fuel, plasma, and waterjet) based on four factors: environmental impact, safety, technical feasibility, and cost. The paper's findings, based on these design considerations, including human and environmental factors, indicate that plasma cutting technology is best suited for the proposed robotic approach to sustainable ship dismantling.

Keywords:
Abstract: AI-equipped robots are beginning to be used in various fields such as healthcare, transportation, caregiving, manufacturing, customer service, and disaster prevention. In some cases, machines autonomously make decisions and judgments without human involvement, raising concerns about safety, fairness, and accountability in the event of accidents. On the other hand, remotely operated avatar robots are also starting to be utilized in customer service, security, and retail settings.

In conjunction with the Sustainable Development Goals and Japan’s Society 5.0 initiative, it is crucial to ensure that AI and robots do not threaten human dignity and contribute to creating a diverse and inclusive society.

To govern technology appropriately, discussions around various principles, including soft laws such as guidelines, are necessary, along with proposed frameworks for implementing these principles in practice. However, the transition from principles to practice is not straightforward. What do safety and fairness mean in the context of the fields? Is the intention to create an inclusive society ironically leading to the emergence of exclusive or overly busy societies?

Society is complex and multilayered. We will recognize the challenges and contradictions and think together about how to overcome them.

Keywords: HRI and social robotics
Abstract: The experience of being ignored and excluded is referred to as ostracism, often triggers aggressive behavior in those being ostracized. Such a relationship between ostracism and aggressive behavior is well documented in human-human interactions and can be explained with the help of some fundamental mechanisms e.g., presence of anger, feelings of relative deprive, and lack of situational control, etc. However, it remains unclear whether such ostracism-aggression relationship extends to human-robot interactions, and whether it can be explained through some fundamental mechanism. To explore such research gaps, a subjective evaluation based experiment was conducted, in which thirty participants were asked to watch two video stimuli and provide their impressions concerning their perceived ostracism, aggression, and behavioral interdependence feelings. The findings revealed that ostracism-aggression relationship is extendable to human-robot interactions and behavioral interdependence serves as one of the key mechanisms, mediating ostracism-aggression relationship in human-robot interactions.

Keywords: Industrial robotics, HRI and social robotics, Socio-economic Impacts
Abstract: This paper presents TeleSign, a novel and flexible gesture-based teleoperation framework inspired by American Sign Language. Using either a static or wearable RGB camera, TeleSign integrates Machine Learning with NVIDIA Isaac Sim and ROS2 to enable intuitive control of a manipulator's end effector through hand gestures, allowing for precise and responsive movements. We evaluated TeleSign through multiple experiments, assessing the performance of a Random Forest classifier, gathering user feedback, and testing its effectiveness in complex manipulation tasks. The results demonstrate strong performance across all areas, highlighting TeleSign’s potential for efficient and intuitive robot teleoperation.

Keywords: HRI and social robotics, Legal and ethical regulations for AI and robotic systems, Legal Aspects
Abstract: Mental states, such as beliefs, intentions, and desires, have been considered key elements of human agency, indicating one's capacity to act deliberately and with awareness. Thus, these mental states have been treated as crucial factors in assessing human accountability. However, it remains unclear whether these mental states are equally important to robot agency and, by extension, to robot accountability. To explore this tension, we investigated how participants judged the relevance of mental states in determining the accountability of an industrial robotic arm or a humanoid robot compared to a human after a workplace accident in which a worker was injured. Our results showed that participants viewed desires, beliefs, and intentions as significantly less relevant when assessing the accountability of robots than humans, and robots were judged as less accountable overall. These findings suggest a need to move beyond human-centered views of agency--particularly those based on assumptions about mental states--when discussing the accountability of robots.

Keywords: HRI and social robotics, Service and assistive robotics, Entertainment robots
Abstract: In this study, our objective was to generate the expressive behaviors of a robot and evaluate their effects on humans when playing a trump game together. By giving the robot the main task of carrying playing cards and the subtask of performing expressive motions using null-space control. In designing the expressive motions, we set a trajectory that would allow for redundancy within the robot's kinematic space and designed motions that will enable the robot to convey emotions to the human players using the robot redundant degrees of freedom. In the experiment, we played the old maid card game using the designed expressive motions and verified what they conveyed to the participants. In some conditions, the subjects had a different perception than we had expected, and in some cases the emotions were not conveyed properly, but in other cases, the intended emotions were conveyed, as there were differences of assessment seen in each motion. Regarding the impression of the robot, it was suggested that performing emotional motions made it easier to feel positive emotions. Designing more detailed emotional behaviors and repeating experiments will lead to the widespread use of interaction robots with limited degrees of freedom and to the expansion of the potential of communication by robots.

Keywords: Service and assistive robotics, The role of AI in the implementation of cognitive robots, HRI and social robotics
Abstract: The development of socially assistive robots is an emerging field aimed at addressing the challenges of aging populations. DarumaTO, a culturally grounded robotic companion inspired by the Japanese Daruma doll, was developed within the e-ViTA initiative to foster engagement, reduce loneliness, and provide a familiar, interactive experience for older adults. DarumaTO was exhibited in Italy, where 148 participants provided structured feedback. Participant’s feedback was supplemented with non-intrusive participant observation. It is the first time that this Buddhist talisman device gets tested in a Western country. The findings highlight the cultural impact on robot perception, demonstrating that religious and cultural background influences how a robot may be recontextualised within the process of technological adoption.

Keywords: HRI and social robotics
Abstract: In daily life peer-to-peer interactions, gender and group size (i.e., number of peers engaged) could affect right to speak (RoS) and peer pressure (PP) perceived. Such a fact is explored in human-human peer interactions. However, in the case of human-robot peer interactions, it is not clear whether gender and group size could affect RoS and PP perceived and consequently help us re-shaping conversations and gain control over those. In this research, we addressed such a research gap. We explored the effect of gender and group size on RoS and PP perceived. Two subjective evaluation based experiments were conducted: one depicting online and one depicting face-to-face interactions between human and robot peers. 2X2 mixed design was adopted for each experiment, where gender with two levels i.e., males and females, was a between subject factor, group size with two levels i.e., one and two robot peers, was a within subject factor, and RoS and PP perceived were the measures. In online interactions, a significant interaction effect between gender and group size was observed on RoS perceived. While, in case of PP perceived, a significant main effect of gender was observed. Contrarily, in face-to-face interactions, a significant main effect of the group size was observed. While, in case of PP perceived, no effect of gender and/or group size was observed.

Keywords: Human Movement Modeling, Machine Learning and Robot Learning, HRI and social robotics
Abstract: When performing tasks such as handovers in human-robot interactions (HRI), the robot's position must be carefully planned to avoid causing injury and strain to users. As with robot safety systems that focus on reducing risks in transient interactions, planning ergonomic HRI points can be done to reduce risks in static interactions. Programming ergonomic interaction points can pose a unique challenge due to the diversity and variability in human body shapes and sizes. Furthermore, consistent interaction using the same point, even in an ergonomic position, can lead to the risk of developing repetitive strain injuries (RSI). A method of creating a human-robot ergonomic work envelope is proposed. By utilizing an ergonomic work envelope rather than specific points will provide a robot with more flexibility during path planning for HRI applications. An experiment was conducted to simulate human-robot handover interactions. The data from this experiment was then used to create a machine learning model of a human-robot ergonomic work envelope customized to a specific user. The model was used to predict ergonomic scores of HRI points based on the robot Tool Centre Point (TCP) position. The human-robot ergonomic work envelope can allow a robotic system to dynamically adjust handover points to promote ergonomic interaction and prevent repetitive strain by avoiding fixed handover positions.

Keywords: HRI and social robotics, Machine Learning and Robot Learning
Abstract: The expressivity of robots plays a crucial role in enhancing robots’ communication capabilities with humans. Previous approaches for generating expressive movements re- lied on manually crafted motions, learning from demonstra- tions, or adapting existing expressive motions. However, these approaches are labor-intensive and typically limited to specific robot platforms, tasks, and expressivity, which constrain their applicability and generalization. To overcome these limitations, we propose a generative model that extracts expressive features from human movements and transfers them to robots. By learning expressivity directly from humans, our approach eliminates the need for datasets tailored to specific robots embodiments and ensures that the generated expressivity is easily perceived by human observers. We validate our method through simulations and experiments on a real robot, demonstrating that our model effectively captures and transfers the expressivity. We also highlight the model’s capacity to enable the creation of a broad range of expressions beyond those used in training.

Keywords: HRI and social robotics
Abstract: The advent of powerful Large Language Models (LLMs) enables an era where humans and AI agents collaboratively engage in dynamic discussions to make decisions together. While prior research has largely focused on LLMs as assistant tools for human decision-making, their potential to co-create decisions through iterative knowledge integration remains less explored. This study introduces a novel model of human-LLM co-creative decision-making, formulated within the distributed Bayesian inference framework. We specifically interpret the iterative cycle, where the LLM proposes options based on its knowledge and the human selects based on preferences, as an instance of the Sampling-Importance-Resampling (SIR) algorithm. To validate our model, two experiments were conducted using LLM agents as substitutes for human participants under controlled conditions: a cooperative card-guessing task and a travel brainstorming task. Findings from the card-guessing task revealed that iterative information integration from both agents leads to a progressively updated decision distribution, consistent with SIR dynamics. Furthermore, the brainstorming task demonstrated that our proposed co-creative model facilitates robust dynamic knowledge integration and adaptive convergence of decisions. By framing co-creative decision-making as the emergence of a shared posterior distribution from integrating diverse knowledge, this research establishes a theoretical foundation for advancing human-AI symbiotic societies.

Keywords: Machine Learning and Robot Learning, Industrial robotics
Abstract: Composite learning can guarantee the exponential stability of adaptive robot control under a condition of interval excitation (IE) that weakens the classical condition of persistent excitation. However, the existing composite learning law must be real-time updated even if tracking and estimation errors have already converged, resulting in the waste of computational resources. This paper proposes a dynamic event-triggered composite learning robot control (ET-CLRC) strategy to relax computing burden, where a dynamic ET condition with adaptive thresholds is designed such that control and adaptation laws are updated only when the predefined condition is satisfied. The proposed method is distinguished from existing ET adaptive robot control methods in the following aspects: 1) The designed event consists of tracking errors and internal variables without involving real-time updated information on control and adaptation laws; 2) the practical exponential convergence of tracking and estimation errors under the IE condition is derived while avoiding the Zeno behavior. Experiments on an industrial robot with seven degrees of freedom have shown that the proposed ET-CLRC can achieve control performance similar to that of real-time updated CLRC while significantly relaxing the computational burden.

Keywords: Machine Learning and Robot Learning, Autonomous Vehicles, Traceability and Safety
Abstract: This paper explores the convergence of traditional and explainable AI (xAI) saliency maps to improve interpretability in deep learning models. It compares a classic saliency map approach (VOCUS2) with a current xAI method (Grad-CAM++), analyzing their advantages, disadvantages, and implications in the context of traffic sign detection for autonomous vehicles. The study examines the implementation of both traditional and xAI saliency maps, their shortcomings, and relevant social concerns. It also evaluates how the models perform when faced with different types of noise and variations.

Keywords: Machine Learning and Robot Learning, Robot Vision and Sensing
Abstract: Foundation Models for Robotics (FMRs) promise to bring large-scale, generalist intelligence to embodied systems, yet they remain limited in their ability to integrate perception, action, and reasoning in physically grounded environments. In this paper, we argue that advancing FMRs requires drawing inspiration from biological systems—specifically human cognition, development, and sensorimotor learning. We outline five key bio-inspired principles for future FMRs: (1) memory architectures incorporating semantic, episodic, and procedural structures; (2) grounded structured reasoning, as exemplified by embodied chain-of-thought (E-CoT) processes; (3) integration of multimodal sensorimotor feedback, including touch and proprioception; (4) self-motivated learning through simulated play and intrinsic exploration; and (5) neural efficiency through sparse expert activation, functional specialization, and modular reasoning. These elements enable generalization, compositionality, and robustness—traits long demonstrated by humans but underrepresented in current robotic models. While this work does not address reliability and safety in depth, we identify them as essential future directions for developing trustworthy, human-aligned FMRs.

Keywords: Autonomous Vehicles, Machine Learning and Robot Learning, Human Vehicle Interaction
Abstract: Teleoperation is a technology that enables remote control of robots, rovers, and vehicles, and it is widely used in various fields such as surgical robotics, lunar exploration, and unmanned shared vehicle relocation. However, teleoperation systems rely on wireless communication to exchange control commands and sensor data, which introduces communication delays that can degrade control stability. While various approaches have been studied to mitigate the effects of such delays, they continue to pose a major challenge in teleoperation systems. In previous research, we proposed PT-DCD (Predictor for Teleoperation under Dynamic Communication Delay), a data-driven, model-free predictor based on a deep learning network, specifically the Long Short-Term Memory (LSTM). PT-DCD showed the potential to mitigate the effects of communication delay by predicting real-time control commands. Specifically, when the PT-DCD was validated under varying outlier ratios, its delay-reduction performance was consistently observed across all environments. Therefore, in this paper, we evaluate whether PT-DCD can perform effectively in environments it has not been trained on, assessing its generalization capability. Performance is measured using three metrics: average trajectory error, goal point error, and acceptable error ratio. Consequently, experimental results show that PT-DCD effectively reduces the impact of communication delays and improves teleoperation stability.

Keywords: HRI and social robotics, Industrial robotics, Humanoid Robots
Abstract: This paper introduces a bilateral teleoperation system for UAVs that employs a hybrid control scheme combining rate and non‐linear position modes. By continuously switching between these modes, the system achieves both agile manoeuvring and precise positioning under communication delays. Validation is carried out using a dynamic model for the master robot with a Novint Falcon haptic device and a simplified model for the slave robot in Gazebo-ROS2. Performance metrics including task completion time, mean squared error, and force feedback demonstrate enhanced stability and efficiency, suggesting promising applications in inspection, environmental monitoring and search and rescue.