Keywords: Computer Vision and Visual Servoing
Abstract: Humans adjust their lower limb control strategies based on the terrain they are walking on. This principle also applies to users of robotic orthoses or prostheses. However, currently commercialized robotic orthoses and prostheses do not adapt their control strategies according to terrain; instead, they apply the same walking control as if on flat ground. This leads to awkward and uncomfortable walking patterns for the wearer. The root cause of this issue is that the robot is unable to recognize and classify different terrains. To address this, we propose a terrain classification system based on stereo depth cameras. The proposed classifier categorizes the data from the stereo depth camera into five distinct terrain types: level ground, ramp ascent, ramp descent, stair ascent, and stair descent. By leveraging the 3D information provided by the stereo depth camera, the system is able to effectively differentiate between flat ground and slopes, which are challenging to distinguish using an RGB camera. The terrain classification system achieved an accuracy of 87.06%. This demonstrates that effective terrain classification can be accomplished using only the 3D data from the stereo depth camera.

Keywords: Computer Vision and Visual Servoing, Dynamics and Control, Modeling, Identification, Calibration
Abstract: Uncalibrated visual servoing (UVS) aims to adaptively control robots without explicit calibration, promising performance in unstructured or dynamic environments. In principle, integrating recursive least squares to estimate the image Jacobian online can enhance tracking accuracy and resilience to uncertainties. This study implements an RLS-based UVS framework on a three-cable parallel robot with an eye-in-hand camera. Experiments and simulations compare the adaptive approach to a baseline using a static, globally calibrated Jacobian. While the RLS method demonstrated the ability to fine-tune the Jacobian over time, steady-state accuracy and convergence speeds were often similar or inferior to the static approach, particularly under large initial offsets or poor initial conditions. These findings highlight the sensitivity of adaptive UVS to initialization quality and the complexity of achieving practical gains in noisy, nonlinear settings. They underscore the need for improved strategies, such as refined initialization or hybrid methods, to fully realize the theoretical potential of RLS-based uncalibrated servoing in challenging real-world scenarios.

Keywords: Computer Vision and Visual Servoing, Mechanism and Design, Object Recognition
Abstract: Laboratory automation is an expensive and complicated endeavor with limited inflexible options for small scale labs. We developed a prototype system for tending to a bench-top centrifuge using computer vision methods for color detection and circular Hough Transforms to detect and localize centrifuge buckets. Initial results showed that the prototype is capable of automating the usage of regular bench-top lab equipment. Experimental results showed the computer vision system to have successful detection rate of 98%, a 70% success rate for removing test tubes from a centrifuge and a 95% success rate for inserting them.

Keywords: Modeling, Identification, Calibration, Computer Vision and Visual Servoing, Actuation and Actuators
Abstract: This paper presents the design, calibration, and error evaluation of a compact three-axis (X-Y-Z) robot equipped with an Intel RealSense LiDAR L515 camera. The camera’s role is to detect and provide the (x, y, z) coordinates of source and target points, so the manipulator can move its end-effector to those points. The X-axis is laid out horizontally, the Y-axis is mounted on the X-axis, and the Z-axis (end-effector) is mounted on the Y-axis, enabling it to reach a workspace of 50 cm (X-direction), 35 cm (Y-direction), and 30 cm vertically (Z-direction). The LiDAR camera offers depth perception and point cloud data to accurately locate features of the plant in three-dimensional space. We outline how we achieved an efficient camera-robot calibration process and how the system uses coordinate transformations to command the manipulator from a source to a destination with minimal error. Results show that the end-effector can consistently reach target points in the environment with a relatively small positioning error, demonstrating the feasibility of our setup.

Keywords: Simultaneous Localization and Mapping (SLAM), Intelligent Robotic Vehicles, Autonoums Vehicle Navigation
Abstract: Recent advancements in autonomous vehicle odometry estimation have been largely driven by the integration of various sensor technologies. Among these, Light Detection and Ranging (LiDAR) and cameras play a crucial role; however, both exhibit inherent limitations. In particular, cameras, which are widely utilized, are highly susceptible to illumination changes. In contrast, LiDAR is robust to such variations, making it a powerful tool in Simultaneous Localization and Mapping (SLAM). To enhance LiDAR performance, numerous sensor fusion approaches incorporating Inertial Measurement Units (IMUs) have been proposed. Nonetheless, LiDAR-based methods still face challenges in accurately estimating vehicle states in environments with repetitive patterns. This paper introduces a novel framework to improve LiDAR odometry estimation accuracy in repetitive pattern environments by leveraging the complementary strengths of both cameras and LiDAR. Specifically, we employ a visual feature tracking-based approach that utilizes 2D range images generated from 3D point cloud data. The use of 2D projected range images enables robust feature extraction while maintaining resilience to illumination changes. The proposed method is evaluated on real-time vehicle state estimation tasks using datasets containing repetitive patterns. Experimental results demonstrate that our approach outperforms traditional LiDAR-based methods, validating the effectiveness of incorporating LiDAR vision techniques in such challenging environments.

Keywords: Simultaneous Localization and Mapping (SLAM), Object Recognition
Abstract: This study presents a novel semantic loop closure method that emphasizes the uniqueness of objects to enhance robustness against false matching in simultaneous localization and mapping (SLAM). Loop closure, a critical technique for detecting revisits during SLAM, has traditionally relied on image feature comparison methods, such as visual bag of words. However, these approaches are highly sensitive to viewpoint, often fail to identify revisits. Semantic segmentation-based methods offer improved robustness but have predominantly focused on the geometric distribution of semantic objects, which increases susceptibility to false positives in environments with repetitive patterns. By incorporating object uniqueness into loop closure detection, the proposed method addresses these limitations effectively. Experimental results show that it achieves greater robustness against false positives than conventional semantic loop closure methods while also reducing false negatives compared to image feature-based approaches.

Keywords: Foundations of Sensing and Estimation, AI Reasoning Methods for Robotics, Rehabilitation and Healthcare Robotics
Abstract: In recent years, wearable healthcare devices and robots have become increasingly crucial in the face of population aging and the rising prevalence of chronic diseases, including cardiovascular disease, which remains a leading cause of mortality worldwide. These conditions amplified the demand for advanced, continuous monitoring of cardiovascular health. While electrocardiogram (ECG) signals offer comprehensive diagnostic insights, their reliance on multiple electrodes often limits practicality. In contrast, photoplethysmogram (PPG) signals are more convenient to acquire but lack the rich detail of ECG. In this paper, we propose a novel PPG-to-ECG reconstruction method that combines an Inception CNN for multi-scale feature extraction with a Transformer architecture for capturing global dependencies. Our proposed method achieves robust ECG signal reconstruction even under high-noise conditions by effectively preserving local morphological details and leveraging long-range contextual information. We validate the proposed approach on diverse datasets spanning everyday life and intensive care unit (ICU) settings, demonstrating high accuracy and generalizability. Experimental results indicate an RMSE of 0.26, corresponding to a 10% improvement over state-of-the-art methods. These findings highlight the feasibility of reliable, real-time ECG reconstruction from PPG signals alone, paving the way for scalable and accessible healthcare monitoring solutions in clinical and wearable contexts.

Keywords: Force and Tactile Sensing, Object Recognition, Contact: Modeling, Sensing and Control
Abstract: As robots increasingly interact with unstructured environments, understanding extrinsic contact is crucial for precise manipulation. In particular, estimating the direction vector of an object in contact with an external surface enables tasks such as alignment, insertion, and controlled motion. We propose a novel approach to estimating the direction vector of extrinsic line contact using vision-based tactile sensors. By tracking the relative motion of a grasped object and leveraging kinematic and frictional constraints, the proposed method accurately infers the contact direction without requiring prior geometric information about the object. The system employs a high-resolution tactile sensor and a motion-tracking algorithm to extract object displacement data, which is then transformed into the world frame for robust estimation. Experimental validation with a robotic manipulator demonstrates that the proposed method achieves an overall mean angular error of 8.71˚, confirming its effectiveness in real-world applications. This approach enhances robotic perception and manipulation capabilities, making it a reliable solution for tasks that require precise handling of extrinsic line contact, such as assembly, insertion, and surface interaction.

Keywords: Foundations of Sensing and Estimation, Range, Sonar, GPS and Inertial Sensing, Modeling, Identification, Calibration
Abstract: Olfactory navigation is one of the most primitive mechanisms of exploration used by organisms. Navigation by machine olfaction (artificial smell) is a very difficult task to both simulate and solve. With this work, we define olfactory inertial odometry (OIO), a framework for using inertial kinematics, and fast-sampling olfaction sensors to enable navigation by scent analogous to visual inertial odometry (VIO). We establish how principles from SLAM and VIO can be extrapolated to olfaction to enable real-world robotic tasks. We demonstrate OIO with three different odour localization algorithms on a real 5-DoF robot arm over an odour-tracking scenario that resembles real applications in agriculture and food quality control. Our results indicate success in establishing a baseline framework for OIO from which other research in olfactory navigation can build, and we note performance enhancements that can be made to address more complex tasks in the future

Keywords: Haptics, Soft Robotics, Actuation and Actuators
Abstract: High-spatial-resolution wearable tactile arrays have drawn interest from both industry and research, thanks to their capacity for delivering detailed tactile sensations. However, investigations of human tactile perception with high-resolution tactile displays remain limited, primarily due to the high costs of multi-channel control systems and the complex fabrication required for fingertip-sized actuators. In this work, we introduce the Soft Haptic Display (SHD) toolkit, designed to enable students and researchers from diverse technical backgrounds to explore high-density tactile feedback in extended reality (XR), robotic teleoperation, braille displays, navigation aid, MR-compatible somatosensory stimulation, and remote palpation. The toolkit provides a rapid prototyping approach and real-time wireless control for a low-cost, 4×4 soft wearable fingertip tactile display with a spatial resolution of 4 mm. We characterized the display’s performance with a maximum vertical displacement of 1.8 mm, a rise time of 0.25 second, and a maximum refresh rate of 8 Hz. All materials and code are open-sourced to foster broader human tactile perception research of high-resolution haptic displays.

Keywords: Haptics, Telerobotics, Robotic Hands
Abstract: Haptic feedback is pivotal to telemanipulation; it equips human operators with intuitive physical information about a distant environment. Anthropomorphic robotic hands demonstrate a similar aptitude for complex remote manipulation. Past efforts to enrich robotic hands with haptic feedback often required prohibitively expensive specialized hardware (>50,000 USD grippers and >5,000 USD gloves) that only displayed unidirectional force feedback. In this work, we present the Hand-in-Hand (HiH): an inexpensive system to realize hand telepresence with 3D force feedback via bilateral robot control. By transforming an inexpensive robotic hand (<500 USD each) into a wearable haptic exoskeleton, users can intuitively control and feel what a distant physical agent touches in real-time. Experimentation reveals: an average motion latency of 63 ms over WLAN; an average joint position tracking RMSE of 3.73 deg; and the force feedback magnitude peaking at ~6 N. Safety is passively ensured via sacrificial exoskeleton parts that prevent excessive loads from harming wearers. Issues regarding stability and transparency are partially addressed using saturated virtual friction. Nonetheless, the HiH presents a novel, intuitive, and low-cost approach to haptic telemanipulation with humanoid robotic hands.

Keywords: Modeling, Identification, Calibration, Multisensor Data Fusion, Wheeled Mobile Robots
Abstract: Accurate LiDAR-camera extrinsic calibration is essential for perception in autonomous ground vehicles, but even a precisely known initial transformation can drift over time due to minor collisions or mechanical shifts. In this paper, we present an online calibration method that detects and corrects such drift without relying on special targets. Our approach continuously checks the existing extrinsic parameters by comparing newly estimated ground-plane parameters and edge-based reprojection errors. Whenever the reprojection error exceeds a preset threshold, a joint optimization refines the LiDAR-camera transform by incorporating both plane and edge constraints. Experimental validation in a simulation environment shows that the proposed framework detects small extrinsic misalignments promptly and effectively restores accurate sensor fusion over extended operation.

Keywords: AI Reasoning Methods for Robotics, Robotic Systems Architectures and Programming, Manipulation Planning and Control
Abstract: Advancements in task planning and human-robot collaboration are driving innovation in robotics. The emergence of sophisticated artificial intelligence (AI), particularly large language models (LLMs), presents significant opportunities for enhanced robotic autonomy and flexible collaboration. In this work, we propose a system that leverages an LLM to interpret natural language prompts and generate task plans. This process integrates environmental data from a vision-language model (VLM) and utilizes an action-function library defining the robot’s capabilities. In addition, we develop an intuitive graphical user interface (GUI) that not only connects to the AI task planner, but allows for user oversight throughout the planning and execution process. To validate our approach, we conducted experiments using a dual-arm robotic system to perform a complex, multi-step task: installing a wire onto a power supply. Our system demonstrates significant potential for improving task flexibility and adaptability in human-robot collaborative settings. These findings pave the way for more autonomous and versatile robotic systems in industrial and collaborative applications.

Keywords: Robotic Hands, Rehabilitation and Healthcare Robotics, Mechanism and Design
Abstract: This paper presents a novel approach to the design and fabrication of powered prosthetic hands using Fused Deposition Modeling (FDM) enhanced by 4D printing principles. To overcome limitations in conventional methods, such as high part count, weight, and assembly complexity, and inherent drawbacks of FDM, such as large joint clearances and anisotropic strength, the study introduces thermally responsive self-adaptive mechanisms after printing. Two key mechanisms are proposed: a self-tightening RCJ that minimizes joint clearance and a self-aligning bending/twisting unit that modifies print orientation via controlled post-print deformation. These mechanisms are integrated into a fully 3D-printed prosthetic hand, eliminating the need for assembly. Experimental validation demonstrates improved mechanical precision and structural adaptability, highlighting the potential of this strategy to enable low-cost, customizable, and functionally robust prosthetic devices through single-step

Keywords: Performance Evaluation and Optimization, Rehabilitation and Healthcare Robotics
Abstract: Care robots are defined as robots or devices that use robot technology to provide physical and mental assistance to the elderly or disabled who have difficulty maintaining their daily lives. Currently, care robots are being developed for the purpose of providing various daily life assistance to care recipients (disabled people including industrial accident disabled people, elderly people, etc.) and caregivers (caregivers, family members, etc.). From 2019 to 2021, our research institute conducted safety-related research projects on four types of care robots (transfer, defecation, bedsores and posture change, meals) as part of a Ministry of Health and Welfare project, and the projects were successfully completed. However, as time passes and technological development advances, the products need to be improved. In addition, five types of care robots (indoor movement, bathing assistance, flexible wearable, communication, and integrated monitoring) have been added for projects to be conducted from 2023, and research and development has begun. Therefore, it is time to improve existing products and evaluate additionally developed products. In this study, we derive performance test items and apply test methods for performance evaluation of care robots, so that it can be used as a standard to confirm the performance and safety of the functions of care robots. Performance test items are intended to provide test results by setting evaluation criteria, including the functions of each robot and the intended use scenario.Companies developing care robots will be able to derive product improvement points through company feedback through the results of this study. Since the test items were applied through pilot tests at the current prototype stage, not all performances of actual commercialized products or care robots with new functions can be covered in this study. However, we believe that it can be used as a good reference for developing performance test methods for the relevant performance.

Keywords: Industrial Robots, Robotic Systems Architectures and Programming, Contact: Modeling, Sensing and Control
Abstract: To autonomously collect soil in uncultivated terrain, robotic arms must distinguish between different granular materials and penetrate the correct material. We develop a prototype that collects soil in heterogeneous terrain. If mounted to a mobile robot, it can be used to perform soil collection and analysis without human intervention. Unique among soil sampling robots, we use a general-purpose robotic arm rather than a soil core sampler.

Keywords: Robot Surveillance and Security, Robotics in Hazardous Applications, Legged Robots
Abstract: Infrastructure inspection is becoming increasingly relevant in the field of robotics due to its significant impact on ensuring workers’ safety. The harbor environment presents various challenges in designing a robotic solution for inspection, given the complexity of daily operations. This work introduces an initial phase to identify critical areas within the port environment. Following this, a preliminary solution using a quadruped robot for inspecting these critical areas is analyzed.

Keywords: Multi-Robot Systems, Wheeled Mobile Robots
Abstract: Task allocation and path planning are critical challenges in coordinating heterogeneous multi-robot systems for agricultural applications. This research focuses on automating lavender harvesting, where robots with varying capabilities must collaboratively navigate complex field layouts to efficiently complete harvesting tasks.

We propose two heuristic approaches to address the specific problem of multi-robot coordination for lavender harvesting. The first approach utilizes a Large Language Model (LLM) to generate harvesting plans. By providing the LLM with small-scale examples and iteratively refining prompts with detailed descriptions of task attributes, robot capabilities, and environmental conditions, it produces feasible task allocations and paths for each robot while minimizing overall operational time. The second approach employs a greedy heuristic algorithm, which starts with an initial feasible solution and iteratively improves it by optimizing task allocation and robot paths while ensuring all constraints are satisfied. This method guarantees feasibility by directly incorporating task requirements and robot constraints into its optimization process.

Both approaches are validated through simulations of lavender fields with varying sizes, layouts, and numbers of robots. The results demonstrate the effectiveness of these methods in achieving efficient task allocation and path planning for heterogeneous multi-robot systems. This work highlights the potential for these approaches to advance automation in lavender harvesting, contributing to increased efficiency and sustainability in agricultural operations, further considering fuel and payload constraints as well as required harvesting techniques.

Keywords: Humanoids, Mechanism and Design, Wheeled Mobile Robots
Abstract: Balloid, a compact ball-shaped humanoid robot for teaching robotics to middle/high school students, is introduced in this study. With its hybrid design, it can switch between walking on its two legs and driving using its separately-driven wheels, allowing it to navigate a variety of surfaces better than today’s educational robots. This paper covers Balloid's mechanics and control system. The mechanical design and building section describes the construction of Balloid's legs and shoulder-mounted wheels. The control system development section explains the developed software that uses trigonometry to calculate Balloid’s lower body actions and uses the differential drive control to make its upper body movements. In addition, experimental results for standing and driving are presented. Future plans include exploring a momentum-based rolling mode for better energy efficiency. This project aims to provide students with hands-on experience that bridges the gap between simple wheeled robots and humanoids. Balloid’s mechanical design and construction details will be shared openly for people to use and modify free for STEM education.

Keywords: Contact: Modeling, Sensing and Control, Wheeled Mobile Robots
Abstract: In this study, we propose a floor material classification method under driving motion using a piezoelectric actuator– sensor pair. The piezoelectric pair consists of an actuator and a sensor. When the pair contacted to the floor while driving motion, the actuator was operated and the sensor collected signals from the floor simultaneously. The collected signals were preprocessed to change it as input data of machine learning. With this method, we classified six floor materials and one no contact situation as a high accuracy of 95.4%.

Keywords: Autonoums Vehicle Navigation, Physical and Cognitive Human-Robot Interaction, Dynamics and Control
Abstract: This study introduces Memory-Augmented Model Predictive Controller (MAMPC), which enhances safe navigation in cluttered environments during human-following scenario by utilizing a buffer of previous optimal control inputs and their associated costs. By strategically reusing partial solutions and refining critical segments of the prediction horizon in real time, MAMPC demonstrates superior obstacle anticipation and collision avoidance compared to traditional Model Predictive Controllers across various complex scenes.

Keywords: Actuation and Actuators, Legged Robots, Mechanism and Design
Abstract: This study proposes a Surface-mounted Permanent Magnet Synchronous Machine (SPMSM) design strategy for applying Field Weakening Control (FWC), a technique commonly used in electric vehicles, to actuators in quadruped robots. By extending the motor's speed range, FWC enables higher locomotion speeds without exceeding voltage limits of battery. We analyze the key motor characteristics required for effective FWC implementation in legged robot actuators and validate our approach through RAISIM simulations with reinforcement learning (RL). The results demonstrate that FWC can significantly enhance maximum locomotion speed of HOUND2 from 11.0 m/s to 14.0 m/s in simulation, providing a foundation for high-performance legged robot actuation.

Keywords: Haptics
Abstract: Although research into the use of shape displays as 3D design tools exists, the most appropriate and versatile haptic technologies for such a task are costly and require a high level of peripheral electronics that act as a barrier to scalability, making their integration into appropriate settings difficult. Additionally, the resultant niche nature of this technology means that the different tools and methods for conveying 3D information using haptic feedback are largely unrealized and under-reviewed. This work presents the results of the first phase of a broader set of works concerning surface haptic displays. We evaluated a 'tactile height map' method of conveying 3D information between two different pin array configurations, consisting of low- and high-density arrays.

A user study exploring how users perform when assembling 3D objects using this 'tactile height map' method and these pin arrays found that pin density resulted in a significant difference in the time it took to reassemble a 3D model; however, it did not significantly affect assembly accuracy. These results can inform how future commercial surface displays can be leveraged to support complex design tasks, such as 3D modeling, effectively. Our future research adds a rendering method and expands the number of haptic surface displays to determine how 3D information can best be conveyed using tactile feedback as the main feedback mechanism.

Keywords: Robot Surveillance and Security, Robotics in Hazardous Applications, Search and Rescue Robotics
Abstract: An end-to-end learning approach for snake robot locomotion using deep reinforcement learning is proposed in this paper. The reward functions tailored to each gait of a snake robot were designed by leveraging the natural locomotion patterns of snakes. The comparative analysis between a reinforcement learn-ing-based control and conventional cyclic control was discussed in this research.

Keywords: Mechanism and Design, Wheeled Mobile Robots
Abstract: This work-in-progress extended abstract introduces CLAW II, a novel wheel-less legged mechanism that maintains smooth-rolling motion and obstacle-climbing capabilities without a conventional wheel. Building on CLAW I, which integrated a circular wheel with leg segments, CLAW II eliminates the wheel entirely, relying solely on the curved leg geometry for continuous rolling-like motion while reducing weight and mechanical complexity. To validate CLAW II, we are developing a mobile robot equipped with CLAW II mechanisms. This robot will be tested for obstacle climbing, multi-terrain mobility, and seamless rolling motions.

Keywords: Biomimetic and Bioinspired Robots, Dynamics and Control, Legged Robots
Abstract: Wall-climbing robots face significant challenges in maintaining stability during climbing. This study presents a bio-inspired tail mechanism that enables a quadrupedal robot to self-right instead of falling when encountering instability. Inspired by the biomechanics of lizards, the proposed mechanism leverages dynamic tail actuation to reorient the robot’s body against vertical surfaces to prevent it from falling. In this work, the mechanism for this tail and the inverse kinematics calculations for position control are discussed.

Keywords: Dynamics and Control, Modular Robots
Abstract: This paper introduces the dynamic rolling locomotion of the Variable Topology Truss (VTT) system. While existing research has explored motion planning and control of truss systems, including our previous work on rolling locomotion and path planning, these studies primarily focused on quasi-static motion - a methodology that inherently limits locomotion speed and efficiency. We are developing a rolling locomotion method that can maintain the VTT system's momentum during locomotion. A preliminary version of the rolling algorithm was tested in a simulation environment and the results were analyzed. Our findings establish the foundation for enhancing the locomotion capabilities of the VTT system and provide critical insights for future hardware implementation.

Keywords: Multi-Robot Systems, Range, Sonar, GPS and Inertial Sensing, Aerial and Flying Robots
Abstract: External factors, including urban canyons and adversarial interference, can lead to Global Positioning System (GPS) inaccuracies that vary as a function of the position in the environment. This study addresses the challenge of estimating a static, spatially-varying error function using a team of robots. The central idea is to use sensed estimates of the range and bearing to the other robots in the team to estimate changes in bias across the environment. This abstract describes a work-in-progress algorithm for this problem that uses a quadratic optimization formulation to find a self-consistent set of pointwise bias estimates, followed by a Gaussian Process Regression (GPR) to form a bias map estimate across the full environment. We also describe an approach that uses informative path planning techniques to plan movements for the robots to improve the accuracy of these estimates. Preliminary results in simulation show the promise of the approach.

Keywords: Search and Rescue Robotics, Aerial and Flying Robots, Deep Learning for Visual Percepton
Abstract: This extended abstract presents a deep learn- ing–guided fire extinguishing system aimed at mitigating wild- fires under adverse conditions such as strong winds, nighttime, drought, and smoke. The system combines a YOLO-based object detection algorithm with a unique Nona Filter to enable real-time recognition and priority-based target tracking of fires and smoke. In controlled experiments, the GFED system successfully identified and tracked fire sources at distances up to 30 meters, maintaining consistent performance even when multiple fire and smoke instances appeared in the same frame. The current work will target aerial deployment under challenging environments like strong winds, nighttime, high altitude operations. Additional enhancements include compu- tational fluid dynamics, 6-degree-of-freedom analysis, sensor integration, and further optimization of the Nona Filter. With continued development, GFED shows strong potential to evolve into a fully autonomous wildfire suppression solution.

Keywords: Human-Robot Augmentation
Abstract: Introductory robotics courses often rely on theoretical lectures, occasionally supplemented by physical labs that offer hands-on experience. However, such labs are costly and resource-intensive, making them impractical for many classroom settings. To overcome these limitations, we introduce RAISE (Robotics with Augmented Instruction for Student Engagement), an Augmented Reality (AR) application designed for standard mobile devices. RAISE overlays 3D robot models onto real-world environments, enabling students to interactively explore core robotics concepts--such as rigid-body motion and forward kinematics--while visualizing coordinate frames and screw axes in real time. By leveraging AR, the platform aims to enhance conceptual understanding and engagement beyond traditional methods without cost and resources required for physical labs. Future work will evaluate its educational impact by comparing RAISE-enhanced instruction with conventional lecture-based approaches using a range of learning metrics. Planned technical updates include support for additional robot models, user-designed assemblies, advanced analytics, and expanded coverage of more complex robotics topics—positioning RAISE as a comprehensive, accessible, and adaptable educational tool.

Keywords: Underwater Robotics
Abstract: Relying on RGB cameras for underwater robotics presents challenges, particularly due to varying light conditions that reduce image processing effectiveness. This work-in-progress paper explores image focus detection methods designed for dynamic underwater environments. We use the Laplacian operator to measure focus and evaluate its effectiveness through lab and underwater experiments with the Reach Alpha 5 manipulator arm. Our goal is to enable underwater robots to dynamically adjust camera positioning for clearer imaging. While effective in many scenarios, the method requires manual intervention due to the lack of standardized thresholds and reliance on raw Laplacian values. Future improvements, such as adaptive thresholding and normalization, could enhance robustness and applicability. This approach lays the foundation for real-time focus optimization in underwater robotics, benefiting autonomous inspection, manipulation, and exploration.

Keywords: Multi-Robot Systems, Robot Surveillance and Security, Multisensor Data Fusion
Abstract: The Multiple Object Trajectory Estimation (MOTE) system, based on multi-hypothesis tracking (MHT), is designed to provide a unified target state estimate for a heterogeneous fleet of autonomous agents, addressing the need for real-time situational awareness and collaborative perception. By utilizing sensor fusion, our perception systems generate observations consisting of object class and 3D positions for Objects of Interest (OOI). These observations are shared between agents via radio communications and ingested by independent instances of MOTE, enabling each agent to maintain multi-target state estimates. The prototype was verified via deployment on a multi-vehicle autonomous fleet, and enabled consistent tracking for a dynamic target across varying fields of view.

Keywords: Multi-Robot Systems, Search and Rescue Robotics
Abstract: Swarm robotics offers scalable and robust solutions for tasks such as foraging, yet congestion near central collection zones remains a critical challenge, especially with increasing swarm sizes. Traditional solutions, such as static path planning or local repulsion-based methods, often fail to prevent inter-robot collisions or bottlenecks near the collection zones. This research presents a comparative study of three strategies to mitigate congestion when returning resources to the central collection zone. The research herein focuses on tightly packed environments where, in theory, robots should follow a pre-planned spiral, either square or circular, with congestion detection as described in the first two strategies. The third strategy introduces an adaptive path that allows robots to make reactive movements in response to congestion. Furthermore, we explore the use of a deep reinforcement learning (DRL) approach that trains policies in a centralized manner but executes them in a decentralized fashion, thereby preserving swarm robotic principles. Both spiral strategies integrate multiple entry points into the central collection zone and dynamic re-routing upon congestion detection. Experimental evaluation in the ARGoS physics-based simulation environment demonstrates significant improvements in task completion time, collision reduction, and system throughput. These results indicate that structured congestion-aware trajectories can significantly improve swarm foraging efficiency.

Keywords: Wheeled Mobile Robots, Multisensor Data Fusion, Mechanism and Design
Abstract: Accurate weed management is essential for mitigating significant crop yield losses, necessitating effective weed suppression strategies in agricultural systems. Integrating cover crops (CC) offers multiple benefits, including soil erosion reduction, weed suppression, decreased nitrogen requirements, and enhanced carbon sequestration, all of which are closely tied to the aboveground biomass (AGB) they produce. However, biomass production varies significantly due to microsite variability, making accurate estimation and mapping essential for identifying zones of poor weed suppression and optimizing targeted management strategies. To address this challenge, developing a comprehensive CC map, including its AGB distribution, will enable informed decision-making regarding weed control methods and optimal application rates. Manual visual inspection is impractical and labor-intensive, especially given the extensive field size and the wide diversity and variation of weed species and sizes. In this context, optical imagery and Light Detection and Ranging (LiDAR) data are two prominent sources with unique characteristics that enhance AGB estimation. This study introduces a ground robot-mounted multimodal sensor system designed for agricultural field AGB mapping. The system integrates optical and LiDAR data, leveraging machine learning methods for data fusion to improve biomass predictions. The best machine learning-based model for dry AGB estimation achieved an R^2 of 0.88, demonstrating robust performance in diverse field conditions. This approach offers valuable insights for site-specific management, enabling precise weed suppression strategies and promoting sustainable farming practices. The integration of high-resolution optical and LiDAR data from a robot-mounted system, combined with machine learning techniques, establishes a scalable framework for automated biomass estimation in large-scale agricultural field conditions, enhancing decision-making in precision agriculture.

Keywords: Robot Surveillance and Security, AI Reasoning Methods for Robotics, Industrial Robots
Abstract: Analog gauges remain widely used in industrial facilities, requiring routine manual monitoring that increases workload and exposes workers to hazardous environments. Recently, mobile robots have been increasingly deployed for automated surveillance tasks, reducing human intervention in hazardous environments. Traditional gauge reading methods rely on classical computer vision or deep learning models, but they face limitations such as sensitivity to lighting conditions and high data collection costs. To address these challenges, this study proposes a gauge reading approach utilizing a Multimodal Large Language Model (MLLM) combined with Chain-of-Thought (CoT) reasoning to improve accuracy without requiring extensive training data. Preliminary experimental results show that the CoT-based model achieves high accuracy in recognizing gauge panel elements such as unit labels and major markings but exhibits lower performance in needle position detection and final value estimation. These findings highlight both the strengths and limitations of CoT-based approaches, emphasizing the need for improved accuracy in needle position detection as a key focus for future research.

Keywords: Aerial and Flying Robots, Biomimetic and Bioinspired Robots, Mechanism and Design
Abstract: Exterior wall tasks (e.g., inspection, cleaning, and painting) are still predominantly performed manually, leading to significant risks and high costs due to the need for additional equipment. The growing construction of high-rise buildings, bridges, and irregularly shaped structures, along with the increasing use of diverse materials such as glass and metal, has further escalated the complexity and risks associated with exterior wall operations. Existing wall-climbing robots have been developed to address these challenges; however, they often rely on magnetic, vacuum, or pneumatic systems, which suffer from limitations such as material dependency, low energy efficiency, slow mobility, and difficulty in overcoming obstacles. To overcome these constraints, this paper presents a hybrid wall-climbing robot platform that integrates the rapid maneuverability of drones with the stability of six-legged walking robots. By leveraging the thrust of drone propellers and the contact forces of robotic legs, the proposed system achieves stable and energy-efficient adhesion and movement on walls, regardless of the surface material. Inspired by the perching and take-off behaviors of insects, the robot operates efficiently without the need for advanced control algorithms or high computational resources. The developed robot has been experimentally validated for its performance on walls with various materials and shapes, demonstrating key capabilities such as stable adhesion, efficient walking speed, and optimized energy consumption. Furthermore, it has been tested in real-world environments, demonstrating its potential for practical deployment. This technology is expected to significantly reduce operational costs and accident risks, providing substantial socio-economic benefits for exterior wall applications.

Keywords: Physical and Cognitive Human-Robot Interaction, Human-Robot Augmentation
Abstract: This paper presents a novel framework and describes initial progress towards designing and simulating Augmented Reality (AR) assisted Human-Robot Interaction (HRI) in industrial environments using Virtual Reality (VR). While AR has shown promise for robotic control systems, its potential benefits for non-operator stakeholders in construction and manufacturing remain largely unexplored. We propose a Virtual Reality (VR) simulation environment that enables the design and testing of AR visualizations before physical implementation. Our system provides a platform for creating context-aware AR visualizations that communicate robot intentions, such as movement previews and operational boundaries, to improve situational awareness and safety for workers in shared spaces. The framework aims to contribute to the advancement of ubiquitous robotics by bridging the gap between robotics engineers and end-users through intuitive visual communication systems.

Keywords: Dynamics and Control, Aerial and Flying Robots, Modeling, Identification, Calibration
Abstract: As the presence of autonomous drones in civilian operations continues to grow, ensuring their safe operation is crucial, as failures can lead to loss of control, system damage, property destruction, environmental harm, and even human injury. On the other hand, the inherently unstable and underactuated dynamics of quadrotors make them particularly vulnerable to system faults, especially rotor failures. In this paper, we introduce a fault-tolerant control strategy using a run time safety assurance filter based on model predictive control (MPC) to provide safety guarantees to the control input of a linear quadratic regulator (LQR) designed with the purpose of trajectory following. Our method incorporates a real-time fault detection and isolation system while backup trajectories are created for different fault modes. We demonstrate the performance of our proposed approach in a 3D simulation environment using a model of the Crazyflie 2.0 drone.

Keywords: Multisensor Data Fusion, Object Recognition, Autonoums Vehicle Navigation
Abstract: With the rapid advancement of probabilistic inference methods, diverse artificial intelligence technologies, and high-performance computing hardware, technologies related to autonomous navigation have achieved considerable development. Compared to other types of vehicles, ships operate in environments with significant variability and must sustain long-duration missions at sea. Consequently, maritime situational awareness systems for detecting objects around the vessel must demonstrate high performance and reliability, while also ensuring cost-effectiveness in terms of system development and maintenance. This paper proposes a multi-object tracking system designed for autonomous ships, taking into account their unique operational characteristics. The system is composed of multiple detection sensors and navigation sensors, and features an AI-based detection algorithm integrated with a probabilistic data fusion architecture. The structure consists of two processing stages based on the purpose of data handling, and it is designed with scalability in mind. The performance of the proposed architecture and algorithm is demonstrated through two types of experimental results. Furthermore, this paper identifies the limitations of relying solely on situational awareness systems for commercial operations and underscores the inevitability of introducing a systematic and standardized method for generating and sharing positional information in autonomous ships. Taking insights from structured environments in robotics, we suggest that this approach offers a practical pathway toward achieving both economic viability and safety, thereby accelerating the commercialization of autonomous ships given the current level of technological maturity.

Keywords: Legged Robots, World Modelling, Contact: Modeling, Sensing and Control
Abstract: We present a soft contact model to simulate diverse soft terrains, enabling robust legged locomotion through reinforcement learning. The model extends a standard spring-damper formulation with Stribeck-Coulomb friction and introduces randomized parameters, such as stiffness, damping, and friction coefficients, to capture the variability of real-world soft surfaces, including soil and mattresses. By replacing the default contact model in the simulator with our formulation, we train a locomotion policy using an existing learning framework. The resulting policy demonstrates stable walking on both flat and inclined soft terrains with the Unitree Go1 robot in simulation. Notably, it generalizes to rigid ground without explicit training, highlighting improved robustness across contact conditions. This work offers a lightweight and flexible alternative to high-fidelity contact modeling for scalable locomotion training.

Keywords: Performance Evaluation and Optimization
Abstract: This paper presents the Probability-Based Manipulability Score (PBMS), a new metric for comparing articulated manipulators with different degrees of freedom. PBMS uses a log-scaled score in a voxelized workspace to capture the effects of kinematic redundancy. This overcomes the upper-bound limitations of conventional indicators, which constrain performance index even when the degrees of freedom increase, and enables comparison across manipulators with different degrees of freedom. Simulation was performed comparing TOCABI's arm with a test manipulator in a common workspace to validate the approach. The results demonstrate that PBMS can effectively guide the task-specific design of redundant manipulators.

Keywords: AI Reasoning Methods for Robotics, Industrial Robots, Motion Planning and Obstacle Avoidance
Abstract: The Online Bin Packing Problem (OBPP) aims to determine the optimal loading position for each incoming item to maximize bin utilization, a critical challenge in various industrial applications. While many studies have focused on learning-based policies and heuristic approaches to enhance packing efficiency, stability constraints have largely been overlooked. In this work, we propose a computationally efficient method to validate stable loading positions for incoming items without requiring exact knowledge of their physical properties, such as mass. Our approach leverages the concept of Load-Bearable Convex Polygons (LBCPs), which provide substantial support forces to ensure structural stability. We further integrate our static stability validation framework into a state-of-the-art deep reinforcement learning (DRL) model, guiding it to learn physics feasible packing strategies. Experimental results demonstrate that our stability-aware DRL model achieves comparable packing efficiency while ensuring robust bin stability, offering a significant advancement in practical OBPP applications.

Keywords: AI Reasoning Methods for Robotics, Performance Evaluation and Optimization
Abstract: Robots and intelligent assistants are increasingly performing tasks autonomously in household settings. While navigation-based tasks are straightforward, open-ended tasks require reasoning on the basis of commonsense knowledge. Towards fostering the development of systems that can use and reason on commonsense knowledge to tackle open-ended tasks, we propose RoboCSKBench, a natural language-based multi-task benchmark to assess embodied commonsense knowledge capabilities of agents and systems interacting in dynamic household environments. Our benchmark combines various resources (e.g. knowledge graphs, manipulation benchmarks, crowdsourcing) to provide data for five different, commonly encountered household tasks: Tidy Up, Tool Usage, Meta-Reasoning, Table Setting and Procedural Knowledge. Each task comprises of data and evaluation metrics supporting the evaluation of systems inducing embodied commonsense knowledge. While the benchmark consists of five tasks at the time of writing, it can be extended with further tasks in the future.

Building on the benchmark, we assess the capabilities of three state-of-the-art large language models on the various tasks of the benchmark. Our results indicate a diverse foundation, with model performance varying across different tasks, suggesting that no single model clearly outperforms the others. On the contrary, all models exhibit limitations, leaving room for further optimization and improvement.

Keywords: AI Reasoning Methods for Robotics, Performance Evaluation and Optimization, Robotic Systems Architectures and Programming
Abstract: Actor-critic methods, like Twin Delayed Deep Deterministic Policy Gradient (TD3), depend on basic noise-based exploration, which can result in less than optimal policy convergence. In this study, we introduce Monte Carlo Beam Search (MCBS), a new hybrid method that combines beam search and Monte Carlo rollouts with TD3 to improve exploration and action selection. MCBS produces several candidate actions around the policy’s output and assesses them through short-horizon rollouts, enabling the agent to make better-informed choices. We test MCBS across various continuous-control benchmarks, including HalfCheetah-v4, Walker2d-v5, and Swimmer-v5, showing enhanced sample efficiency and performance compared to standard TD3 and other baseline methods like SAC, PPO, and A2C. Our findings emphasize MCBS’s capability to enhance policy learning through structured look-ahead search while ensuring computational efficiency. Additionally, we offer a detailed analysis of crucial hyperparameters, such as beam width and rollout depth, and explore adaptive strategies to optimize MCBS for complex control tasks. Our method shows a higher convergence rate across different environments compared to TD3, SAC, PPO, and A2C. For instance, we achieved 90% of the maximum achievable reward within around 200 thousand timesteps compared to 400 thousand timesteps for the second-best method.

Keywords: Deep Learning for Visual Percepton, Computer Vision and Visual Servoing
Abstract: We present an improved pixelNeRF adaptation that integrates depth features, image global features, and the SIREN module to improve novel view synthesis (NVS) in sparse-view settings. While pixelNeRF effectively reconstructs scenes from limited views, it struggles with blurriness and insufficient high-frequency details, particularly in occluded regions. To address this, depth features provide structural guidance, global features enhance consistency in sparsely observed areas, and SIREN improves fine-grained detail capture. Experiments on ILSH and DTU datasets show that the proposed method reduces blurriness and improves occlusion reconstruction, validated through qualitative and quantitative evaluations. Future work should refine feature aggregation to better handle complex environments, demonstrating the potential of multi-scale feature integration for NVS in sparse-view scenarios.

Keywords: Deep Learning for Visual Percepton, Humanoids, Object Recognition
Abstract: Humanoid robots competing in RoboCup (an inter- national robot soccer competition) must perceive their environments under highly dynamic and often unpredictable conditions. Requiring a vision system for localization and path planning, teams typically need to collect training image data for the machine learning object detection model. This paper presents an empirical comparison of three commonly used methods - handheld camera, gimbal-mounted systems, and rollable tripods. Experimental results show that the gimbal-mounted approach consistently outperforms the other two, yielding superior precision and recall when detecting crucial soccer field landmarks and the game ball. These results highlight that data collection methods which effectively simulate the robot’s actual visual experience during gameplay lead to more robust and reliable vision models. Inspired by these findings, we implemented a revised vision pipeline in our latest humanoid robot, ARTEMIS, capturing data directly from its onboard stereo camera system. This approach proved instrumental in achieving reliable object detection in real-time, even under severe motion blur and degraded field conditions during the dynamic matches, resulting in our eventual victory. We discuss the advantage and limitations of each data collection method, emphasizing the critical role of matching the robot’s real-world visual experience to achieve champion-level performance.

Keywords: Deep Learning for Visual Percepton, Multisensor Data Fusion, Object Recognition
Abstract: This paper proposes a deep learning-based approach for pallet detection and pose estimation using RGB-D sensors, addressing key challenges in forklift-operated logistics and manufacturing environments. Existing research often faces limitations such as environmental constraints and data scarcity. To overcome these issues, our method combines synthetic data generation with Diffusion model-based data augmentation techniques, generating diverse pallet datasets with varying sizes and shapes using advanced 3D simulation tools. A modified YOLOv11 network is introduced to detect pallet bounding boxes and estimate the center and corner points of cuboids. The network is trained on the generated data and evaluated using real-world RGB-D data in real-time. The proposed approach significantly improves the precision of pallet detection and pose estimation in complex environments, contributing to logistics automation and offering broader implications for various industrial applications.

Keywords: Motion Planning and Obstacle Avoidance, Robotic Systems Architectures and Programming, AI Reasoning Methods for Robotics
Abstract: This paper presents a novel framework for memory-based navigation for terrestrial robots, utilizing a customized multimodal large language model (MLLM) to interpret visual inputs and generate navigation commands. The system employs a Unitree GO1 robot equipped with a camera to capture environmental images, which are processed by the customized MLLM for navigation. By leveraging a memory-based approach, the robot efficiently reuses previously traversed paths, reducing the need for re-exploration and enhancing navigation efficiency. The hybrid controller in this work features a deliberation unit and a reactive controller for high-level commands and robot alignment. Experimental validation in a hallway-like environment demonstrates that memory-driven navigation improves path retracing and overall performance.

Keywords: Autonoums Vehicle Navigation
Abstract: This paper presents the design and development of a low-maintenance, modular lunar rover for long-term service and maintenance operations on the Moon. The rover is engineered for extended durability (~1 year) low-backlash cycloidal drive systems, and solar-resistant materials to minimize wear and maintenance. Key features include a 6- degree-of-freedom (DOF) robotic arm capable of lifting 50 kg (81 N under lunar gravity). To ensure autonomous operation in a dynamic lunar environment, the rover integrates advanced sensors, including a Zed Mini Camera, Micro Lidar sensors, and IMU modules, controlled by a Jetson Nano-based system. Autonomous navigation and payload manipulation are enabled through computer vision models trained on convolutional neural networks (CNNs), with PID-controlled dynamic adjustments. The rover is powered by a Lithium Iron Phosphate battery, allowing for hot-swappable operation and extended activity. Additionally, Gazebo simulations will be used to refine control algorithms before deployment. This design aims to enhance long- term operational efficiency and reduce logistical costs for sustained lunar exploration.

Keywords: Autonoums Vehicle Navigation, Deep Learning for Visual Percepton, Computer Vision and Visual Servoing
Abstract: A critical task for autonomous vehicles is not only detecting surrounding objects but also predicting their orientation and heading, particularly for nearby vehicles. Understanding the direction a neighboring vehicle is facing and its potential trajectory enables autonomous systems to make informed navigation decisions, avoid collisions, and operate safely in traffic. This paper develops and evaluates two passive-sensing methods that leverage machine learning to predict vehicle orientation from two-dimensional (2D) images. The first approach employs deep learning to classify a vehicle’s orientation into one of eight general directions. The second method utilizes a cascaded approach with object detection, bounding box area analysis, and regression to predict a more precise orientation of the vehicle. A dataset of 1,424 labeled images, each annotated with the relative heading difference of the distant vehicle with respect to the observing vehicle, was collected and used in both approaches. The findings of this research indicate that the second, cascaded approach is particularly effective, achieving a Mean Absolute Error of 5.06 degrees, demonstrating its potential for robust vehicle tracking applications.

Keywords: Autonoums Vehicle Navigation, Industrial Robots, Wheeled Mobile Robots
Abstract: This paper presents a robust and precise Autonomous Mobile Robot (AMR) system tailored for small to medium-sized warehouses, specifically addressing the challenges of navigating misplaced racks in constrained environments. We developed a compact AMR capable of effective operation in spatially limited environments, optimizing it for the smaller racks commonly found in these warehouses. The key contribution is the development of a precise rack center navigation algorithm using 2D LiDAR, allowing the AMR to detect rack legs, navigate between them, and accurately position itself for collision-free and efficient lifting operations, even when racks are misaligned. Experimental validation in real warehouse settings demonstrates the system's superior performance in terms of path stability, error tolerance, and adaptability, proving its potential to significantly enhance logistics automation in small to medium-sized warehouses.

Keywords: Autonoums Vehicle Navigation, Search and Rescue Robotics, Robotic Systems Architectures and Programming
Abstract: Exploring narrow and multi-floor indoor environments presents significant challenges due to their confined spaces and structural complexity. This paper introduces a novel exploration strategy based on Multi-Criteria Decision-Making (MCDM) to address these challenges effectively. The proposed algorithm dynamically manages exploration coverage and utilizes ray-casting techniques tailored to the size of the environment to identify exploration candidates efficiently. Additionally, it incorporates a robust staircase detection and traversal mechanism using 3D LiDAR sensors, enabling seamless exploration across multiple floors. Experimental validation in real-world maze-like environments demonstrated the algorithm's capability to thoroughly explore confined spaces, detect and overcome staircases, and resume exploration on new floors. The results confirmed the algorithm's effectiveness in achieving comprehensive exploration and robust performance, validated through experiments conducted under diverse and challenging environmental conditions.

Keywords: Autonoums Vehicle Navigation, World Modelling, Foundations of Sensing and Estimation
Abstract: This paper presents a proof of concept for integrating real-time audio classification into autonomous vehicle systems using the open-source autonomous driving simulator, CARLA. With the increasing need for autonomous vehicles to operate safely in their environment, the addition of auditory signal perception (e.g., emergency sirens) can improve navigation in urban settings. Using support vector machines, we developed a binary classification model capable of identifying sirens within the simulated environment, allowing simulated autonomous vehicles to detect and respond to emergency signals. Using CARLA, our open source framework, we can synthesize realistic urban driving scenarios, collecting and processing audio data. Results demonstrate the potential of auditory perception systems in autonomous vehicle development, improving the vehicle’s situational awareness and paving the way for further developments in audio-responsive autonomous driving technology. This research showcases the flexibility of CARLA for auditory simulation and highlights the potential of audio detection to improve autonomous vehicle safety and environmental awareness. To facilitate further research and development in this area, we have made our implementation open-source.

Keywords: Intelligent Robotic Vehicles, Autonoums Vehicle Navigation
Abstract: Autonomous navigation in complex environments, such as underground parking lots, poses significant challenges due to the absence of prior maps and reliance on real-time perception. This paper proposes a comprehensive framework for mapless exploration and navigation, integrating a Semantic Occupancy Grid Network (SoCNet) and a Navigator module. SoCNet predicts local semantic occupancy grids from sensor data, achieving an average pixel accuracy of 0.8234 on test maps. The Navigator constructs and updates a global semantic occupancy grid using a Bayesian approach, incorporating distancebased weighting to mitigate uncertainties in distant predictions. Exploration targets, termed Topology Nodes, are sampled and scored based on proximity and semantic likelihood, guiding the robot via an A* planner. Evaluated in the Isaac Sim environment across multiple trials, the framework successfully explored all spaces in 11 out of 12 trials (91.7% success rate), despite occasional revisits to known areas. Our results demonstrate robust adaptability and efficiency, offering a practical solution for autonomous navigation in unmapped, dynamic settings.

Keywords: Motion Planning and Obstacle Avoidance, Autonoums Vehicle Navigation, Intelligent Robotic Vehicles
Abstract: This paper presents H*, an enhanced A* algorithm designed to improve the realism and efficiency of pathfinding for real-world agents, especially in robotics. H* uses hexagonal grid decomposition and an improved geometric heuristic to model traversable space more effectively. By incorporating the agent’s velocity and turning radius, H* generates smoother, more realistic paths and detects sharp turns tailored to kinematic constraints. These enhancements aim to reduce travel time while maintaining feasibility for physical agents. Experimental results demonstrate the algorithm’s practical benefits in realistic scenarios.

Keywords: Learning from Humans, Cognitive Human Robot Interaction
Abstract: Often, robots are asked to execute primitive movements, whether as a single action or in a series of actions representing a larger, more complex task. These movements can be learned in many ways, but a common one is from demonstrations presented to the robot by a teacher. However, these demonstrations are not always simple movements themselves, and complex demonstrations must be broken down, or segmented, into primitive movements. In this work, we present a parameter-free approach to segmentation using techniques inspired by autocorrelation and cross-correlation from signal processing. In cross-correlation, a representative signal is found in some larger, more complex signal by correlating the representative signal with the larger signal. This same idea can be applied to segmenting robot motion and demonstrations, provided with a representative motion primitive. This results in a fast and accurate segmentation, which does not take any parameters. One of the main contributions of this paper is modification of the cross-correlation process by employing similarity metrics that can capture features specific to robot movements. To validate our framework, we conduct several experiments of complex tasks both in simulation and in real-world. We also evaluate the effectiveness of our segmentation framework by comparing various similarity metrics.

Keywords: Learning from Humans, Grasping, Manipulation Planning and Control
Abstract: Imitation learning is a powerful technique for teaching autonomous agents a variety of tasks. However, many imitation learning algorithms suffer from the fundamental problem of error propagation. Typically, imitation learning is framed as a supervised learning problem; however, during live operation, the agent operates in a state space generated by its own actions instead of expert demonstrations. This co-variate shift can lead to agents either failing to complete designated tasks, taking dangerous actions which can lead to damage or harm, or simply copying expert behavior without completing the task (copycat problem). The problems of co-variate shift and expert copying is especially important in safety-critical environments, such as assistive robotics, where simple errors can bear high costs. While algorithmic solutions exist for this problem, these often rely on constraining the agent policy or simply improving decision making without resolving the issue of error propagation. To address these challenges, we present an efficient solution for resolving error propagation by introducing the tactile modailty in fine-grained grasping and manipulation tasks. To this end, we present Visuotactile Diffusion Policy, a policy learning framework which allows for automated failure recovery. The purpose of this study was to explore an out-of-the-box technique for preventing co-variate shift and the copycat problem, especially in grasping tasks for assistive robotic systems. The primary contribution of this study was to develop a robotic system and policy learning framework capable of automated failure recovery in grasping tasks. Along with this, we demonstrate how tactile sensing can lead to more robust robotic control policies and provide a generalizable solution for co-variate shift in robotic control tasks.

Keywords: Learning from Humans, Physical and Cognitive Human-Robot Interaction
Abstract: To learn manipulation skills, robots need to understand the features of those skills. An easy way for robots to learn is through Learning from Demonstration (LfD), where the robot learns a skill from an expert demonstrator. While the main features of a skill might be captured in one differential coordinate (i.e., Cartesian), they could have meaning in other coordinates. For example, an important feature of a skill may be its shape or velocity profile, which are difficult to discover in Cartesian differential coordinate. In this work, we present a method which enables robots to learn skills from human demonstrations via encoding these skills into various differential coordinates, then determines the importance of each coordinate to reproduce the skill. We also introduce a modified form of Elastic Maps that includes multiple differential coordinates, combining statistical modeling of skills in these differential coordinate spaces. Elastic Maps, which are flexible and fast to compute, allow for the incorporation of several different types of constraints and the use of any number of demonstrations. Additionally, we propose methods for auto-tuning several parameters associated with the modified Elastic Map formulation. We validate our approach in several simulated experiments and a real-world writing task with a UR5e manipulator arm.

Keywords: Learning from Humans, Robotic Hands, Grasping
Abstract: This paper presents an unsupervised deep-learning method for controlling dexterous robotic hands by mimicking human hand motions. We introduce a cross-domain similarity metric to capture the spatial and kinematic relationships between human and robot hands. Using this metric, our approach learns a shared latent space that aligns motion features across the two embodiments. The framework consists of two separate encoders that map human and robot hand data into the latent space, along with a robot decoder that generates feasible robot hand motions. During inference, only the human hand encoder and the robot hand decoder are needed to seamlessly retarget human hand movements to the robot hand, enabling scalable and flexible motion retargeting without requiring paired human-robot data. To demonstrate real-world applicability, we integrate our motion retargeting system with Mediapipe, a human hand pose estimator, enabling real-time robotic hand control from RGB video input.

Keywords: Manipulation Planning and Control, Learning from Humans
Abstract: Putting a lid on a container is a very common and crucial task in daily life. In this paper, we propose a novel learning-free method for robots to `imagine' the matching of unseen open containers and lids via physical simulation. After reconstructing the objects with the Gaussian process distance field, open container imagination is conducted initially to generate the footprint. The footprint is analyzed to determine the relative pose between the container and lid. Then the optimal matching pose is identified by carrying out matching imagination. Experiments were conducted in real-world scenarios. Our method outperforms an LLM-based method, reaching a success rate of 90% when the robot autonomously puts lids on containers. The code is available on our GitHub page.

Keywords: Physical and Cognitive Human-Robot Interaction, Cognitive Human Robot Interaction, Search and Rescue Robotics
Abstract: Advancements in autonomy, navigation, and sensor systems have led to the increased deployment of drones in high-risk applications, such as mapping operations. While drones can mitigate the dangers associated with these missions, human trust in drones is essential for their effective use. This study explores the influence of key factors, including drone appearance, capabilities, protective cage, and noise on human trust. We implemented two different methodologies: (1) an isolated approach, in which the effects of each drone’s appearance and capabilities were studied independently, and (2) a comparative approach, where participants evaluated two drones with different appearances and capabilities in direct comparison. The experiment results indicate that while drone appearance influences human trust, drone capabilities have a significantly greater impact. Additionally, comparing the two methodologies revealed that the comparative approach directs participants’ attention more effectively to the studied factors. One of the primary contributions of this work is the introduction of a tested and effective method to investigate the effects of different factors on trust between humans and drones. Our findings can help robot designers develop drones suited for diverse scenarios by identifying the features most valued by human operators.

Keywords: Soft Robotics, Actuation and Actuators, Biomimetic and Bioinspired Robots
Abstract: This paper presents a motion control strategy for a magnetically actuated rod-like soft robot, enabling it to transition from free space into a tube and overcome structural boundaries. Although soft robots have shown promise in navigating constrained environments, initiating entry into narrow channels and transitioning across sudden changes in geometry, such as the boundary between open space and a confined tube, remains a significant challenge. To address this limitation, we introduce a crawling-based transition mechanism that allows the soft robot to actively engage with the tube entrance, facilitating smooth entry without relying on external guiding structures. We developed a modeling framework to analyze the propulsion dynamics, considering elastic energy storage, frictional interaction, and magnetic actuation. Experiments confirmed that propulsion efficiency depends on the stored elastic energy and how it is released. Our results suggest that controlled oscillatory actuation enables the robot to overcome boundary constraints, which could be an available approach for navigation in confined environments. This work advances magnetically driven soft robotic locomotion, with potential applications in minimally invasive procedures and targeted drug delivery.

Keywords: Industrial Robots, Modular Robots, Grasping
Abstract: Additive manufacturing introduces challenges in quality assurance due to the high variability and volume of produced components. Traditional manual inspection methods, such as measurements taken with calipers, are time consuming, labor intensive, and prone to human error. This paper presents RIPTIDE, a quality confirmation system designed to enhance inspection accuracy and efficiency. RIPTIDE integrates a robotic pick and place mechanism with a scanning procedure to generate high fidelity three dimensional models of manufactured parts. By eliminating human intervention, this system improves consistency, reduces inspection time, and streamlines the validation process for mass volume. The proposed approach demonstrates significant potential in optimizing additive manufacturing workflows by ensuring reliable and scalable quality control.

Keywords: Multisensor Data Fusion, Grasping, Industrial Robots
Abstract: This paper introduces a novel method for precise object grasping point estimation using a palm-type line laser sensor. Active stereo sensors have difficulty in accurately determining object positions and spatial distances, which poses a challenge in robotic grasping. The proposed approach improves object recognition by accurately detecting positions, widths, and spaces, significantly improving the accuracy of contact points and distance measurement between objects. This enables a robot gripper to insert tool tips without collision, enhancing overall operability. Particularly in complex environments, this method substantially improves robotic manipulation capabilities, making it more effective in industrial automation and smart manufacturing.

Keywords: Robotics in Hazardous Applications, Industrial Robots
Abstract: The ability to manufacture spare parts, complete repairs, and carry out other important manufacturing activities is a major concern for users in expeditionary environments (battlefields, remote research stations, or disaster relief areas). The challenges that arise include a limited source of energy, security concerns, an unreliable supply chain, poor local infrastructure, harsh weather, and urgency not typically encountered in regular manufacturing environments. This article developed a conceptual model for the challenges encountered in expeditionary manufacturing, with a focus on applications that use robotic systems to complete or assist in the fabrication. A case study was completed to demonstrate the concepts for a realistic scenario. This work is useful for designers and system planners who wish to use robotic systems (including CNC machines and 3D printers) to support manufacturing activities within an expeditionary environment.

Keywords: Search and Rescue Robotics, Intelligent Robotic Vehicles, Autonoums Vehicle Navigation
Abstract: Efficient evacuation route planning is critical for enhancing emergency response systems in disaster scenarios. Unlike traditional navigation systems that rely on pre-existing data and provide traffic-based routing under normal conditions, our method integrates robotic scouts–comprising drones and ground vehicles–to dynamically assist in evacuation planning during disasters. We propose an effective method for synthesizing evacuation policies that enable robotic scouts to guide evacuees through disaster-affected areas. By leveraging real-world disaster assessment data mapped onto a roadmap, we model the disaster environment and formulate the problem of generating evacuation policies in a stochastic framework using a Markov Decision Process (MDP). Within this framework, we assign location-specific costs on the roadmap based on the degree of structural damage in surrounding areas. Through policy iteration, we solve the MDP to synthesize the optimal evacuation policy for robotic scouts, ensuring effective routes from impacted zones to safe locations. Our simulation results based on real-world data from previous disaster assessments and performance analysis demonstrate the effectiveness of our method and validate its potential to significantly improve disaster management and emergency response strategies.