Keywords: Physical and Cognitive Human-Robot Interaction

Keywords: Telerobotics, Legged Robots, Robotics in Hazardous Applications
Abstract: This paper presents our research aimed at deploying the Spot quadruped robot for inspection tasks in port environments to enhance worker safety. The study is organized into two phases. In the first phase, we conducted interviews with harbor personnel to gather their perspectives on safety-critical concerns and to assess the perceived suitability of using Spot in this setting. In the second phase, we assessed an immersive teleoperation framework designed for operating Spot, particularly within container yards where hazardous situations may arise. The immersive setup—based on a Head-Mounted Display (HMD) and 5G-enabled remote control—was evaluated in a user study and compared with a conventional interface consisting of a gamepad and a 2D monitor. The evaluation focused on operator workload, usability, simulator sickness, and participants’ performance in identifying simulated hazards. Overall, the findings highlight the practical relevance and technical viability of employing the Spot quadruped for port inspections to minimize workers’ exposure to dangers.

Keywords: Autonoums Vehicle Navigation, Intelligent Robotic Vehicles, Robotic Systems Architectures and Programming
Abstract: In heavy snowfall, LiDAR scans often contain substantial snow-induced noise, which can severely contaminate point clouds. When accumulating multiple scans to build a map, small registration errors can accumulate over time and lead to drift. Conventional VGICP summarizes points at the voxel level and uses these voxel summaries for registration; however, when snow points leak into the voxel summaries, registration becomes unstable and the accumulated error increases. We use a per-point snow probability to downweight likely snow points during alignment and voxel-map updates. Experiments on a snowy Boreas sequence show that the proposed method substantially reduces long-term drift, improving the scan-to-map translational ATE from 42.063 m to 20.234 m compared with conventional VGICP.

Keywords: Humanoids, Dynamics and Control
Abstract: This paper presents a reinforcement learning (RL)-based adaptive model predictive control (MPC) framework for robust bipedal locomotion. Due to the underactuated nature of bipedal robots and their sensitivity to external disturbances and contact variations, stability-aware real-time trajectory generation is essential. Conventional MPC-based approaches rely on manually tuned constant weights, limiting adaptability to varying environments, while end-to-end RL methods often lack interpretability and explicit stability guarantees. To overcome these limitations, this paper proposes an RL-based adaptive cost shaping strategy integrated into a model-based MPC framework. Rather than directly generating control actions, RL adjusts state-dependent cost weights according to physically meaningful stability metrics, such as ZMP margin and center-of-mass deviation. This enables real-time regulation of the trade-off between stability and agility while preserving the structure and interpretability of model-based control. In addition, an adaptive soft-constraint mechanism is introduced to improve optimization feasibility and computational efficiency. By incorporating slack variables into stability-related constraints and allowing RL to modulate their penalty terms, the proposed method enhances disturbance rejection and stability margins without compromising real-time performance.

Keywords: Simultaneous Localization and Mapping (SLAM), Motion Planning and Obstacle Avoidance
Abstract: This paper presents the design, implementation, and laboratory evaluation of a visual SLAM-based autonomous navigation system for an unmanned ground vehicle (UGV) intended for structured plantation environments. The proposed system integrates an RGB-D camera with RTAB-Map for localization and mapping, an A* algorithm for global path planning using occupancy grid maps, and a Pure Pursuit controller for path tracking. Experimental validation is conducted under controlled laboratory conditions using three path scenarios: straight-line motion, single-obstacle avoidance, and circular trajectory tracking. Performance is evaluated based on lateral tracking error and traversal time. Results demonstrate stable localization and successful path execution across all scenarios, while highlighting the influence of waypoint discretization and heading correction on tracking accuracy. The study provides practical insights into system integration, controller tuning, and implementation trade-offs, serving as a foundation for future field deployment in plantation environments.

Keywords: Intelligent Robotic Vehicles, Autonoums Vehicle Navigation, Wheeled Mobile Robots
Abstract: This paper presents an area-based coverage planning method for ultraviolet (UV) disinfection robots using a centroidal Voronoi tessellation (CVT)-based automatic waypoint generator. Unlike conventional cleaning robots that typically rely on reactive strategies such as zigzag or wall-following motions, UV disinfection robots must consider spatial irradiation characteristics to effectively disinfect an entire indoor environment. Since UV radiation propagates radially with a limited effective range, coverage planning must account for circular irradiation regions rather than simple surface traversal. To address this requirement, we propose a radius aware CVT-based waypoint generation framework for space disinfection. The number of waypoints is determined based on the effective UV radius and the free space area of an occupancy grid map. Lloyd’s algorithm is then applied to distribute the waypoints uniformly over the navigable space, enabling area based coverage rather than boundary focused traversal. A line-of-sight constraint is incorporated to reflect realistic UV propagation conditions in indoor environments. Experimental results on multiple occupancy grid maps demonstrate that the proposed method provides a novel automatic waypoint generation framework for UV disinfection robots. Given an input map, the algorithm computes area-based irradiation waypoints by considering the effective UV radius and spatial distribution of free space. The framework can be further extended by incorporating semantic map information as weighting factors, enabling adaptive and prioritized disinfection planning in future work.

Keywords: Micro/nanosystems, Physical and Cognitive Human-Robot Interaction, Deep Learning for Visual Percepton
Abstract: In recent years, the growing demand for intracytoplasmic sperm injection (ICSI) in assisted reproductive technology (ART) has led to a critical shortage of skilled embryologists. In this study, we propose an assistive system for the entire oocyte rotation procedure in ICSI. The system provides adaptive pipette operation guidance responsive to the operational scene based on polar body detection using U-Net. To address the inherent ambiguity in artificial intelligence (AI)-based polar body detection, we quantify the ambiguity in inference by the variance-based ambiguity score (VAS) using the deep ensembles method. The system enhances its reliability by requesting operator confirmation for operational scene transitions when the calculated VAS indicates high ambiguity. The subjective experiment using porcine oocytes indicated that the proposed system enhances the efficiency of oocyte rotation while achieving high operators’ trust in the system.

Keywords: Medical Robotics and Computer-Integrated Surgery, Foundations of Sensing and Estimation, Computer Vision and Visual Servoing
Abstract: Controlling the pan-tilt of the endoscope using angle knobs is difficult. The difficulty varies depending on the operator’s hand size and grip. In this study, we propose a system that automates the angle knob operation, which places a high physical burden on the operator. Motors are attached to the angle knobs to enable automatic pan-tilt control.The system uses depth anything to estimate the farthest point in endoscopic images in real time. This point is used as the insertion direction, and the endoscope tip is automatically controlled so that the point is kept at the center of the image.With this system, manual operation of the angle knobs can be eliminated, reducing the operator’s workload and making colonoscope insertion easier. The effectiveness of the proposed system was demonstrated through insertion experiments.

Keywords: Wheeled Mobile Robots, Autonoums Vehicle Navigation, Motion Planning and Obstacle Avoidance
Abstract: In this study, we propose a localization-aware path planning method for heterogeneous mobile robots that avoids low-reliability regions based on localization accuracy. A reliability map applicable to heterogeneous sensing systems is constructed and incorporated into the path cost function to enhance navigation stability. Simulation results demonstrate that the proposed method autonomously avoids low-reliability regions and can be applied to different motion mechanisms, including omnidirectional and differential-drive robots.

Keywords: Multi-Robot Systems, Intelligent Robotic Vehicles, Multisensor Data Fusion
Abstract: To address wildlife intrusion in farmland, we propose a ROS 2-based multi-UGV cooperative deterrence system. The proposed system utilizes mmWave radar for real-time animal trajectory detection and tracking, combined with GNSS/LiDAR hybrid positioning to ensure UGV reliability in complex farmland environments. This research incorporates generative models to dynamically generate unpredictable cooperative repelling paths based on grid-based map information (obstacles, animal locations, UGV distributions). Current work-in-progress (WIP) has achieved subsystem development and preliminary validation.

Keywords: Micro/nanosystems, Computer Vision and Visual Servoing, Haptics
Abstract: Micromanipulation tasks such as somatic cell nuclear transfer (SCNT) require frequent switching between a large field-of-view (FOV) image for exploration and a high-resolution image for precise positioning. This switching of the displayed FOV size disrupts workflow and demands visuomotor recalibration after each size change. Furthermore, maintaining stable and precise movements during high-resolution positioning under such frequent transitions is inherently difficult. This paper proposes and validates viscous force feedback based on FOV for a view-expansive microscope with a pen-type haptic device. The system scales pipette velocity to maintain consistent visual displacement regardless of the displayed FOV size. In addition, viscous damping is applied when the high-resolution image is displayed. We conducted a participant experiment simulating micromanipulation workflows with frequent switching of the displayed FOV size. The results show that viscous force feedback based on FOV significantly reduces path length and targeting time during high-resolution image display, suppressing overshooting and hand tremor.

Keywords: Micro/nanosystems, Learning from Humans, Computer Vision and Visual Servoing
Abstract: Intracytoplasmic sperm injection (ICSI) requires operators to adjust suction pressure to hold oocytes manually. This manual adjustment places a significant technical burden on operators. We propose a system that automates this adjustment using an artificial intelligence (AI) model. This model employs ensemble learning to predict the appropriate distance from the holding pipette tip to the oocyte cytoplasm by emulating expert manipulations. Experiments involving oocyte rotation tasks verified the system's effectiveness. The results demonstrated successful pressure adjustment and statistically significant improvements in both mental workload and system usability.

Keywords: Telerobotics, Contact: Modeling, Sensing and Control, Learning from Humans
Abstract: Collecting high-quality demonstrations for imitation learning commonly relies on leader--follower teleoperation. In contact-rich manipulation, especially tasks that require motion along object geometry such as crease formation in paper wrapping, successfully collecting high-quality demonstrations remains difficult. Existing force-feedback and compliance-control approaches typically represent interaction as point-wise or net force at the end-effector, which limits geometry-aware manipulation and can lead to object damage or conservative, slow operation. To address this problem, we propose tactile cooperative control for paper folding with edge tracing. Using spatially distributed contact information from a vision-based tactile sensor, the method allocates control degrees of freedom between the operator and the follower robot based on the contact state. The robot autonomously stabilizes contact-critical degrees of freedom (DoFs), while the operator controls the remaining DoFs to execute the folding motion. We compared the proposed method with a leader--follower baseline using admittance control in a user study on a paper folding task. The proposed method improved contact maintenance and increased the success rate without damaging the paper or the box. We also found that it enabled faster operation and reduced task completion time.

Keywords: Legged Robots, Wheeled Mobile Robots, Telerobotics
Abstract: Mobile robots with various locomotion modes are expected to adapt to diverse environments such as disaster sites. Hybrid mobile robots use multiple motors and perform various locomotion, such as wheeled locomotion and walking locomotion. ARMS, one of the Hybrid mobile robots, perform seven locomotion and operated by an operators. For operating ARMS, their twelve motors' angles requires a human's seven degree-of-freedom operation. However, these degrees of freedom for the operation are still large. This paper proposes a motor angle generation method for ARMS to reduce the degrees of freedom of the operation. The proposed method generates nine motor angles from an operator's two-dimensional command by using a decoder of a variational autoencoder. The variational autoencoder is trained to compress nine motor angles to two latent variables for the operation. We validated the generation of the nine motor angles for pushing forward locomotion of ARMS according to a trajectory of the latent variables given by an operator.

Keywords: Aerial and Flying Robots
Abstract: Reliable pose estimation relative to vertical structures is critical for UAVs operating near infrastructure such as walls and building facades. Accurate estimation of surface distance and yaw alignment is necessary to ensure stable motion and safe navigation in confined or cluttered environments. Conventional approaches rely primarily on 2D LiDAR or vision-based sensing, which may degrade in the presence of smoke, fog, dust, or transparent surfaces. Millimeter-wave (mmWave) radar offers improved robustness under such conditions but provides sparse measurements with limited angular resolution, making precise surface orientation estimation difficult when used alone.This paper presents a radar–LiDAR fusion framework for wall-relative pose estimation of a UAV. A lightweight 2D LiDAR is used to extract planar surface geometry and estimate yaw orientation, while mmWave radar measurements are incorporated for surface validation and robustness under sensing degradation. The proposed method estimates both relative distance and yaw alignment with respect to vertical structures. Experimental evaluation on a UAV platform demonstrates improved consistency and reliability compared to single-sensor approaches, particularly under challenging sensing conditions.

Keywords: Computer Vision and Visual Servoing, Cognitive Human Robot Interaction, Object Recognition
Abstract: 。インテリジェントスペース(iSpaces)は、 環境はセンシングによって自律的にサービスを提供します ユーザーやその周囲に関する情報。公の場で 環境、従来型の相互作用方法など、 個人のデバイスや大声の音声でサービスに合図を送る ユーザー 騒音やプライバシーによって制限される空間は、 干渉 他の人たちと共に。私たちはマルチモーダルの会話を提案します インターフェース これは(i)カメラによる手を上げたトリガーを統合しています 検出、 (ii) 視覚言語モデルに基づくサービス提案 (VLM) 運搬物体の推定、(iii) 局所音声 プレゼンテーション 指向性スピーカーを使用し、(iv)ジェスチャーベース 了解/ 拒絶の返答。キャリードオブジェクトモ

Keywords: Robotic Systems Architectures and Programming, Intelligent Robotic Vehicles
Abstract: RoboCup@Home imposes strict time limits, cas cading penalties, and a prohibition on human intervention— conditions under which raw capability matters less than reliability. This work-in-progress paper presents a strategic analysis of our championship-winning RoboCup@Home 2025 Open Platform League (OPL) entry, Anubis. We argue that the decisive factor in our victory was a deliberate variance minimizing deployment strategy: trading bonus opportunities for deterministic inputs, deploying hardware capabilities only when on-site validation permitted, and accepting failure modes outside our reliability envelope rather than attempting risky recovery. We support this claim with three competition case studies and final scoring data.

Keywords: Autonoums Vehicle Navigation, Motion Planning and Obstacle Avoidance, Contact: Modeling, Sensing and Control
Abstract: This paper presents a fundamental study on outdoor autonomous navigation of a mobile robot using a low-cost sensor configuration and a ROS-based navigation framework. The proposed system integrates GNSS, IMU, and wheel encoder measurements through an Extended Kalman Filter (EKF) to achieve reliable localization in outdoor environments subject to sensor noise and GNSS degradation. Waypoint-based global path planning and cost-map-based navigation are implemented using the Robot Operating System Navigation Stack. For real-time path tracking, a Pure Pursuit control algorithm is employed to generate stable motion commands based on a look-ahead point on the reference path. Outdoor experiments demonstrate successful point-to-point navigation and autonomous path following under varying GNSS fix conditions. Experimental results showed an average cross-track error of 0.38 m under RTK-FIX conditions, demonstrating stable outdoor path tracking performance. The results confirm that practical outdoor autonomous navigation can be achieved using low-cost sensors and standard navigation tools. Future work will focus on extending the system to unstructured outdoor environments and evaluating robustness under more challenging conditions.

Keywords: Evolutionary Robotics, Wheeled Mobile Robots
Abstract: This paper proposes an efficient navigation framework for a mobile robot operating in unknown environments. The framework integrates a global path planner based on the A* algorithm for Target Tracking (TT) and a local path planner for Obstacle Boundary Following (OBF). The TT behavior employs a dual-mode Fuzzy proportional-integral -derivative (PID) controller, meticulously designed for decoupled control of linear and angular velocities. This controller precisely controls the robot towards target points provided by the A* global planner. The OBF behavior employs a fuzzy controller with robot-obstacle distance information acquired from a laser rangefinder. This behavior enables the robot to bypass obstacles and avoid collisions. To enhance the navigation performance, an evolutionary computation algorithm is employed optimize the parameters of the controllers in each behavior. Simulations with comparative analyses show the effectiveness and efficiency of the proposed navigations scheme.

Keywords: Medical Robotics and Computer-Integrated Surgery, AI Reasoning Methods for Robotics, Learning from Humans
Abstract: Learning-based autonomous surgical assistance has shown promise in minimally invasive surgery. However, most existing systems are trained offline and deployed as fixed models, despite substantial distribution shifts in visual conditions, anatomy, and surgeon behavior. Under such shifts, models may produce confident but incorrect predictions, posing risks in safety-critical settings. To address this challenge, we propose an uncertainty-guided online continual learning framework that integrates selective supervision with online model updates during deployment. Predictive uncertainty from a probabilistic model is used to decide when to request supervision, promoting safety and supervision efficiency. For online learning in the stream, we adopt experience replay to incorporate new knowledge while mitigating forgetting. We evaluate the framework on a 2D target estimation task in collaborative tissue resection under a non-stationary stream, while fixing the online update rule to focus on uncertainty-guided querying. As a result, it reduces average anytime error while capturing a larger fraction of high-error samples compared to uniform querying, improving learning efficiency and risk-aware intervention. These results highlight the benefit of uncertainty-guided selective supervision for online continual learning under distribution shift.

Keywords: Computer Vision and Visual Servoing, Manipulation Planning and Control
Abstract: While most phenotyping platforms rely primarily on image-based measurements, advanced plant characterization requires the integration of active physiological sensing modalities such as chlorophyll fluorescence. We present an autonomous robotic platform designed to perform targeted fluorescence measurements on plant leaves. The system combines 3D plant reconstruction, geometric analysis, and motion planning to localize suitable measurement points and generate collision-free trajectories for a robotic manipulator.

A dense 3D model of the plant is reconstructed from multi-view data and used to extract candidate leaf surfaces based on orientation, accessibility, and sensing constraints. These targets are then integrated into a task-level planning framework that guides the end-effector to precise contact or near-contact configurations required for point-based fluorescence acquisition. The platform enables automated, repeatable, and spatially resolved physiological measurements that go beyond passive imaging.

By tightly coupling perception, geometric reasoning, and manipulation, the proposed system provides a robotics-driven approach to high-resolution plant phenotyping and opens new directions for autonomous agricultural inspection and plant-aware manipulation.

Keywords: Simultaneous Localization and Mapping (SLAM)
Abstract: Accelerated Coordinate Encoding (ACE) is a highly effective method for estimating camera poses from single RGB images. However, because it processes frames independently, it ignores the temporal continuity of video sequences. This independence often leads to unstable 3D coordinate predictions, especially in areas with repetitive textures or input noise, resulting in geometric outliers that confuse pose optimization solvers. In this paper, we introduce a simple, lightweight post-processing filter to stabilize the network's output without the need for expensive retraining. By mapping 2D features between consecutive frames using internal descriptors extracted directly from the ACE network encoder and applying a spatial distance threshold, we filter out erratic predictions before pose estimation. Early experiments on the Microsoft 7-Scenes dataset show that a 240cm threshold successfully targets geometric outliers, improving high-precision (2cm/2deg) localization accuracy by 0.9%.

Keywords: Robotics in Hazardous Applications, Mechanism and Design, Wheeled Mobile Robots
Abstract: This paper proposes, an improved version of the 3 in sewer pipe inspection robot equipped with an emergency evacuation mechanism. The low traction force and poor step-over capability, which were challenges of the first version, have been improved by simply connecting the propulsion units. The coupled propulsion units feature a differential mechanism capable of posture changes via a single wire, enabling adaptation to pipe diameter variations. To traverse obstacles like pipe joints, a control method was devised that detects obstacle contact through current load on the drive wheel motors and slackens the wire. This method was verified through simulated pipe experiments. Load comparisons were made using current waveforms applied to the drive wheels. Our proposed control method significantly improved the step-over capability of the new articulated robots.

Keywords: Force and Tactile Sensing, Modeling, Identification, Calibration, Contact: Modeling, Sensing and Control
Abstract: Multi-axis force/torque (F/T) sensors, which are indispensable components in robotic applications despite their high cost, have been widely integrated into various systems for human–robot interaction (HRI), robot joint stiffness control, force control, and related applications. However, these sensors are highly vulnerable to shocks and overloads. Once damaged, they are generally considered irreparable due to permanent deformation of the sensor’s elastic body, resulting in significant economic loss and environmental waste caused by the disposal of damaged force/torque sensors. In this paper, we propose a methodology to repurpose a damaged sensor into a partially functional force/torque sensor, even though it cannot be fully restored to its original operating condition. Specifically, we identify partially functioning strain gauges and recalibrate them into physical quantities such as Fx, Fy, and Tz using a normally functioning reference multi-axis force/torque sensor. Experimental results show that the root mean square errors (RMSEs) for Fx, Fy, and Tz are 0.34382 N, 0.3285 N, and 12.24216 N·mm, respectively, demonstrating the feasibility of the proposed sensor reuse approach.

Keywords: Learning from Humans, Manipulation Planning and Control, Telerobotics
Abstract: Imitation learning for tight-tolerance manipulation is often data-limited because task success can depend on brief alignment, release, and contact phases. We propose Button-Assisted Stage Reweighting (BASR), a low-overhead procedure for recording operator-perceived action importance during teleoperation using a clutch-style trigger. Stage 0 indicates a released clutch or no command, while stages 1–3 encode normal-to-critical motion during active operation. The recorded stages are used only during training, either by sampling important segments more often or by assigning larger imitation losses, while the policy receives no stage input at inference. On real-robot pick-and-place, the largest gain appears in the tighter 2mm tolerance setting, where the success rate increases from 40.0% for the baseline ACT policy to 75.0% with oversampling and 65.0% with loss weighting. We additionally report supplementary exploratory peg-in-hole and USB insertion results to motivate future evaluation on contact-rich precision tasks.

Keywords: Grasping, Deep Learning for Visual Percepton, Manipulation Planning and Control
Abstract: Teleoperation remains essential for executing complex maneuvers in hazardous or remote environments. However, relying solely on direct control in modern leader–follower setups imposes severe cognitive load and physical fatigue on the operator, especially during contact-rich tasks such as precise grasping. To bridge the gap between full autonomy and pure teleoperation, we propose a real-time, intent-driven shared autonomy framework. The system implicitly infers the operator’s objective through continuous visual action recognition and blends user inputs with an automatic grasp-approach trajectory. Using a distributed ROS 2 architecture, the vision pipeline localizes the target object and detects an “Approach the cup” primitive, which automatically triggers Contact-GraspNet to generate a target pose without explicit mode switching. A translation-focused State-Fusion Shared Control (SFSC) mode then governs the approach so that the operator can concentrate on rotational alignment and finger actuation of the robotic gripper. In a preliminary NASA-TLX evaluation on a cup pick-and-place task (5 participants, 10 trials per condition),the prototype reduced overall workload by 34.7% relative to direct teleoperation.

Keywords: Computer Vision and Visual Servoing, Cognitive Human Robot Interaction, Robotic Systems Architectures and Programming
Abstract: We present Team Tidyboy’s on-device, modular vision system for language-driven service robots in RoboCup@Home. The system is designed for robust operation under clutter, occlusion, and dynamic human interaction, and provides reusable perception outputs for manipulation and human–robot interaction. It combines YOLO and P-YOLO-3D for object perception and 3D localization, YOLO-Pose and rule-based cues for person tracking and gestures, DeepFace for identity verification, and an on-demand VLM layer for semantic attribute grounding, emphasizing task-level usage and stable interfaces to language and action planning.

Keywords: Modeling, Identification, Calibration, Humanoids, Social and Socially Assistive Robotics
Abstract: Recent advances in automatic speech recognition (ASR) have been driven by foundation models trained on massive datasets with large parameters. However, deploying these models on edge AI devices, especially robotic platforms, remains a significant challenge due to limited computational resources. Existing solutions often rely on cloud-based APIs or traditional DNN frameworks, which may raise privacy concerns or fall short of state-of-the-art performance. This paper presents a novel solution that enables ASR decoders, originally designed for GMM-HMM or DNN-HMM architectures, to support modern foundation models such as wav2vec2, HuBERT, and Whisper. Our approach facilitates the seamless integration of cutting-edge, highly accurate speech recognition capabilities into edge AI systems, including robotic platforms, without compromising user privacy or performance.

Keywords: Medical Robotics and Computer-Integrated Surgery, AI Reasoning Methods for Robotics, Cognitive Human Robot Interaction
Abstract: As collaborative robot-assisted surgery advances, effective natural language communication between surgeons and robotic assistants becomes increasingly relevant. A structural challenge in this communication is the pragmatic ambiguity of verbal instructions. Commands often omit explicit references to tools or targets, leading to uncertain execution parameters. This paper evaluates a multi-expert framework designed to interpret, ground, and disambiguate verbal surgical instructions using the visual context of the operative field. The system utilizes a two-stage affordance-based reasoning pipeline: first, a vision-language model (VLM) analyzes the endoscopic image to identify available tools; second, an affordance-augmented reasoning model cross-references the instruction with a knowledge base to determine the physical feasibility of the action. To measure reliability, we apply a dual-set conformal prediction method to quantify confidence and determine when to request human clarification. Evaluating a dataset of 180 instructions mapped to the CholecT50 dataset across six VLM configurations reveals that combining specialized models (Qwen3-VL-32B for visual grounding and Ministral-14B for reasoning) achieves a disambiguation rate of 73.3% while classifying 26.7% of instructions as requiring clarification. The results indicate that foundational VLMs with structured affordance logic and statistical confidence measures provide a viable, safe approach to handling ambiguous commands in surgical environments.

Keywords: Humanoids, Telerobotics, Learning from Humans
Abstract: Deformable object manipulation, such as package poly-bagging, remains a significant bottleneck in logistics automation due to high degrees of freedom and unpredictable material behavior. While semi-automated systems such as AutoBaggers exist, they still rely heavily on manual labor, exposing workers to repetitive strain and safety risks. This study presents a comprehensive learning-based framework to automate the AutoBagger task using a humanoid platform. We propose a hybrid teleoperation system that integrates robot-egocentric stereo vision, exoskeleton-based dual-arm control, and hand pose retargeting to collect high-quality demonstration data in confined industrial workspaces. To ensure operational stability, we employ Action Chunking with Transformers combined with a smoothing framework to mitigate discontinuous motions. Experimental results on the ROBROS IGRIS-C humanoid demonstrate successful execution of the full bagging pipeline, including grasping, insertion, and sealing. We further validate the feasibility of the proposed system through zero-shot deployment in an actual industrial environment.

Keywords: Computer Vision and Visual Servoing, Object Recognition, Simultaneous Localization and Mapping (SLAM)
Abstract: LiDAR SLAM technology is still costly due to high sensor price, computing load, and maintenance costs. As an alternative, we have developed Virtual Geometric Matching (VGM), a self-localization method that works by matching a pre-made 3D map and the AMR’s camera image. However, VGM still has a narrow use case as it has a high initial cost; the 3D map’s construction needs sizable time and workforce. This paper proposes an improved VGM with no initial cost by using 3D CAD models of existing equipment as the 3D map.

Keywords: Aerial and Flying Robots, Autonoums Vehicle Navigation, Motion Planning and Obstacle Avoidance
Abstract: UAV inspection of photovoltaic plants often relies on visual servoing along panel rows and GNSS-based transitions between rows. Since GNSS errors can be comparable to row spacing, they may cause failures during row-to-row navigation. This work proposes two complementary solutions: Post-TSP optimization to shorten critical inter-panel transitions, and a visual inter-panel navigation strategy to reduce GNSS reliance. Experiments on ten real PV plants in ROS 2 and CoppeliaSim, using GNSS traces recorded from a DJI Mavic 2 Enterprise Dual drone, show that the visual strategy achieves full completion in the valid analyzed runs across all tested GNSS error levels.

Keywords: AI Reasoning Methods for Robotics, Search and Rescue Robotics, Robotics in Hazardous Applications
Abstract: This work presents a Planning Domain Definition Language (PDDL)-based framework tailored to SAR scenarios, designed from requirements elicited through interviews with first responders (FRs). The domain models essential tasks for a quadruped robot, including environment assessment, victim detection, health evaluation, and safe path planning. The framework is evaluated through extensive simulations with varying environmental complexity. Results show that planning performance is mainly affected by the number of floors and rooms. These findings highlight scalability challenges and suggest the need for abstraction in large environments.

Keywords: Underwater Robotics, Autonoums Vehicle Navigation
Abstract: This late-breaking result reports a feature-based approach to underwater terrain-aided navigation (UTAN) using multibeam echosounders. The implementation combines Bag-of-Words (BoW)-based global localization with standard particle filtering to explore UTAN from an unknown initial state without prior pose information and recovery from the kidnapped robot problem (KRP) during operation. Prior to deployment, the reference map is partitioned into submaps, and handcrafted features are extracted and used to train a BoW, allowing each submap to be indexed for comparison. During the mission, once the vehicle has gathered enough bathymetric measurements, the newly constructed submap index is compared with the reference submap indexes to identify candidate matches. The vehicle pose is subsequently estimated through registration. The global localization results are used to initialize particles for continuous tracking. Experiments conducted on a sea trial dataset suggest that the combined approach can provide global navigation without prior pose information, while also maintaining robustness against KRP during the mission.

Keywords: Robotic Systems Architectures and Programming, Behavior-Based Systems, Legged Robots
Abstract: This paper presents an implementation-centered system integration result for mission-level autonomy of a quadruped robot in a realistic unstructured simulation environment. The proposed stack integrates a behavior-tree-based Integrated Hybrid-Task Planner, an operator mission interface, a Navigation2-based navigation layer, and an elevation/traversability-aware decision module into a single executable pipeline. The planner supports both operator-defined manual mission ordering and learning-enabled automatic mission sequencing through the same downstream execution logic. Validation is conducted in a patrol scenario on a site-like simulated field map, where mission poses are specified through the interface and executed by the BT-centered framework. The main contribution is a verified software architecture that unifies mission specification, sequencing, navigation, and terrain-aware decision support, providing a practical basis for mission-sequence planning, capability-aware task allocation, and heterogeneous multi-robot extension.

Keywords: Multi-Robot Systems, Wheeled Mobile Robots, Simultaneous Localization and Mapping (SLAM)
Abstract: In multi-robot navigation systems, semi-static obstacles that appear and disappear on planned global paths must be carefully handled. Constructing an accurate global obstacle map from distributed observations of multiple robots is challenging due to localization and perception errors. This paper proposes a multi-robot obstacle map management method that integrates distributed observations into an object-level global map. The proposed method performs three procedures, including data association, persistence-based probability update, and outlier removal with bounding box estimation. Real-world experiments with three mobile robots demonstrate that the proposed method reliably performs obstacle registration and removal while achieving centimeter-level positional accuracy.

Keywords: Autonoums Vehicle Navigation, Wheeled Mobile Robots, Simultaneous Localization and Mapping (SLAM)
Abstract: Reliable autonomous navigation of outdoor mobile robots requires robust localization and integrated system operation. This paper proposes an autonomous navigation system for outdoor mobile robots that leverages the 3D LiDAR-based localization method proposed in our prior work. The proposed system consists of an offline mapping phase and an online navigation pipeline that includes localization, global path planning, and local motion control. Through a 540 m autonomous navigation experiment on a university campus, the proposed system achieved centimeter-level accuracy in both localization and path tracking. The results validate that the localization method operates effectively within an integrated navigation system.

Keywords: Dynamics and Control, Modeling, Identification, Calibration
Abstract: We propose a hybrid dynamics modeling framework for high-speed autonomous surface vehicles (ASVs) that achieves both condition-invariant residual learning and fast online adaptation to changing operating regimes. The key idea is to separate invariant dynamics from condition-dependent effects, enabling robust modeling across vessel speeds and environmental disturbances. The model is built on a physics-informed nominal baseline that embeds the vessel’s three-degree-of-freedom rigid-body equations of motion. To compensate for planing-regime mismatch and unmodeled disturbances, we train a feature network with adversarial objectives that suppress speed- and domain-discriminative cues. Condition-dependent variations are then represented by a low-dimensional coefficient matrix updated online, allowing rapid adaptation without retraining the feature network. The resulting hybrid model preserves physical structure while improving generalization across operating regimes and environmental conditions.

Keywords: Motion Planning and Obstacle Avoidance, Wheeled Mobile Robots
Abstract: This paper introduces Spanning Forest Coverage (SFC), a grid-based complete coverage path planning framework capable of handling arbitrary start and goal endpoint constraints while maintaining the polynomial-time efficiency of traditional Spanning Tree Coverage (STC). Operating on a quotient graph abstraction, SFC utilizes a two-stage approach. First, dynamic programming computes an optimal, acyclic root coverage path along a serially connected sequence bridging the designated endpoints. Second, a Minimum Spanning Forest (MSF) is constructed to encompass the remaining workspace, seamlessly integrating local loops generated by each tree into the main trajectory. Simulations confirm that SFC minimizes redundant traversals and consistently outperforms existing baseline planners in constrained environments.

Keywords: Multi-Robot Systems, Wheeled Mobile Robots, Industrial Robots
Abstract: Autonomous Mobile Robots (AMRs) are increasingly used in logistics warehouses and manufacturing sites, but decentralized AMR systems may suffer from deadlocks in constrained spaces such as narrow corridors and intersections. This paper proposes a graph-based local coordination method that detects deadlock risk zones from overlapping robot paths and plans deadlock-free paths on a graph map generated from a 2D occupancy grid map. Conflict-Based Search is used for local path planning, and an Action Dependency Graph is adopted for robust execution. Simulation results in a warehouse environment show that, in the tested scenario, the proposed method improves the task schedule success rate from 80% to 100% and the average throughput from 1.5 to 1.97 tasks/min compared with the existing decentralized method.

Keywords: Underwater Robotics, Foundations of Sensing and Estimation, Autonoums Vehicle Navigation
Abstract: Underwater localization is challenging because GNSS signals are unavailable underwater and infrastructure- based acoustic navigation often requires costly deployment. Geomagnetic localization is a promising alternative because magnetic signals can be measured passively without exter- nal infrastructure. However, practical map-based geomagnetic localization is sensitive to temporal variation, which creates a mismatch between a prebuilt reference map and online magnetic measurements. This paper presents a geomagnetic field-based underwater localization method under temporal variation for underwater vehicles. The proposed method is built on a Rao-Blackwellized particle filter (RBPF), where planar position is represented by particles and the remaining states are estimated conditionally. To improve measurement discrimination, recent magnetic intensity measurements are compared in a queue form rather than at a single time step. To mitigate temporal bias, a particle-wise mean offset is estimated between the measured and predicted queues and compensated before likelihood evaluation. In 20 Monte Carlo runs, the proposed method reduced the mean XY localization error from 75.41±4.28 m to 42.35±1.78 m compared with the queue-based method without temporal compensation.

Keywords: Wheeled Mobile Robots, Autonoums Vehicle Navigation, Motion Planning and Obstacle Avoidance
Abstract: Reliable traversability estimation is essential for autonomous robots operating in outdoor environments. This paper reports recent field deployment of a self-supervised traversability learning framework on a wheeled mobile robot. The deployed pipeline produces per-cell traversability over an online elevation map and supports downstream path planning. We summarize qualitative system behavior and illustrate representative cases in a real-world outdoor scene. Finally, we share practical observations from operating the robot outside controlled conditions.

Keywords: Learning from Humans, Cognitive Human Robot Interaction, AI Reasoning Methods for Robotics
Abstract: In a ubiquitous manufacturing environment, humans remain at the core of performing complex, unstructured manual tasks, which requires systems to have the ability to seamlessly and objectively understand human operational skills. This study proposes a contactless skill assessment framework based on the topology of the Optimal Action Manifold (OAM). We constructed a multidimensional biomechanical phase space through monocular pose estimation and proposed a two-level analytical mechanism of 'macro-topology–micro-primitive.' At the macro level, Dynamic Time Warping (DTW) and Principal Component Analysis (PCA) map the expert’s high-dimensional degree-of-freedom bundling strategies onto a low-dimensional OAM with strict statistical boundaries, achieving time-warping-resistant quantitative scoring. At the micro level, the extraction of local absolute kinematic primitives reversely maps the abstract deviations to specific joints, enabling highly interpretable correction. Experiments conducted in two extreme manufacturing scenarios—machining (strong constraints, low dimensions) and sand casting (weak constraints, high dimensions)—validated the framework’s robust generalization capability in capturing dynamics divergence across Bernstein synergy spectra. This work transforms implicit human skills into computationally tractable topology, laying a foundational perceptual path for programming-free imitation learning in ubiquitous robotics.

Keywords: Modeling, Identification, Calibration, Dynamics and Control
Abstract: This paper introduces a data-driven method for modeling the dynamics of surface vehicles by integrating a deep learning framework with the Koopman Operator. Data-driven modeling has recently gained significant attention for its ability to represent complex dynamic behaviors that traditional models often fail to capture. In particular, ship dynamics involve strong nonlinearities and disturbances due to hydrodynamic forces, making conventional approaches less effective. The Koopman Operator offers a promising solution by providing globally linearized models that better capture the underlying dynamics of surface vehicles. Our proposed method accurately identifies an end-to-end model for surface vehicles without requiring prior knowledge of the vehicle. During validation, it demonstrates high prediction accuracy for surge, sway, and yaw velocities, highlighting the potential of data-driven methods to address challenges in dynamic system modeling.

Keywords: Robotic Systems Architectures and Programming, AI Reasoning Methods for Robotics, Performance Evaluation and Optimization
Abstract: As service robot software complexity grows, ISO 22166-202 defines a Software Information Model (SIM) for modular service robot software, but SIM-based test generation remains labor-intensive and error-prone. This paper proposes a SIM-grounded multi-LLM pipeline that uses an ISO 22166-202-conformant XML SIM instance, including retType error codes, to generate and refine executable pytest-based test code for ROS 2 modules. The pipeline assigns test design, test-code generation, and test-result analysis to three LLM agents connected through iterative feedback and specification-based grounding rules. Experimental results show that the number of passing tests increased from 1/17 to 17/17, with no interface mismatches or undefined API members observed in the final iteration.

Keywords: Modeling, Identification, Calibration, Dynamics and Control
Abstract: This paper presents a physics-informed neural modeling framework for the Cybership II surface vessel. A MATLAB/Simulink-based simulator was constructed to generate trajectory data under randomized thruster and rudder commands. Ten hydrodynamic damping coefficients were first identified using regression models derived from simplified three-degree-of-freedom surge, sway, and yaw dynamics. The identified coefficients were then embedded into a physics-informed loss for training a multilayer perceptron (MLP), which predicts next-step body-fixed velocities from the current velocities and control commands. The learned model was evaluated through closed-loop rollout and compared with trajectories from the original simulator.

Keywords: Social and Socially Assistive Robotics, Rehabilitation and Healthcare Robotics, Object Recognition
Abstract: This paper proposes an active wake-up assistance system based on Human-Robot Interaction (HRI) using a robot manipulator to enable independent awakening for individuals with hearing impairments. Conventional alarm systems for the hard-of-hearing primarily rely on visual signals or indirect vibratory stimuli through bedding; however, these methods often suffer from sensory adaptation and habituation, limiting their reliability. To overcome this limitation, we design an active system that delivers direct physical stimuli to the user's body to ensure reliable awakening. With a 7-DoF manipulator, the proposed system executes a four-stage sequential pipeline comprising YOLOv8-based arm detection, MediaPipe-based wrist localization, MoveIt-based motion planning and task execution, and repetitive physical stimulation. Experimental results demonstrate that the proposed pipeline was successfully executed at the user-specified wake-up time, suggesting that active physical interaction can enhance the reliability of wake-up assistance.

Keywords: Soft Robotics, Contact: Modeling, Sensing and Control
Abstract: We present a modular co-simulation platform for modeling and benchmarking deformable object manipulation. The system couples SOFA, as a high-fidelity finite-element reference, with rigid-body simulators (MuJoCo, NVIDIA Isaac Sim/Newton) for robot kinematics and scene dynamics. We validate the approach using a sphere-on-cube compression benchmark utilizing the Hertz contact model as reference. Results show that the framework reproduces contact behavior while highlighting the impact of solver coupling and synchronization on accuracy

Keywords: Legged Robots, Modeling, Identification, Calibration, Dynamics and Control
Abstract: We present the first closed-loop learning-based lo- comotion on MiNI-Q, a 270 g quadruped with eight continuous-rotation legs. A 26k-parameter PPO policy, trained in simulation with a system-identified actuator model, is deployed fully onboard: an ESP32-S3 runs the compiled policy at 50 Hz while a co-resident ESP32-C3 handles peripheral I/O over I2C, with no off-board compute and no wireless link in the control loop. The policy tracks linear and angular velocity commands on hardware after addressing a small set of observation-space fidelity issues characteristic of sub-kilogram legged platforms.

Keywords: Intelligent Robotic Vehicles, Motion Planning and Obstacle Avoidance, Wheeled Mobile Robots
Abstract: This paper presents a Late-Breaking Results report on an autonomous wheelchair platform designed for future indoor–outdoor mobility support. The platform integrates a 3D LiDAR, RGB camera, IMU, wheel odometry, UWB modules, GNSS, and ROS 2-based control modules on an embedded NVIDIA Jetson computer. At the current stage, preliminary experiments were conducted separately in indoor and outdoor environments rather than as a fully continuous indoor–outdoor navigation scenario. Indoor experiments focused on UWB-assisted localization in GNSS-denied spaces, where the final drift was reduced from 0.82 m to 0.31 m compared with odometry-only estimation. Stopping behavior was also evaluated in both indoor and outdoor environments under static-obstacle and pedestrian-crossing scenarios. Across 40 stopping trials, the platform achieved 35 successful stops, including 19/20 trials for static obstacles and 16/20 trials for pedestrian-crossing scenarios. These preliminary results demonstrate the feasibility of the proposed wheelchair platform and provide a basis for future continuous indoor–outdoor navigation and quantitative safety evaluation.

Keywords: Robot Surveillance and Security, Legged Robots, Multi-Robot Systems
Abstract: We present an integrated simulation testbed as an initial step toward learning-based mission planning for heterogeneous quadruped robots in orchard environments. The testbed combines a quadruped walking controller, an orchard world map, and task-level execution support in a unified simulation framework. A single quadruped robot has been successfully deployed to navigate toward designated mission points within the orchard scene, demonstrating end-to-end feasibility of the integrated system. This platform establishes the simulation foundation required for future research on scalable multi-robot mission planning under varying robot team sizes and task distributions.

Keywords: Physical and Cognitive Human-Robot Interaction

Keywords: Rehabilitation and Healthcare Robotics, Force and Tactile Sensing, Contact: Modeling, Sensing and Control
Abstract: Autonomous navigation toward target ultrasound views remains challenging in robotic ultrasound imaging due to the absence of global anatomical registration and explicit geometric models. This paper presents a similarity-based local iterative navigation method for coarse localization of target abdominal ultrasound views. The proposed approach is formulated as a local search problem within a bounded, image-defined reference space, where navigation decisions are derived from ultrasound image similarity without recovering absolute anatomical coordinates or relying on global path planning.

A robotic ultrasound system is developed to iteratively estimate the relative displacement between the current probe position and a user-specified target image. At each navigation step, relative position estimation is performed using ultrasound image similarity with respect to a preconstructed local reference image set, and the estimated displacement is used to guide discrete planar probe motion under closed-loop control. To reduce the dependence on single-point similarity estimation, a Top-K neighborhood aggregation strategy is adopted for position estimation, in which multiple high-similarity candidates are combined through similarity-weighted averaging.

Experiments conducted on an abdominal ultrasound phantom provide an initial proof-of-concept that the proposed method can guide the probe toward the target image from different initial positions within a local region. The results show that similarity-based relative displacement estimation provides directionally consistent guidance for coarse localization, while probe–surface contact conditions are regulated throughout navigation. These findings suggest the feasibility of similarity-driven local navigation for coarse ultrasound view localization within a constrained search space.

Keywords: Physical and Cognitive Human-Robot Interaction, Foundations of Sensing and Estimation, Rehabilitation and Healthcare Robotics
Abstract: Wearable electromyography (EMG) interfaces enable natural human-robot interaction, and low-density configurations are preferred for practical deployment due to their simplicity and cost-effectiveness. However, limited spatial information challenges reliable and timely pattern recognition. This study compares three segmentation strategies for three-channel EMG gesture recognition - fixed sliding window, event-gated adaptive thresholding, and Teager-Kaiser energy operator (TKEO) - evaluated on six gestures from eight subjects under leave-one-subject-out cross-validation with BiLSTM+Attention classification. Mean classification accuracies ranged from 59.17% (fixed window) to 64.39% (adaptive), with no statistically significant differences among methods (p > 0.18). Temporal evaluation showed mean onset-to-decision latencies of 113.9 ms (TKEO), 137.7 ms (fixed window), and 149.7 ms (adaptive) - all within the 300 ms real-time feasibility threshold. Cross-subject variability (26-88%) substantially exceeded inter-method differences, suggesting that subject-specific factors are a more critical performance determinant than segmentation strategy. These results are intended as a direction-confirming study for temporal processing design in low-channel wearable EMG systems; statistical generalization to larger populations is reserved for future work

Keywords: Rehabilitation and Healthcare Robotics, Deep Learning for Visual Percepton, Performance Evaluation and Optimization
Abstract: Robotic Patient Simulators (RPS) generate pain-related facial expressions in response to trainee actions, lending caregiving training the realism of interacting with an actual patient. However, artifacts such as unnatural expression transitions or eye–mouth asynchrony can cause trainees to misjudge pain intensity and timing, undermining training fidelity. PainDiffusion, a diffusion-based model for naturalistic pain expression synthesis, is governed by several inference-time hyperparameters—text guidance strength, sampling temperature, action-unit (AU) scaling, and denoising steps—whose effects on output quality remain poorly understood. We propose a lightweight, reproducible protocol that systematically explores these hyperparameters on a fixed PainDiffusion baseline through one-factor-at-a-time sweeps. Twenty human raters evaluated the outputs on a 5-point naturalness scale. Results indicate that moderate text guidance and sampling temperature yield the highest naturalness ratings, whereas increasing AU intensity consistently degrades perceived quality. Free-text comments further identify eye–mouth asynchrony and localized temporal jitter as key factors contributing to perceptual quality degradation. The proposed workflow provides a practical engineering baseline for selecting stable inference configurations and supports future extension to larger-scale evaluation studies.

Keywords: Rehabilitation and Healthcare Robotics, Neurorobotics, Human-Robot Augmentation
Abstract: Surface electromyography-based hand gesture recognition presents considerable potential for wearable human computer interaction. However, its streaming performance typically degrades due to inter-subject variability and variations in the electrode placement and wearing conditions. In unlabeled test streams, online normalization can become unreliable when its running baseline statistics are contaminated by active segments or low-quality signals. In this study, we propose a causal preprocessing method called streaming statistics-based input normalization, which mitigates the feature distribution shifts without using target labels, future samples, or test-time model updates. The proposed method applies normalization at each hop while updating the baseline statistics only from the windows detected as rest-like and passing a signal-quality criterion, thereby reducing the update contamination. It maintains feature statistics at the short and long time scales and adaptively mixes them based on their discrepancy to handle both short-term fluctuations and long-term drift. We evaluated the proposed method on a public benchmark dataset with eight channel forearm surface electromyography, time-domain features, and a ResNet-18 classifier under a 40-subject leave-one subject-out streaming protocol (8-class subset). It achieved an accuracy of 74.0 ± 8.9% and macro-F1 of 39.8 ± 9.1%, thereby improving over no normalization with an increase of +8.4%p in the accuracy and +14.4%p in the macro-F1. We additionally reported deployment-oriented observables such as baseline ready rate and time-to-ready to quantify the feasibility of the proposed streaming normalization. These results support streaming statistics-based input normalization as a promising strategy for long-term online electromyography streaming recognition.

Keywords: Medical Robotics and Computer-Integrated Surgery, Deep Learning for Visual Percepton, Computer Vision and Visual Servoing
Abstract: Monocular depth estimation in endoscopic environments is constrained by the scarcity of diverse and high-quality paired RGB-depth data. Although synthetic data can provide geometric ground truth, the sim-to-real appearance gap degrades transfer performance. Conversely, unconstrained neural rendering techniques like 3D Gaussian Splatting (3DGS) can synthesize photorealistic views, but they routinely struggle to reconstruct accurate underlying geometry in textureless, non-Lambertian biological tissues. This paper evaluates whether interpolated synthetic RGB-depth data generated by a geometrically anchored, prior-guided 3DGS model improves monocular depth estimation. By initializing 3DGS from real endoscope frames via an external geometric prior, the scene optimization is restricted to prioritize appearance over geometry mutation. Novel continuous viewpoints are then generated via Spherical Linear Interpolation (SLERP) and rendered into paired RGB-depth sets. We evaluate a Depth-Anything-V2 (ViT-Base) setup comparing real-only supervision versus real+GS supervision with matched optimization bounds and held-out real evaluation. In a test set of 77 endoscope images, the augmentation strategy reduces the absolute relative error (AbsRel) from 0.0198 to 0.0168 and the RMSE from 1.933 to 1.718 mm. The analysis indicates that calibrated prior-guided synthetic data mitigate target-domain overfitting and improve robust geometric generalization for low-data surgical tasks.

Keywords: Robotic Hands, Mechanism and Design, Grasping
Abstract: Wire-driven robotic hands are widely used for achieving compact, lightweight, and anthropomorphic designs. However, conventional architectures suffer from path dependency, in which configuration-dependent wire routing causes unintended coupling between joints. This phenomenon degrades motion accuracy, repeatability, and controllability, and makes independent joint actuation within a finger difficult, particularly in multi-joint and bidirectional motion scenarios. This paper presents a wire-driven robotic hand with reduced path dependency based on a joint-centered wire routing architecture. In the proposed design, all actuation wires are routed through the geometric centers of the joints so that joint rotation does not induce changes in effective wire length or transmission characteristics. As a result, mechanical interference between joints is eliminated, enabling independent actuation of multiple finger joints with predictable transmission behavior. A five-fingered robotic hand with 25 degrees of freedom was designed and fabricated to validate the proposed concept. Kinematic modeling and manipulability analysis were conducted to examine the motion characteristics enabled by the proposed structure. Experimental evaluations focusing on single-joint actuation confirmed that joint motions do not affect the wire routing or actuation behavior of other joints. The results demonstrate accurate and repeatable joint motions in both forward and reverse directions, verifying the effectiveness of the proposed joint-centered wire routing approach for reducing path dependency in wire-driven robotic hands.

Keywords: Robotic Hands, Cognitive Human Robot Interaction, Mechanism and Design
Abstract: Scalable learning of dexterous manipulation remains bottlenecked by the difficulty of collecting natural, high-fidelity human demonstrations of multi-finger hands due to occlusion, complex hand kinematics, and contact-rich interactions. We present WHED, a wearable hand-exoskeleton system designed for in-the-wild demonstration capture, guided by two principles: wearability-first operation for extended use and a pose-tolerant, free-to-move thumb coupling that preserves natural thumb behaviors while maintaining a consistent mapping to the target robot thumb degrees of freedom. WHED integrates a linkage-driven finger interface with passive fit accommodation, a modified passive hand with robust proprioceptive sensing, and an onboard sensing/power module. We also provide an end-to-end data pipeline that synchronizes joint encoders, AR-based end-effector pose, and wrist-mounted visual observations, and supports post-processing for time alignment and replay. We demonstrate feasibility on representative grasping and manipulation sequences spanning precision pinch and full-hand enclosure grasps, and show qualitative consistency between collected demonstrations and replayed executions.

Keywords: Robotic Hands, Force and Tactile Sensing, Mechanism and Design
Abstract: This paper presents a robotic gripper designed to enable the simultaneous acquisition of visual and visuotactile information within a shared field of view using a single camera. The proposed gripper employs a passive linkage mechanism that changes the mounting angle of the fingertip camera in coordination with the opening and closing motions of the gripper. Consequently, global visual information is obtained during the approach phase, whereas local visual and visuotactile information can be captured after grasping without additional actuators or sensors. Based on the geometric relationship between the passive linkage structure and camera field of view, design constraints were formulated, and the design parameters were determined numerically. A prototype gripper was fabricated, and experiments were conducted to verify the view transition and the simultaneous observation of an object-side marker and tactile-surface markers within a shared field of view. The experimental results confirmed that the proposed mechanism achieved the intended view transition and that, under a simple vertical contact condition, marker displacements could be measured repeatably.

Keywords: Robotic Hands, Soft Robotics, Grasping
Abstract: Soft grippers enable stable grasping through compliant structures and multipoint contact; however, manipulating the pose of a grasped object while maintaining contact remains challenging, particularly when large pose changes are required. Conventional approaches based on locally estimated drive matrices are limited to small motions owing to the actuator stroke and approximation validity. This study proposes a selective actuation strategy for redundant parallel-driven soft grippers to generate large object pose changes through the iterative application of small motions. By exploiting actuation redundancy, predefined actuator groups that satisfy the force-closure conditions are alternately activated. The partial drive matrices corresponding to each actuator group are extracted from a single, globally estimated drive matrix, enabling the discrete switching of control inputs without re-estimation. After reaching the vicinity of the target pose, the control scheme is switched to a full-actuator configuration for high-precision position and orientation regulation. Experiments using a fabricated redundant soft gripper demonstrated that large pose changes, which are difficult to achieve with a single control action, can be realized step-by-step while maintaining stable grasping.

Keywords: Soft Robotics, Mechanism and Design, Biomimetic and Bioinspired Robots
Abstract: Conventional robotic grippers such as suction cups and parallel-jaw claws often struggle with the diverse shapes, surface textures, and fragility of real-world objects. Inspired by the elephant’s trunk, we present the TRUNK-gripper, a soft multifunctional system that integrates gripping, twisting, wrapping and absorption to enhance grasp stability and surface adaptability. Its soft-skirt structure enables vacuum sealing on irregular surfaces, twisting motions to extract embedded items, and compliant contact for gently enclosing fragile objects. We introduce a 100-object benchmark comprising 49 adversarial shapes, 20 fragile items, and 31 household goods, and use it to evaluate the TRUNK-gripper under real-world, multi-challenge grasping scenarios. Achieving a 93.4% success rate (467/500), the TRUNK-gripper significantly outperformed a parallel-jaw gripper (80.6%, 403/500) and a standard suction gripper (1.2%, 6/500). These results underscore the potential of biologically inspired, contact-based multifunctional grasping for robust performance across a wide range of manipulation tasks.

Keywords: Intelligent Robotic Vehicles, Autonoums Vehicle Navigation, Learning from Humans
Abstract: Autonomous robots operating in off-road environments must reason about environmental properties that are only partially observable and change over time. Frame-wise perception alone is insufficient to capture such properties reliably, especially in cooperative leader-follower settings with heterogeneous sensing capabilities. This work introduces a memory-aware environmental knowledge pipeline that transforms transient semantic observations into persistent, decision-relevant knowledge. Local perception is abstracted into a compact class representation and maintained in short term and long term memory using explicit confidence modeling. Abstract knowledge is selectively shared between a leader and a perception-degraded follower without exchanging raw sensor data or dense maps. The system is evaluated using dataset playback, focusing on anticipation and knowledge quality. Results show that memory-aware cooperation enables earlier reliable awareness and more stable internal environmental representations under partial observability.

Keywords: Autonoums Vehicle Navigation, Robotic Systems Architectures and Programming, Intelligent Robotic Vehicles
Abstract: Autonomous mobile robots typically require geometrically consistent maps for navigation, which must be created and maintained using onboard sensors. This requirement limits practical deployment because map construction and maintenance impose significant operational costs. In this paper, we propose a navigation framework that uses architectural CAD drawings as a "weak spatial reference" instead of a precise map. This approach separates global referencing and local motion generation by utilizing LiDAR-based localization in the CAD coordinate frame for global guidance, while generating local motion through a vision-based deep reinforcement learning policy.

Experimental results demonstrate that the proposed framework maintains stable navigation even in situations where conventional map-based methods fail due to inconsistencies between the CAD drawings and the real environment. In particular, long-range navigation was achieved as the vision-based policy adaptively determined traversable areas based on perception, even when localization was unstable. These results present the potential for a practical autonomous navigation system that directly utilizes human-oriented drawings without the need for maps for robots.

Keywords: Autonoums Vehicle Navigation, Computer Vision and Visual Servoing, Wheeled Mobile Robots
Abstract: This paper proposes a novel navigation method for the autonomous weeding robot R4 (Ritsumeikan Road-weed Removal Robot), designed to reduce labor costs in roadside weeding operations. In visual navigation systems, misrecognition of road regions often generates paths that deviate from the actual road geometry. To address this, the proposed method employs road-region detection and vanishing-point detection to identify road geometry and generates trajectories using linear or curved mathematical models depending on the road conditions. This study focuses on the preliminary verification of the proposed system to explore its feasibility for roadside environments. The fundamental characteristics and potential of the navigation method are evaluated through preliminary driving experiments, providing insights for future robust autonomous navigation.

Keywords: Autonoums Vehicle Navigation, Motion Planning and Obstacle Avoidance
Abstract: Autonomous loose fruit collection in oil palm plantations requires reliable tree trunk localization and stable navigation in GPS-denied outdoor environments. This paper presents a low-cost RGB-D vision-based framework for tree trunk detection and circular navigation of an agricultural mobile robot. The proposed method combines depth-based geometric edge detection with HSV colour-space validation to improve robustness under varying illumination. A vision-driven circular waypoint generation strategy enables 360° navigation around individual trees without prior mapping or external localization. To enhance radial tracking accuracy, a depth based PID controller regulates linear velocity to compensate for odometry drift and perception noise during circular motion. The system is implemented on a TurtleBot3 Waffle Pi equipped with an Intel RealSense D435i camera and evaluated under varying distances and lighting conditions. Experimental results demonstrate reliable trunk detection within 1.15–2.15 m and significant reduction in radial deviation under closed-loop control compared to open-loop tracking. The proposed approach provides a practical and computationally efficient solution for GPS-free agricultural robot navigation in plantation environments.

Keywords: Autonoums Vehicle Navigation, Intelligent Robotic Vehicles, Deep Learning for Visual Percepton
Abstract: Drivable area detection in heavy snowfall remains challenging due to occluded lanes and ambiguous boundaries. Furthermore, conventional supervised learning requires costly manual annotation. To address this, we propose a robust weakly supervised framework using GPS-based ego-trajectories. By projecting historical GPS data onto images, we automatically generate labels, allowing the model to learn physical drivability rather than visual boundaries. Additionally, we employ a coarse-to-fine strategy and Test-Time Augmentation (TTA) to improve the reliability of predictions performance. Experiments demonstrate that our method achieves highly reliable detection in snowy environments without any manual labeling.

Keywords: Wheeled Mobile Robots, Force and Tactile Sensing, Multi-Robot Systems
Abstract: Traditional indoor robot localization commonly relies on onboard sensors such as cameras and LiDAR, which are susceptible to occlusion, illumination changes, and limited sensing range. Pressure-based localization offers an infrastructure-level alternative by measuring ground reaction forces through embedded floor sensors, enabling global position estimation within the instrumented area. However, multi-robot scenarios introduce significant challenges. When multiple robots move simultaneously on a deformable surface, their deformation fields overlap, making it difficult to distinguish individual robot positions. While finite element modeling provides physically accurate predictions, it is computationally expensive, and purely data-driven Gaussian Process (GP) models often lack physical consistency. This paper proposes a physics-informed Gaussian Process framework for pressure-based multi-robot localization. The proposed method incorporates plate mechanics into a GP surrogate model and models multi-robot pressure responses as the sum of individual robot-induced deformation fields. Dual-robot experiments demonstrate accurate and robust global localization, effectively separating multiple robots from combined pressure observations.

Keywords: Multi-Robot Systems, Dynamics and Control, Intelligent Robotic Vehicles
Abstract: Heterogeneous dual-robot collaborative systems offer significant operational flexibility but present coordination challenges due to kinematic mismatches. While Integral Sliding Mode Control (ISMC) effectively handles nonlinear dynamics, traditional fixed-gain ISMC suffers from a fundamental trade-off between robustness and chattering suppression. To address this, this paper proposes a TD3-based Adaptive Acceleration-Level ISMC strategy. Unlike conventional velocity-level approaches, the proposed method operates at the acceleration level to generate smoother commands and employs a Twin Delayed Deep Deterministic Policy Gradient (TD3) agent as a dynamic gain scheduler. By integrating a lookahead mechanism, the agent anticipates curvature changes and optimizes sliding mode parameters—switching gains (𝑲𝟏,𝑲𝟐), and integral gain (𝑲𝟑)—in real-time. This allows for aggressive disturbance suppression during maneuvers while minimizing control effort on straight paths. Simulation results in a high-fidelity ROS 2/Gazebo environment demonstrate that the proposed approach significantly reduces chattering and achieves superior tracking precision (lower RMSE and maximum error) compared to fixed-gain baselines.

Keywords: Multi-Robot Systems, Performance Evaluation and Optimization
Abstract: The complete collection of sparse resources in large, unknown environments remains a challenging problem for autonomous robot swarms. Previous studies have shown that a substantial portion of total mission time is consumed during the final stage of collection, where only a small fraction of randomly scattered resources remain. Consequently, many existing swarm foraging algorithms (search and collection) focus on collecting most resources within a limited time window, rather than improving end-stage efficiency for collecting all resources. We propose a grid-based stochastic foraging strategy that explicitly reduces redundant visits and accelerates late-stage collection. The unknown search area is partitioned into a grid map, which is maintained by a lightweight central server. To maintain scalability, both robots and the server operate within limited memory and computational constraints. Robots execute decentralized search behaviors while exploring assigned regions and communicate summarized trajectory information to the server upon returning to the center only. The server updates the grid-level visitation counts based on robot-reported locations, producing a global estimate of the exploration density. For each new foraging trip, a robot selects its next search area from a local 3times 3 neighborhood of grids probabilistically with the lowest visitation count, thus biasing exploration toward under-visited regions while maintaining stochasticity. Extensive simulation experiments demonstrate that the proposed strategy consistently outperforms the canonical centrally placed baseline foraging algorithm (CPFA). Compared to CPFA, the proposed method reduces the total collection time by up to 33% and improves collection efficiency by more than 48% during the final stage of the mission. These results indicate that the proposed strategy is robust, flexible, and scalable for near-complete and complete resource collection in robot swarms and can serve as a general enhancement for stochastic swarm foraging methods under limited onboard resources.

Keywords: Multi-Robot Systems, Performance Evaluation and Optimization
Abstract: Multi-Robot Task Allocation (MRTA) is a central challenge in decentralized multi-agent systems, where teams of robots must cooperatively assign and execute tasks under limited communication while optimizing global performance objectives. Auction-consensus algorithms, such as the Consensus-Based Bundle Algorithm (CBBA), provide scalable decentralized coordination with provable convergence, but rely on hand-crafted greedy scoring functions that often lead to suboptimal task allocations. This paper proposes a learning-enhanced auction-consensus framework in which CBBA’s deterministic bidding mechanism is replaced by a neural bidding policy trained using reinforcement learning. Under a centralized training and decentralized execution paradigm, agents learn to compute task bids from partial local observations while retaining the standard auction and consensus phases for decentralized coordination. The learned bidding policy is trained using Proximal Policy Optimization with rewards shaped by proximity to globally optimal solutions obtained via mixed-integer linear programming. Multiple neural architectures are evaluated, including a Neural Additive Model, the Long Short-Term Memory (LSTM) model, and the Set Transformer Model. Experimental results across varying swarm sizes demonstrate that learned bidding policies can improve solution quality over classical CBBA while preserving decentralized execution. The proposed approach highlights the effectiveness of integrating reinforcement learning with classical distributed coordination algorithms, offering a scalable pathway toward higher-quality decentralized multi-robot task allocation.

Keywords: Multi-Robot Systems, Telerobotics, World Modelling
Abstract: Mixed Reality (MR) interfaces have been extensively explored for controlling mobile robots, but there is limited research on their application to managing teams of robots. This paper presents HORUS: Holistic Operational Reality for Unified Systems, a Mixed Reality interface offering a comprehensive set of tools for managing multiple mobile robots simultaneously. HORUS enables operators to monitor individual robot statuses, visualize sensor data projected in real time, and assign tasks to single robots, subsets of the team, or the entire group, all from a Mini-Map (Ground Station). The interface also provides different teleoperation modes: a mini-map mode that allows teleoperation while observing the robot model and its transform on the mini-map, and a semi-immersive mode that offers a flat, screen-like view in either single or stereo view (3D). We conducted a user study in which participants used HORUS to manage a team of mobile robots tasked with finding clues in an environment, simulating search and rescue tasks. The study compares an integrated MR supervisory interface (shared map + goal-based navigation + task tools + mini-map teleoperation) against an individual teleoperation-only baseline. The results suggest HORUS can improve mission efficiency and perceived usability for multi-robot operation, and demonstrates its potential to advance human-robot collaboration in dynamic, team based environments

Keywords: AI Reasoning Methods for Robotics, Performance Evaluation and Optimization
Abstract: As Vision-Language-Action (VLA) models continue to scale in the number of parameters, the computational cost and resource requirements for domain-specific fine-tuning have become significant barriers to practical robotic deployment. While Parameter-Efficient Fine-Tuning (PEFT) techniques like LoRA (Low-Rank Adaptation) offer a potential solution, they often fail to match the task success rates of their fully fine-tuned counterparts. In this paper, we propose a novel hierarchical VLA architecture that achieves state-of-the-art performance while maintaining high parameter efficiency. Our model decomposes control into a high-level System 2 for instruction-conditioned semantic context encoding—a frozen PaliGemma-3B backbone with 0.12B trainable LoRA parameters and a reactive System 1 for multimodal fusion and action generation. To optimize training efficiency, System 2 takes only a single egocentric image, while System 1 recovers missing context by integrating wrist-view images via ResNet-34 and proprioceptive state history encoded with a single linear projection layer. This information is fused through a Transformer Encoder, and final action trajectories are refined via a Transformer-parameterized conditional flow-matching decoder. To improve task performance, we generate diverse candidates by sampling from (N) independently initialized Gaussian noise vectors and using different numbers of denoising steps (K) per sample, and then select the executed action using a Cal-QL (Calibrated Q-Learning)-based critic. Evaluated on the standardized LIBERO benchmark, our proposed model achieved a 98.1% average success rate, outperforming contemporary fully trained models across all task suites. These results demonstrate that strategic architectural design can enable parameter-efficient models to exceed the performance of full-scale fine-tuning, offering a viable path for high-performance robotics under constrained computational resources.

Keywords: Motion Planning and Obstacle Avoidance, AI Reasoning Methods for Robotics, World Modelling
Abstract: Large Language Models (LLMs) have demonstrated strong generalization and reasoning capabilities across a wide range of domains, including embodied decision making and robotics. Despite this progress, existing reinforcement-based embodied agents often struggle with novel tasks requiring complex spatial reasoning and multi-object manipulation when relying solely on the robot’s egocentric view. In this paper, we propose SPaRL (Spatially-aware Reinforcement Learning with Language), a novel framework that explicitly integrates a structured 3D scene graph into an LLM-based reinforcement learning policy. By representing objects and their spatial relationships in a compact, semantically meaningful graph and conditioning it on task instructions, SPaRL provides the policy with explicit relational context beyond raw visual inputs. The scene graph is dynamically pruned to retain instruction-relevant objects and relations, serialized into natural language, and jointly processed with visual observations and task descriptions by a frozen LLM backbone. We evaluate SPaRL on language-conditioned rearrangement tasks in Habitat. Results show that incorporating an instruction-conditioned scene graph consistently improves performance over a vision-only LLM policy. Across curriculum training, SPaRL achieved improved performance on the Language Rearrangement benchmark. These results suggest that explicit spatial representations provide useful inductive bias for object reasoning, particularly as task difficulty increases.

Keywords: Learning from Humans, AI Reasoning Methods for Robotics, Physical and Cognitive Human-Robot Interaction
Abstract: In this paper, we study sample-efficient preference learning for human-robot interaction under forced-choice feedback constraints. In many embodied settings, users select a preferred option from executed robot behaviors, whereas existing preference learning and bandit formulations often assume the availability of a no-choice option. We formulate preference learning as a contextual preference bandit under forced-choice winner feedback and propose a confidence-guided query selection strategy that jointly considers predicted preference and model uncertainty. We provide a regret analysis establishing sublinear cumulative regret and demonstrate through simulation and real-world experiments that the proposed approach requires fewer user interactions to identify user-aligned behaviors than existing baselines. A video illustrating this work can be found on our project page https://chanwookim971024.github.io/SaFe/.

Keywords: Computer Vision and Visual Servoing, Manipulation Planning and Control, AI Reasoning Methods for Robotics
Abstract: Vision-Language-Action(VLA) models have demonstrated strong performance in robot manipulation by leveraging pre-trained vision-language models to map observations directly to actions. However, existing approaches reason primarily at the visual or semantic level, lacking explicit understanding of the physical interactions that fundamentally govern manipulation tasks. In this paper, we propose Physics Reasoning VLA, a method that enables VLA models to explicitly reason about physical interactions prior to acting, grounded in two fundamental quantities: contact points, which specify where the target object interacts with the robot or surrounding environment, and contact forces, which describe the magnitude and direction of force applied at those locations. Rather than directly mapping observations to actions, our model first predicts contact points and forces at the pixel level via learnable physics queries, then incorporates the resulting physics-aware features alongside visual and language inputs to generate actions. To prevent physics reasoning from disrupting pre-trained visual and linguistic representations, we further introduce a hybrid attention mechanism that applies full attention over image, language, and proprioceptive tokens, while applying causal attention over phyics query and action tokens. We evaluate our method on the RoboCasa simulation benchmark, emonstrating that physics reasoning consistently improves performance over the vanilla π0 baseline, with an average success rate improvement from 36.2% to 43.6%.

Keywords: Evolutionary Robotics, Legged Robots, Dynamics and Control
Abstract: Recent advances in reinforcement learning have demonstrated multitask capabilities for individual robots, while separate lines of research have explored morphology-invariant locomotion across heterogeneous platforms. However, combining multitask learning with morphological diversity remains an open challenge. In this study, we extend a common-locomotion policy framework to the multitask setting by conditioning a single shared policy on target velocity commands. This design allows the policy to solve multiple velocity-tracking tasks across robots with different morphologies within a unified architecture. Our long-term objective is modular robot control, where adaptability and transferability across heterogeneous platforms are fundamental requirements. Furthermore, we evaluate robustness to morphology variations at test time and demonstrate that policies trained to capture morphology-invariant locomotion principles can adapt effectively to structural changes without task-specific retraining.

Keywords: Social and Socially Assistive Robotics, Manipulation Planning and Control
Abstract: Assistive mobile manipulators must operate reliably in cluttered home environments while remaining accessible to users with limited mobility or visual impairments. Large language model (LLM)–based planners enable flexible interpretation of natural language instructions; however, purely sequential action generation often results in ambiguous grounding, omission of critical intermediate steps, and unsafe execution under uncertainty.

We present an enhanced Mobile Assistive Robotic System (MARS) that integrates voice-centric remote interaction with structured task management and fault-tolerant execution. The system comprises: (i) a web-based, voice-first user interface employing a WebRTC bridge for cross-device access; (ii) an LLM-driven Task Manager that converts user commands into structured program representations and employs clarification and confirmation loops for underspecified or safety-critical actions; and (iii) a hierarchical Skill Controller with explicit precondition checks and automated retry mechanisms.

Empirical evaluation on a physical mobile manipulator in a mock apartment testbed (200 trials per configuration) shows that the enhanced MARS achieves a 92% task success rate, outperforming a baseline system (78%) and substantially reducing failures due to ambiguous intent and transient perception or manipulation errors. A user study with 50 participants with visual impairments further indicates that voice-centric interaction with confirmation prompts improves system accessibility and perceived transparency.

Keywords: Dynamics and Control, Physical and Cognitive Human-Robot Interaction, Modeling, Identification, Calibration
Abstract: The pointing task is an important part of human-robot interaction. To bridge the gap between human intuition and the underactuated dynamics of a blimp, this paper proposes an integrated interaction framework. The core of the framework is a reformulated 2D VITE model acting as an intent observer, which infers the human’s target destination from early-stage wand gestures. By incorporating a coupling compensation control law, the framework provides a predictive trajectory reference that is inherently smooth and respects blimp constraints. Trajectory reconstruction experiments demonstrate the effectiveness of this architecture, showing that the model can accurately capture and predict complex 2D human pointing intent, providing a foundation for preemptive robot responses.

Keywords: Physical and Cognitive Human-Robot Interaction, Cognitive Human Robot Interaction, Legged Robots
Abstract: This paper proposes a method for conveying an impression of age through robot motion. Expressing the aging process enhances the robot's realism as a living creature, thereby forming a stronger emotional bond between the robot and its users. To achieve this, we designed the gait of a physical quadruped robot to reflect the age-related characteristics of real dogs by selectively applying specific movement variables. Based on the behavior of elderly dogs, we defined three core motion features, namely slower speed, longitudinal sway, and disoriented pauses. We generated composite motions by systematically combining the presence or absence of these features and subsequently evaluated the resulting impressions. Experimental results demonstrate that robot motion can effectively evoke different age impressions, and that slower speed and disoriented pauses have a strong impact on the perceived age.

Keywords: Cognitive Human Robot Interaction, AI Reasoning Methods for Robotics
Abstract: In robotics, task planning using Vision-Language Models (VLMs) has attracted significant attention. The contextual understanding capabilities of VLMs enable adaptation to dynamic environmental changes and autonomous generation of high-level actions based on semantic region specifications.

However, relying solely on standard images and natural language makes it difficult for VLMs to grasp the geometric structure of a space, causing a problem where accurate target coordinates cannot be inferred in tasks requiring high spatial precision, such as collision avoidance and systematic alignment.

Therefore, this study constructs a system that infers an alignment policy from preceding human actions, and formulates and verifies a task-level placement plan accordingly. Using visual prompts with overlaid grids, we enable the VLM to recognize absolute coordinates. Concurrently, we introduce verification based on geometric calculations and an autonomous self-correction loop for the inferred coordinates to ensure placement reliability. Experimental results confirmed that our system enables the VLM to accurately extract classification and alignment rules from human actions, realizing both the precise placement of new objects at specific locations and systematic arrangement operations.

Keywords: Social and Socially Assistive Robotics, Cognitive Human Robot Interaction, Neurorobotics
Abstract: Understanding whether people attribute mental states to robots during realistic conversations remains an open question in human–robot interaction. In this study, we analysed a large corpus of human–human and human–robot conversations including neurophysiological recordings, using an LLM–based conversation tagger to automatically detect utterances involving mental-state reasoning in the conversation transcript. We found that when participants listened to robot speech tagged as ToM-related, activity increased in core mentalizing brain areas—including the left TemporoParietal Junction (TPJ), and bilaterally in the Medial Frontal and Posterior Cingulate cortices—whereas the equivalent contrast for human speech did not elicit any significant response. These findings suggest that in natural interactive settings, humans recruit brain’s ToM network only when processing robot-produced mental-state content, indicating that adopting an intentional stance toward robots is, in sharp contrast to humans, triggered by specific speech contents.

Keywords: Motion Planning and Obstacle Avoidance, AI Reasoning Methods for Robotics
Abstract: We propose a stack-aware task and motion planning (TAMP) framework that rearranges heterogeneous objects within confined and cluttered environments. These settings involve vertical stacking and variations in object size and shape, which increase planning complexity. However, existing TAMP approaches typically assume uniform object geometries or rely on 2D abstractions that ignore detailed 3D stacking dependencies, leading to inefficient plans and frequent failures in tightly packed environments with densely stacked objects. To overcome these challenges, we introduce a rank-based task planner that assigns each potential relocation a lexicographically ordered score based on four factors: object rank, the number of objects stacked above it, buffer space clearance, and the number of blocking objects. The planner then selects the relocation with the lowest combined score, thereby reducing symbolic distance to the goal. During execution, the planner repairs the plan reactively by preserving valid subplans and applying local adjustments when collisions occur. Experimental results from both simulations and real robot experiments demonstrate higher success rates, reduced planning times, and improved object efficiency under tight spatial and stacking constraints.

Keywords: Motion Planning and Obstacle Avoidance, AI Reasoning Methods for Robotics, Industrial Robots
Abstract: Learning-based sampling can significantly accelerate sampling-based robot path planning, but learned models introduce a risk of failure if their predictions are inaccurate. In this work, we focus on a novel method to integrate a generative model with an RRT-based planner by automatically rejecting low-confidence model samples using an error-bound criterion. Unlike prior approaches that require a hand-tuned ratio between biased (learned) and uniform sampling to maintain probabilistic completeness, our method monitors the model’s performance on- the-fly and falls back to uniform sampling when the learned sampler is deemed unreliable. We demonstrate this approach on a 7-DOF Baxter robot arm scenario. The proposed method improves success rates and planning times under tight time constraints without manual parameter tuning.

Keywords: Motion Planning and Obstacle Avoidance, Contact: Modeling, Sensing and Control, Dynamics and Control
Abstract: This study proposes a high-efficiency flying trigger path planning technique for a coupled line-scan and area-scan Automated Optical Inspection (AOI) gantry system. While this coupled architecture reduces hardware costs and footprint, the continuous unidirectional motion of the line-scan axis introduces irreversible spatiotemporal constraints. These constraints render traditional Traveling Salesman Problem (TSP) algorithms ineffective. To address this, we model the dynamic flow of inspection windows as temporal logic constraints using a Directed Acyclic Graph (DAG). We then implement a two-stage hybrid planning strategy. First, Dynamic Programming (DP) constructs a global optimal path backbone specifically for Critical Defects. Subsequently, a Greedy Algorithm inserts secondary defects into the remaining time windows. Experiments using real production data from AUO Company and randomly distributed weighted coordinates demonstrate the method's efficacy. The approach achieved a weighted processing rate exceeding 95% on synthetic datasets and an average of 93.63% on industrial datasets, while reliably maintaining a high coverage for Critical Defect Weights. The results confirm that combining global optimization with local search effectively resolves flying trigger path planning issues under strict motion constraints and dynamic time windows.

Keywords: Intelligent Robotic Vehicles, Robotic Systems Architectures and Programming, AI Reasoning Methods for Robotics
Abstract: Standard local trajectory planners, such as the Dynamic Window Approach (DWA), often struggle in unstructured environments, frequently succumbing to local minima by entering dead-end traps. This paper presents a robust, ROS 2-native mapless navigation framework based on Deep Reinforcement Learning (DRL) designed to overcome these limitations. We introduce a Procedural Adversarial Trap Generator (ATG) within Gazebo to train a Proximal Policy Optimization (PPO) agent, forcing it to learn generalized recovery strategies for complex geometric obstacles. Unlike traditional planners that reactively navigate into cul-de-sacs, our proposed agent learns to proactively identify and circumnavigate geometric traps from the onset. Extensive experiments demonstrate that the DRL-based approach significantly outperforms the DWA baseline, achieving a success rate of 88% in complex maze scenarios where the baseline success rate drops to 8%. Furthermore, we validate the system's efficacy through a zero-shot transfer to a physical Unitree Go2 quadruped robot, demonstrating that the learned proactive avoidance behavior transfers robustly to the real world despite sensor noise.

Keywords: Motion Planning and Obstacle Avoidance, Dynamics and Control, Wheeled Mobile Robots
Abstract: In the rapidly evolving landscape of Industry 4.0 and smart societies, autonomous service robots have become pivotal components in ubiquitous mobility applications, ranging from healthcare assistance to smart manufacturing. However, ensuring stable operation under unknown payload variations and unstructured terrains remains a critical challenge. Unwanted vibrations and parametric uncertainties not only degrade tracking precision but also compromise safety during human-robot interaction. To address these issues, this paper proposes a novel hybrid control framework that synergistically integrates Zero Vibration Derivative (ZVD) input shaping with an Adaptive Sliding Mode Control (SMC) driven by Radial Basis Function Neural Networks (RBFNN). By leveraging computational intelligence, the RBFNN approximates unknown nonlinear dynamics online, thereby significantly mitigating the chattering phenomenon inherent in conventional SMC while enhancing system robustness. Simultaneously, the ZVD shaper effectively suppresses residual vibrations from the reference input. The effectiveness and feasibility of the proposed strategy are rigorously verified through both numerical simulations and real-world experiments on a laboratory robotic platform. The results demonstrate that the proposed approach achieves superior performance in trajectory tracking and vibration suppression compared to traditional methods, confirming its potential for practical deployment in dynamic environments.

Keywords: Wheeled Mobile Robots
Abstract: This paper presents the development of raising-up and laying-down control strategies for a half-drone (HD) wheeled inverted-pendulum (WIP) robot. A control architecture combining dynamic compensations and a LQR feedback regulator is proposed to manage stance transitions. The system's performance is validated through both numerical simulations and experimental trials. Results demonstrate that the robot can successfully transition between grounded and standing states. The study also identifies hardware constraints, such as propeller thrust saturation and feedback noise, and discusses potential mitigation strategies.

Keywords: Mechanism and Design, Wheeled Mobile Robots
Abstract: This paper proposes a new design for the Mobile Module (MoMo), a ceiling-climbing robot, to improve locomotion stability and speed. MoMo travels by mechanically engaging its rail-shaped body with ceiling-mounted protrusions and driving wheels pressed against the ceiling surface. However, friction at the rail–protrusion interface and geometric irregularities at rail joints increase resistance and degrade motion stability. Therefore, structural improvements are necessary to achieve high-speed, stable locomotion. To achieve high-speed, stable motion of MoMo, this paper proposes a two-layer rail structure. In addition, we design several protrusions equipped with thrust ball bearings, as well as a protrusion with a different shape and material. Experimental evaluations confirm that, compared with the conventional mechanism, the proposed configuration improves moving speed and reduces load during operation. Based on these findings, we discuss the effectiveness of each protrusion design and the optimal configuration for enhancing both speed and stability.

Keywords: Mechanism and Design, Wheeled Mobile Robots, Intelligent Robotic Vehicles
Abstract: The rapid growth of e-commerce has increased last-mile delivery demand, while labor shortages continue to raise costs and reduce logistics efficiency. Real delivery routes often include step-like discontinuities such as curbs and consecutive steps, which can constrain wheeled robots optimized for flat ground. This paper presents a delivery robot that switches between wheel mode for efficient flat-ground travel and track mode for stable step negotiation. The robot retracts its wheel modules using electric cylinders so that the tracked crawlers become the primary ground-contact elements during obstacle traversal. A depth-only step-height estimator uses depth images from an RGB-D camera to detect step risers as dominant vertical planes; estimated height is compared with a mechanical limit to trigger mode switching and a three-phase traversal behavior. Feasibility is validated through real-environment experiments involving step heights of 11-14 cm and up to five consecutive steps.

Keywords: Mechanism and Design, Wheeled Mobile Robots, Dynamics and Control
Abstract: This paper presents a T-spoke wheel design for stair-climbing mobile robots and a reinforcement learning (RL)-based locomotion control framework developed in NVIDIA Isaac Lab. Unlike conventional circular wheels that fail to gain traction on stair edges, the proposed T-spoke wheel features radially protruding spoke structures with a T-shaped cross-section, enabling mechanical engagement with stair steps. A parametric wheel model is constructed within Isaac Lab, allowing systematic variation of wheel radius while maintaining fixed T-profile dimensions. A four-wheel-drive (4WD) robot equipped with independently driven T-spoke wheels is trained using Proximal Policy Optimization (PPO) to autonomously climb stairs. A parametric study across five wheel radii (65--145,mm) reveals that the flange gap between adjacent spokes is the critical geometric factor for stair engagement. All T-spoke configurations with sufficient flange gap successfully climbed 7--8 steps on average, while the circular-equivalent baseline (65 mm) achieved only 1% success rate. The results confirm the geometric advantage of the T-spoke design and identify heading deviation as the primary remaining challenge for future improvement.

Keywords: Autonoums Vehicle Navigation, Robotic Systems Architectures and Programming, Motion Planning and Obstacle Avoidance
Abstract: This paper presents the integration and evaluation of an autonomous navigation system for a four-wheel steering (4WS) agricultural robot operating in opposite-phase steering mode within simulated blueberry plantation environments. The system combines opposite-phase steering kinematic modeling with wheel-based odometry fused via an Extended Kalman Filter, graph-based SLAM for environment mapping, Adaptive Monte Carlo Localization for pose estimation, Hybrid-A* path planning for kinematically-constrained trajectory generation, and Model Predictive Path Integral control for local motion execution, all orchestrated through a behavior tree framework within the ROS 2 Navigation Stack (Nav2). Three navigation tasks: perimeter traversal, inter-row serpentine coverage, and an extended multi-region circuit were conducted in a Gazebo-based simulation replicating blueberry row-crop geometry. The system completed all navigation goals without triggering recovery behaviors, validating the feasibility of the Nav2 framework for structured agricultural navigation.

Keywords: Underwater Robotics, Simultaneous Localization and Mapping (SLAM), Range, Sonar, GPS and Inertial Sensing
Abstract: Accurate underwater navigation remains challenging because GPS signals are unavailable underwater, leading to unbounded drift over time. Bathymetric simultaneous localization and mapping (SLAM) addresses this challenge by using relative pose constraints from multibeam echosounder (MBES) measurements. However, the reliability of these constraints can change greatly depending on the terrain and sensor settings. Fixed or manually tuned covariance values can make the optimization unstable if a few bad constraints dominate and distort the estimated trajectory. In this paper, we propose a learning-based uncertainty modeling framework that predicts constraint reliability from registration quality indicators and assigns adaptive weights for pose-graph optimization. Our method keeps a physics-based error-state EKF front-end and standard geometric registration, using learning only for modeling uncertainty. We test the approach on real-world ocean data collected in Pohang, South Korea, using GPS as ground truth. In a 2 km mission with 78 submaps, our method achieves a drift of 0.52% distance traveled, avoiding catastrophic failures seen with naive uniform weighting (3.72% drift) and reaching near the performance of expert-tuned heuristics (0.42% drift). These results show a practical path to robust bathymetric SLAM while keeping the estimation structure interpretable.

Keywords: Underwater Robotics, Dynamics and Control, Motion Planning and Obstacle Avoidance
Abstract: This paper proposes a robust autonomous control framework for the close-proximity inspection of cylindrical marine infrastructures, such as subsea pipelines and offshore platform columns. Close-proximity maneuvers in underwater environments present significant challenges due to nonlinear 6-degrees of freedom (DOF) hydrodynamics, unpredictable flow currents, and payload-induced uncertainties. Traditional model predictive control (MPC) often lacks adaptivity to time varying environments when using fixed cost weights, while pure model-free reinforcement learning (RL) struggles to guarantee safety constraints and stability during training and deployment. To address these limitations, we present Marine Actor-Critic Model Predictive Control (Marine AC-MPC), which integrates a differentiable iterative linear–quadratic regulator (iLQR)-based MPC as the final layer of a neural network policy. In this framework, the actor learns to dynamically adapt MPC cost parameters (time-varying weights) based on real-time observations, while the critic estimates long-horizon performance to guide the learning process. This design preserves the structured constraint-handling of MPC, ensuring adherence to sensor standoff, footprint, and collision-avoidance constraints, while enabling the high adaptivity of RL under disturbances. Quantitative evaluations conducted in MarineGym, a high fidelity underwater simulator, demonstrate that the proposed method outperforms fixed-weight MPC and RL-based proximal policy optimization (PPO) baselines. Notably, Marine AC MPC maintains robust tracking performance and minimizes unnecessary corrective actions even under rapidly shifting current directions and magnitudes. Our results suggest that the integration of structured control and learned adaptivity provides a reliable path toward safe and precise autonomous underwater inspection in complex environments.

Keywords: Underwater Robotics, Autonoums Vehicle Navigation, Intelligent Robotic Vehicles
Abstract: In underwater geomagnetic mapping, autonomous missions face the challenge of navigational drift, which can significantly degrade the quality of the generated maps. This paper presents a Gaussian Process (GP)-based framework for robust magnetic intensity mapping that mitigates the impact of inertial navigation drift over mission time. We introduce a drift-aware kernel that combines spatial and temporal components to account for navigation misalignments. By simulating an Autonomous Underwater Vehicle (AUV) mission, we evaluate the performance of our proposed method against traditional GP regression models. Our results demonstrate that the drift-aware kernel improves map accuracy under navigation drift by modeling drift-induced inconsistencies. The proposed method is validated through quantitative metrics and shown to outperform traditional spatial-only kernels, offering a more robust solution for underwater geomagnetic intensity mapping.

Keywords: Underwater Robotics, AI Reasoning Methods for Robotics, Intelligent Robotic Vehicles
Abstract: We introduce a language-guided mission synthesis framework for adaptive water quality monitoring using autonomous surface vehicles (ASVs). The proposed system enables operators to specify monitoring objectives through structured natural language instructions, which are automatically translated into executable mission plans. Our framework combines a constrained task grammar with a fine-tuned lightweight sequence-to-sequence model (T5-small) that maps language commands to structured mission tokens. These tokens are then compiled by a deterministic interpreter into adaptive mission plans capable of dynamic refinement during execution. To evaluate the approach, we conduct simulation studies using real-world water quality data collected from a physical ASV platform. Experimental results demonstrate high translation accuracy and reliable plan compilation across diverse monitoring scenarios. The system supports interactive mission generation while preserving deterministic execution and safety constraints. This work demonstrates the feasibility of language-driven adaptive mission planning for aquatic environmental monitoring and provides a practical interface for bridging human intent and autonomous ASV operations.

Keywords: Underwater Robotics, Intelligent Robotic Vehicles
Abstract: Long-term in situ monitoring of aquatic environments using autonomous surface vehicles (ASVs) is often limited by onboard battery capacity, which constrains operational duration. This study proposes an energy-adaptive station-keeping control algorithm for a solar-powered ASV that dynamically switches control strategies based on the battery state of charge (SOC).

The proposed method defines three operating modes according to SOC thresholds: a high-accuracy mode with a small tolerable distance, a standard mode balancing accuracy and energy efficiency, and a power-saving mode with an enlarged tolerable distance to reduce energy consumption. A hysteresis mechanism is introduced to prevent frequent mode switching. By adapting the tolerable distance in real time according to energy availability, the algorithm addresses the trade-off between station-keeping accuracy and endurance.

Simulation experiments using Webots were conducted under three irradiance scenarios representing seasonal and cloudy conditions in the Lake Biwa region. Results show that the proposed method achieves 24-hour continuous operation under summer clear-sky conditions while maintaining moderate station-keeping accuracy, outperforming fixed high-accuracy and fixed low-power strategies in overall energy-performance balance.

Keywords: Deep Learning for Visual Percepton, Object Recognition, AI Reasoning Methods for Robotics
Abstract: Vision-language models (VLMs) are increasingly used as perceptual backbones for assistive robots, yet domestic healthcare settings require both accurate localization and visually grounded, trustworthy explanations. This paper evaluates three VLMs, PaliGemma-2, Qwen2-VL, and Florence-2, for object detection and explanation faithfulness in cluttered scenes. A custom Intel RealSense D435i dataset was captured, annotated with bounding boxes, and expanded to 1,250 images using controlled geometric and photometric augmentation. Object-centric explanation labels were generated using an automated, constraint-based pipeline to enforce spatial anchoring and linguistic consistency. Models were fine-tuned and evaluated using mean IoU and mAP@50, together with a modified Semantic Visual Fidelity (SVF) score. Florence-2 achieved the best overall performance (IoU 0.9517, mAP@50 0.992, SVF 0.7100), indicating that stronger localization correlates with higher explanation faithfulness.

Keywords: Deep Learning for Visual Percepton, Computer Vision and Visual Servoing
Abstract: This study addresses the partial-to-complete geometry reconstruction of deformable objects (DOs) from point-cloud observations toward precise DO manipulation. Recent DO reconstruction approaches often adopt implicit neural representations (INRs) to model continuous surfaces as well as capture structural variability. However, these methods typically rely on object-specific shape priors that improve training stability and limit generalization. To figure it out, we introduce ParCo-SDF, a two-stage partial-to-complete signed distance field (SDF) reconstruction framework consisting of temporal geometry encoding followed by FiLM-conditioned SDF prediction. The temporal encoder captures structural similarity across DO sequence, enabling prior-free stable training. FiLM-based conditioning preserves reconstruction expressivity while reducing network complexity. We evaluate the proposed method against a state-of-the-art DO surface reconstruction baseline on a rubber band manipulation dataset, demonstrating robust and high-fidelity reconstruction under severe occlusions.

Keywords: Computer Vision and Visual Servoing, Object Recognition, Performance Evaluation and Optimization
Abstract: Long-term identity consistency is essential for providing personalized services in a smart space where multiple people move freely and interact with each other. Multi-object tracking (MOT) methods frequently suffer from ID switches due to occlusions and close approaches, which disrupt the continuous collection of behavioral histories. This paper proposes a view-aware re-identification (ReID) strategy that improves ID consistency by restricting appearance matching to frontal-view observations. Unlike conventional approaches that utilize features from all viewpoints, the proposed method selectively performs ReID only when a person is observed from the front, thereby stabilizing appearance representation and reducing mismatches. We integrate the proposed ReID module into three state-of-the-art trackers: OC-SORT, BoT-SORT, and ByteTrack. Experimental results show that the integrated module improves performance with HOTA improving by up to 26.2% compared to the original trackers.

Keywords: Deep Learning for Visual Percepton, Manipulation Planning and Control, Computer Vision and Visual Servoing
Abstract: This study proposes a three-dimensional mosaic art generation system using a robotic arm. Conventional generation methods are often limited to two-dimensional representations or face challenges regarding physical stability during stacking. To address these issues, we developed a planning method that integrates image segmentation using the Segment Anything Model 2 (SAM2) and depth estimation using Depth Anything V2. This method enables the stacking of parts based on the semantic regions of the image. This approach simultaneously reduces the risk of structural collapse and enhances the visual three-dimensional effect. Real-world experiments demonstrated the system's ability to reproduce the "Mona Lisa" from a set of unknown parts and autonomously construct the 3D structure.

Keywords: Deep Learning for Visual Percepton, Object Recognition, Wheeled Mobile Robots
Abstract: Egocentric cameras are widely used in robotic navigation and manipulation, yet conventional 2D Video Object Tracking (VOT) methods suffer from severe performance degradation under rapid viewpoint changes and frequent frame-out events. Because most existing trackers rely solely on 2D appearance cues, they often fail to recover object identities once targets temporarily disappear. We propose R3DVOT, a 3D-aware tracking framework that augments 2D vision foundation models with spatial geometric reasoning. Its core component, the Position-aware Memory Selection Module (PMSM), lifts mask candidates into a canonical 3D world coordinate system and maintains a persistent world-frame state for position-consistent hypothesis selection. By curating the memory bank with spatially consistent anchors, R3DVOT improves robustness to occlusion and frame-out events. On the VOT benchmark, R3DVOT achieves an AUC improvement of 5.1% over SAM 2 and 1.4% over SAMURAI. Furthermore, on the VOS benchmark, R3DVOT increases the J&F score by 4.6% compared to SAM 2 and 9.0% compared to SAMURAI. These results highlight the effectiveness of 3D spatial continuity in enhancing tracking, segmentation, and long-term identity consistency in robotic perception.

Keywords: Computer Vision and Visual Servoing, Deep Learning for Visual Percepton
Abstract: Autonomous ground robots operating in agricultural environments require reliable perception to navigate safely under unstructured and visually challenging conditions. Oil palm plantations present difficulties due to dense vegetation, uneven terrain, and highly variable illumination. This paper presents a deployment-oriented multi-class semantic segmentation pipeline based on YOLOv8, designed for ground level navigation in plantation-like outdoor environments. Five semantic classes relevant to robot navigation which are oil palm trees, drains, drain edges, uphill terrain, and puddles, are defined and annotated to support terrain understanding and navigation safety. The pipeline integrates multi-class YOLOv8 segmentation with a flexible inference design and illumination-aware preprocessing, where Contrast Limited Adaptive Histogram Equalisation (CLAHE) is conditionally applied during inference under low-light conditions. Qualitative evaluation in a controlled outdoor environment resembling plantation settings demonstrates stable and meaningful segmentation outputs across varying illumination and terrain scenarios, with improved reliability under low-light conditions when conditional preprocessing is enabled. The proposed approach emphasises practical deployment considerations and provides a foundation for integrating semantic perception into autonomous ground robot navigation in plantation environments.

Keywords: Computer Vision and Visual Servoing, Deep Learning for Visual Percepton, Underwater Robotics
Abstract: Real-time image segmentation for underwater robotic systems remains a challenging task due to limited onboard computational resources and harsh underwater imaging conditions. While deep learning-based segmentation methods can achieve high accuracy, they typically require GPU acceleration, which makes real-time deployment difficult on low-power underwater platforms. In this paper, we propose a lightweight, unsupervised underwater image segmentation pipeline without neural networks. The proposed method consists of fast pixel bucketing using locality-sensitive hashing (LSH), K-means clustering, Jensen–Shannon divergence (JSD) based cluster similarity analysis, and dendrogram-based hierarchical meta-group construction. Our system employs a four-level hierarchy—Cluster, Grouped Cluster, Meta-group, and Annotation—enabling a structured workflow from pixel-level segmentation to semantic-level object categorization. Experimental results show that the proposed approach achieves effective separation of divers from background across diverse underwater scenarios, including a water-tank setup, open-water environments, and synthetic data generated by a generative model. We further integrate the system into a ROS2-based real-time processing pipeline and observe approximately 6 fps for the clustering and grouping stages, and about 2.5 fps when including the classification stage.

Keywords: Computer Vision and Visual Servoing, Foundations of Sensing and Estimation, Robotics in Hazardous Applications
Abstract: Perception in dark or texture-less environments remains a major challenge for mobile robots, where conventional photogrammetric pipelines fail due to insufficient illumination and unreliable feature correspondences. In extreme scenarios such as lunar or underground exploration, both lighting and power are severely constrained. To address this issue, we propose an event-only Structure from Motion and Multiview Stereo (SfM-MVS) framework using ultraviolet-activated phosphorescent materials as temporary artificial features in complete darkness. An event camera captures intensity changes emitted by the phosphorescent material, and the raw event data are reconstructed into image sequences using an Event-to-Video (E2VID), which is a Recurrent Neural Network (RNN)-based method. The refined frames are then processed through SfM and MVS for three-dimensional (3D) reconstruction. Experimental results demonstrate that the proposed framework enables effective 3D reconstruction using phosphorescent materials in extremely low-light environments.

Keywords: Manipulation Planning and Control, Motion Planning and Obstacle Avoidance, Intelligent Robotic Vehicles
Abstract: Roadside mowing with vehicle-mounted hydraulic manipulators is a safety-critical maintenance task that remains largely manual due to highly coupled challenges in terrain contact, geometric variability, and obstacle avoidance. This paper studies autonomous control of a 5-DoF mowing arm mounted on a slowly moving carrier vehicle, where the vehicle motion is externally driven and the manipulator must track a target mowing region while maintaining continuous ground contact and avoiding obstacles. We develop a high-fidelity simulation environment in Isaac Lab with procedurally generated terrains and obstacle configurations that emulate common roadside conditions. To bridge the sim-to-real sensing gap, we adopt a teacher--student knowledge distillation framework: a privileged teacher policy is trained with terrain and contact information, and a deployable student policy is distilled using only onboard sensing, including a stereo depth image and proprioceptive signals. Domain randomization over terrain difficulty and system parameters is applied to improve robustness. Simulation results demonstrate accurate mowing-head tracking, stable terrain contact, and effective obstacle avoidance across diverse terrain geometries and randomized dynamics, indicating the feasibility of vision-based learning for autonomous roadside mowing.

Keywords: Force and Tactile Sensing, Mechanism and Design, Robotic Systems Architectures and Programming
Abstract: This paper presents a compact, multi-axis contact force sensing mechanism tailored for buoyancy-assisted robots (BARs), addressing the need for lightweight tactile sensing in constrained environments. The proposed sensor integrates soft elastomeric materials with an embedded 3-axis Hall effect sensor to estimate ground reaction forces in a small form factor accurately. A computational design framework is introduced, combining Neo-Hookean elastomer modeling with magnetic field simulation to identify and estimate sensor performance. Two force prediction models are evaluated: a simulation-data-trained Gaussian Process Regression (GPR) model for design-time estimation, and a hardware-data-trained Multilayer Perceptron (MLP) for real-time inference. Experimental validation on the BALLU platform demonstrates high accuracy across lateral and axial force directions, with the MLP achieving superior performance in complex, coupled deformation scenarios. This dual-model approach enables robust sensor design and deployment, advancing tactile sensing capabilities for safe and agile human-robot interaction.

Keywords: Object Recognition, Haptics
Abstract: Ubiquitous companion robots offer a promising avenue for immediate anxiety relief in children, yet their effectiveness relies on the ability to monitor physiological states continuously and unobtrusively. Current solutions often depend on external wearables, which impose usability barriers and limit the robot's autonomy. This paper investigates the integration of an embedded photoplethysmography (PPG) sensor directly into a pocket-sized companion robot, AffectaPocket, to enable self-contained heart rate monitoring during tactile interaction. We address the significant challenge of motion artifacts inherent in handheld usage by implementing a two-stage filtering pipeline that utilizes an onboard Inertial Measurement Unit (IMU) to reject high-variance segments and a confidence-based smoothing algorithm for recovery periods. We evaluated the system against a commonly used wrist worn sensor in a Within-Subjects Study with 26 participants. Our results demonstrate that the filtering strategy significantly reduced the Mean Absolute Percentage Error and achieved statistical equivalence to the ground truth measurements (p<0.05). Analysis of short-duration interactions shows that the sensor requires stability over longer periods to converge.

Keywords: Force and Tactile Sensing, Contact: Modeling, Sensing and Control, AI Reasoning Methods for Robotics
Abstract: Tactile roughness perception is commonly formulated as either classification or regression, implicitly assuming that tactile signals can be grounded to a stable absolute scale. In practice, however, tactile measurements are highly sensitive to sensing variability, making this numerical grounding unreliable under limited supervision or domain shift.

In this paper, we investigate how the choice of learning objective affects robustness when absolute grounding is unstable. We compare classification, regression, and ranking-based learning on a sandpaper dataset with 5 grit levels and 9,595 samples. Under well-conditioned in-domain settings, all paradigms achieve comparable performance. However, when supervision is scarce (e.g., 5% training data) or when substantial domain shift is introduced through changes in sensor hardware and exploration protocol, their behavior diverges.

Under domain shift, regression exhibits a pronounced collapse in ordinal consistency, whereas ranking-based learning preserves relative ordering more effectively, outperforming regression by 0.290 in Spearman rank correlation. These findings suggest that directly optimizing ordinal consistency yields more robust generalization when tactile roughness labels provide reliable ordering but unstable numerical scales.

Keywords: Computer Vision and Visual Servoing, Force and Tactile Sensing, Contact: Modeling, Sensing and Control
Abstract: Rapid detection of slip and contact deformation in visuotactile sensing requires low-latency estimation of contact-surface motion. Conventional frame-based cameras, however, struggle to capture fast and transient contact phenomena due to limited temporal resolution and motion blur. This paper proposes a low-latency method to estimate two-dimensional optical flow from asynchronous event streams of an event camera. We first derive normal-flow constraints for individual events by local spatiotemporal plane fitting, then aggregate them per grid cell within a Lucas--Kanade framework to resolve the aperture problem and recover full 2D flow. To improve robustness, we discard ill-conditioned estimates in degenerate regions using a confidence measure based on the structure-tensor condition number, and enable online processing via a sliding-window update that incrementally adds and removes constraints. Experiments with an event-camera-based visuotactile sensor show characteristic flow fields consistent with applied normal force, shear force, and rotational torque (convergence/divergence, translation, and rotation). Processing-time measurements further confirm 1,ms-scale flow updates on a 6-thread CPU over a practical range of event rates.

Keywords: Force and Tactile Sensing, Contact: Modeling, Sensing and Control, Soft Robotics
Abstract: High sensitivity, large area tactile sensing, and reliable contact localization are required for safe and adaptive human-robot interaction. Robots need high sensitivity and precise localization to distinguish harmless contact from potentially dangerous interactions and operate safely near humans. Microphone based tactile sensors are studied for their wide sensing range and for capturing dynamic contact events. Earlier studies used the high sensitivity and wide dynamic range of microphones to fabricate large area tactile skins. Yet, most localization algorithms assume a planar sensor surface, which limits their application to robots based on non-planar and curved geometries. This work proposes a contact localization framework for microphone-embedded tactile sensors on non-planar surfaces. This method uses geometric mapping of the sensor surface to compute the geodesic distance matrix of the shortest surface paths between candidate contact points and embedded microphones. Contact vibrations travel along this contact layer and are measured in intensity patterns across microphones, which are compared with precomputed geodesic matrices to estimate the most probable contact location. The approach is verified with a hemispherical sensor having positive Gaussian curvature. The performance is evaluated through indentation experiments using root mean squared error analysis. By extending microphone-based tactile sensing beyond planar assumptions, this work achieved 87.20% localization accuracy on non-zero Gaussian surfaces and laid the foundation for accurate contact localization on curved robotic structures.

Keywords: Legged Robots, Mechanism and Design, Actuation and Actuators
Abstract: Physical joint limits are common in legged robots and can restrict workspace, constrain gait design, and increase the risk of hardware damage. This paper introduces MiNI-Q, a miniature, wire-free quadruped robot with independently actuated, mechanically unbounded 2-DOF leg joints. We present the mechanical design, kinematic analysis, and experimental validation of the proposed robot. The leg mechanism enables both oscillatory gaits and rotary locomotion while allowing the robot to fold to a minimum height of 2.5 cm. Experimentally, MiNI-Q achieves speeds up to 0.46 m/s and demonstrates low-clearance crawling, stair climbing, inverted locomotion, jumping, and backflipping. The wire-free architecture extends our previous Q8bot design, improving assembly reliability at miniature scale. All mechanical and electrical design files are released open source to support reproducibility and further research.

Keywords: Legged Robots, Dynamics and Control
Abstract: Deep reinforcement learning has enabled quadrupedal robots to traverse challenging terrains, yet energy efficiency remains a limiting factor for prolonged autonomous operation. Most existing frameworks rely on fixed or adaptively tuned proportional-derivative (PD) controllers that operate exclusively in the joint space. Such approaches typically lack an explicit mechanism for contact compliance, often applying excessive torque on benign terrains while providing insufficient absorption of reaction forces on irregular surfaces. To address these limitations, we propose HIP, a hybrid impedance and PD control framework that fuses joint-space PD control for trajectory tracking with task-space impedance control for contact compliance. The impedance term, mapped to joint torques via the Jacobian transpose, models compliant foot-tip behavior that absorbs impact energy during ground contact rather than resisting it through rigid control. To coordinate the two control modalities, we further introduce the attention for representation combiner (ARC) network. The ARC network employs a cross-attention mechanism between a gain actor and a joint actor, enabling control gains and desired joint positions to be generated in a coordinated manner. A state estimator augmented with a per-leg stumble estimator provides additional proprioceptive context to both actors for proactive gain adaptation. Simulation experiments across diverse terrains demonstrate that HIP achieves velocity tracking accuracy comparable to existing baselines while delivering improved energy efficiency. Torque decomposition analysis further confirms that the impedance component effectively reduces torque peaks during contact events.

Keywords: Legged Robots, Computer Vision and Visual Servoing, Autonoums Vehicle Navigation
Abstract: Research in robot navigation has been accelerating rapidly with the advent of vision-language models (VLMs) that can explicitly reason about the diverse variables present in real-world driving scenarios. Despite the growing success of AI-based robot navigation, existing datasets and policies remain tightly coupled to specific hardware embodiments, which limits cross-platform generalization. As a result, platforms for which data are substantially harder to collect and far less publicly available than for wheeled robots, such as legged robots, remain chronically under-represented, aggravating the dataset imbalance across robot morphologies. To resolve this data discrepancy, we propose XNav-Pipe, a 2-staged synthetic data generation framework that expands specific-robot driving datasets to general-purpose navigation across robot types. In Stage 1, we imitate prior wheeled-robot driving behaviors using lightweight PPO agents to re-generate data on novel platforms. In Stage 2, we leverage commonsense-aware VLMs to synthesize new navigation behaviors from given instructions and trajectories. Our framework, built on NVIDIA Isaac Lab, allows scalable and diverse simulation of robot navigation across morphologies.

Keywords: Legged Robots, Robotic Systems Architectures and Programming, Mechanism and Design
Abstract: This paper presents the design and pilot implementation of a three-week, co-developed legged robotics workshop for high school students using the Q8bot miniature quadruped platform. The workshop integrated mechanical assembly, fabrication, and software exploration to support participation across a range of prior technical experience. Field observations, student surveys, and peer interviews were collected to examine how students engaged with the robot as a system-level engineering platform. Across three student groups, participants successfully assembled functional robots and engaged in iterative design, debugging, and software modification tasks. Observations from the workshop suggest that the compact, open-source Q8bot platform enabled hands-on interaction with core robotics concepts such as design-for-manufacturing, motion generation, and embedded system constraints. Students demonstrated varying pathways of engagement, from high-level behavioral tuning to deeper architectural modifications. While this study does not attempt to measure formal learning gains, the results provide initial evidence that small-scale, open-source legged robots can support experiential engineering practices and foster systems-oriented thinking in secondary education settings. These findings inform the design of future classroom implementations and highlight directions for more rigorous evaluation of educational outcomes.

Keywords: Foundations of Sensing and Estimation, Mechanism and Design
Abstract: The diffuse and specular reflection of plant leaves, like many materials, exhibits strong angular dependencies that pose challenges for quantitative spectroscopy in agricultural applications. For quantitative use of spectral data, for example in chemometric models of chlorophyll content, such angular effects need to be understood and controlled. We present a robotic point spectroscopy system that integrates a UV–VIS spectrometer with relay optics on a 6-DoF manipulator, enabling angular sweeps of the observation direction from 30◦ to 150◦ while maintaining a 1 mm measurement spot within a 5 mm radius on the leaf surface at a working distance of 100 mm. Using this system, we acquire angle-resolved spectra from basil and coleus leaves and show that, for basil, the normalized difference vegetation index (NDVI) for a fixed leaf location can vary by up to 71% depending on observation angle, whereas coleus exhibits markedly weaker angular dependence. These results indicate that measurement geometry alone can induce significant index variations and highlight multi-angle, single-point robotic spectroscopy as a useful basis for collecting controlled datasets for future quantitative modeling.

Keywords: Performance Evaluation and Optimization, Object Recognition, Foundations of Sensing and Estimation
Abstract: This paper investigates person detection and tracking in an industrial indoor workspace using a LiDAR mounted on an overhead crane. The overhead viewpoint introduces a strong domain shift from common vehicle-centric LiDAR benchmarks, and limited availability of suitable public training data. Henceforth, we curate a site-specific overhead LiDAR dataset with 3D human bounding-box annotations and adapt selected candidate 3D detectors under a unified training and evaluation protocol. We further integrate lightweight tracking-by-detection using AB3DMOT and SimpleTrack to maintain person identities over time. Detection performance is reported with distance-sliced evaluation to quantify the practical operating envelope of the sensing setup. The best adapted detector configurations achieve average precision (AP) up to 0.84 within a 5.0 m horizontal radius, increasing to 0.97 at 1.0 m, with VoxelNeXt and SECOND emerging as the most reliable backbones across this range. The acquired results contribute in bridging the domain gap between standard driving datasets and overhead sensing for person detection and tracking. We also report latency measurements, highlighting practical real-time feasibility. Finally, we release our dataset and implementations in GitHub to support further research.

Keywords: Performance Evaluation and Optimization, Foundations of Sensing and Estimation, Deep Learning for Visual Percepton
Abstract: A cappella rehearsals often rely on subjective listening to judge pitch accuracy and inter-part balance, which can be difficult for novice singers to assess consistently. This study proposes an objective singing-voice evaluation system for a cappella performance that integrates score information (MusicXML) with audio analysis. Based on an open-ended questionnaire survey with experienced university a cappella singers, we focus on two perceptually important factors: pitch accuracy and inter-part loudness balance. The system estimates pitch trajectories using pYIN and CREPE, applies temporal smoothing and MIDI quantization, and computes pitch deviations with respect to score-aligned reference notes. For volume balance evaluation, frame-level RMS is calculated for each part and normalized with rest-aware masking derived from the score, enabling time-varying relative level ratios across parts. The results are visualized in an intuitive GUI that highlights problematic regions along the performance timeline. Listening experiments using controlled degradation conditions show positive correlations between the proposed objective measures and participant ratings, suggesting that the system provides interpretable feedback and supports effective self-assessment during a cappella practice.

Keywords: Foundations of Sensing and Estimation, Performance Evaluation and Optimization, Intelligent Robotic Vehicles
Abstract: Object localization in indoor environments is a fundamental enabling technology for Digital Transformation and service robotics. However, conventional Indoor Positioning Systems require the installation of fixed anchors, which constitutes a major obstacle to deployment. To address this issue, anchorless positioning approaches using Autonomous Mobile Robot (AMR) have been studied to maintain localization accuracy without fixed infrastructure. Nevertheless, in environments where uncertainty remains, for example when the localization distribution spreads predominantly in a specific direction due to environmental structure or observation geometry, the trajectory of the AMR significantly affects localization performance. This study proposes a trajectory planning method that statistically analyzes the shape of the localization distribution to improve localization accuracy. Principal component vectors are extracted from the particle distribution obtained by a particle filter, and the AMR trajectory is sequentially determined to reduce dominant uncertainty. Simulation experiments assuming an indoor environment were conducted, and comparisons were made with a conventional circular trajectory. The performance was evaluated using the magnitude of the localization error, the time required to reach the error threshold, and spatial stability. The results demonstrate that the proposed method suppresses localization error growth and achieves stable and high-accuracy localization compared with the conventional approach.

Keywords: Physical and Cognitive Human-Robot Interaction, Foundations of Sensing and Estimation, Performance Evaluation and Optimization
Abstract: Accurate and low-latency sound source localization is essential for natural human–robot interaction, as it enables timely gaze shifts, body orientation, and responsive nonverbal behaviors. However, many deep-learning-based direction-of-arrival (DoA) methods prioritize accuracy over real-time deployment and often rely on computationally expensive intermediate representations such as template matching. This paper presents DV-SSL, a lightweight real-time 3D DoA estimation framework designed for humanoid robot platforms. Instead of estimating intermediate direct-path features, the proposed model directly regresses a Cartesian unit direction vector, eliminating grid-based post-processing and reducing computational overhead. The architecture incorporates (1) a stateful GRU-based streaming design for continuous inference, (2) frequency pooling to mitigate narrow-band computational complexity, and (3) frequency attention pooling for adaptive band aggregation under noisy and reverberant conditions. The proposed system is trained using large-scale simulated acoustic environments and validated through real-world deployment on a humanoid robot equipped with a multi-channel microphone array. Experimental results demonstrate stable real-time operation and practical 3D localization performance across diverse acoustic conditions, highlighting the suitability of DV-SSL for embodied interactive robotic systems.

Keywords: Simultaneous Localization and Mapping (SLAM), Computer Vision and Visual Servoing
Abstract: The simultaneous localization and mapping community has introduced a growing number of systems adapted for multi-session operations where the operational environment features low-dynamic changes that impact mapping, such as surface wear, weather phenomena, or seasonal change. These systems allow for lifelong operations by a robot within these environments. There is also growing interest in operations in environments where the unique ground texture is the only mapping feature available for use. These ground texture systems are not yet targeted for multi-session low-dynamic-change environments though. This work explores the impact of three different techniques on trajectory estimation accuracy in these multi-session low-dynamic ground texture environments. Of the three, the use of Kullback-Leibler Divergence, as a similarity score and a bias influencing loop closure confidence, is found to have the most success. We show an analysis of all three methods and a deeper exploration of the impact of Kullback-Leibler Divergence. We also introduce a dataset for use by the robotics community that contains multi-session images where the ground changes between sessions and also high-accuracy pose information for use in evaluation.

Keywords: Simultaneous Localization and Mapping (SLAM)
Abstract: Feature matching is a fundamental task in computer vision, essential for applications such as SLAM and 3D reconstruction. However, achieving high accuracy, particularly with complex deep learning models, often incurs prohibitive computational costs. In this paper, we propose a novel feature matching method that integrates an Edge Feature Gate Mechanism into the Joint Mamba architecture. The Joint Mamba framework is selected for its superior computational efficiency. By incorporating the Edge Feature Gate, our method selectively emphasizes structural edge information, thereby enhancing matching precision without significantly increasing the computational burden. This mechanism enables the model to effectively capture geometric details while suppressing computational costs. Consequently, our proposed method realizes high-accuracy matching performance while maintaining efficiency.

Keywords: Computer Vision and Visual Servoing, Performance Evaluation and Optimization, Simultaneous Localization and Mapping (SLAM)
Abstract: Indoor surveillance requires structuring human activities into searchable logs by who, where, and what. Al- though mobile drones provide flexible viewpoints compared to fixed cameras, existing approaches typically operate in image coordinates and do not jointly evaluate metric localization, open-set identification, and action estimation in a unified 3D map coordinate system. This paper presents an integrated framework that processes monocular indoor drone video to estimate camera poses with respect to a pre-built 3D map, computes person ground-plane positions (Where) via geometric back-projection onto the floor plane in map coordinates, and incorporates open-set face-based identification (Who) together with vision-language-model-based action tagging of posture and object-holding state (What), thereby generating surveillance logs directly in the map coordinate system. The contributions are (1) metric ground-plane localization evaluated in map coordinates, (2) open-set identification evaluation including unknown individuals, (3) quantitative evaluation of posture and object-holding tags, and (4) a unified protocol that evaluates metric localization and semantic attributes under a shared map coordinate system. Experiments in real indoor environments demonstrate median localization errors of 5.4–7.7 cm. Action tagging achieves 100% accuracy within the defined discrete action space and controlled evaluation conditions, and identi- fication attains 99.1% for enrolled identities but decreases to 70.0% for unknown individuals, indicating that robust open-set recognition remains the primary challenge.

Keywords: Simultaneous Localization and Mapping (SLAM), Multisensor Data Fusion, World Modelling
Abstract: Handheld LiDAR-IMU mapping enables rapid indoor capture, but the resulting point cloud remains unstructured for downstream as-built modeling. This paper presents a portable Livox MID-70 and Sony Spresense scanner, a ROS 2 stack that compensates a residual LiDAR-IMU clock offset by correcting the per-frame timebase, and online mapping with FAST-LIO2. We also provide a semi-automatic GUI for ROI-driven, seed-based fitting of planes and cylinders and export of lightweight mesh proxies. We evaluate mapping self-consistency on three indoor closed-loop courses and primitive fitting on the generated maps. Cylinder fitting on six pillars achieves MAE 10.6 mm in radius and 75.5 mm in height, and planar extent estimation on six rectangular specimens yields a mean error of 7.806% under ROI perturbations.

Keywords: Multisensor Data Fusion, Autonoums Vehicle Navigation, Simultaneous Localization and Mapping (SLAM)
Abstract: Non-linear optimization in visual-inertial odometry (VIO) significantly enhances localization accuracy and robustness for autonomous robots and MAVs in dynamic environments. However, visual tracking instability, often induced by motion blur or abrupt lighting changes, frequently causes trajectory drift, compromising precise localization. The proposed method maintains visual reprojection constraints within an inner window, while applying marginalized relative pose constraints in an outer window, seamlessly linked by preintegrated inertial measurements. To mitigate uncertainties commonly observed in low-cost sensors, our inertial preintegration explicitly models axis misalignment, scale factors, and g-sensitivity. Crucially, as keyframes exit the inner window, reprojection costs are transformed into relative pose constraints with accurately computed full covariance matrices. Explicitly propagating this full covariance improves estimator consistency and prevents overconfident updates during sustained tracking failures. Furthermore, the bounded double-window architecture strictly limits the joint state dimension, ensuring computational scalability without the excessive overhead typical of global optimization methods. In addition, evaluations on the EuRoC dataset demonstrate that our algorithm outperforms widelyused methods, effectively handling highly noisy and visually blind segments.