Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance
Abstract: Lower-limb exoskeletons offer potential to support balance recovery during perturbed walking, yet the biomechanical effects on dynamic stability remain poorly understood. This study investigated how wearing a 3-DoF ankle exoskeleton influences the recovery response and dynamic stability following forward falling slip perturbations during treadmill walking. Twelve healthy young adults experienced slip perturbations while walking with and without the exoskeleton. Anteroposterior margin of stability (MoS), whole-body angular momentum, and spatiotemporal gait parameters were evaluated for baseline walking and eight consecutive recovery steps without exoskeleton assistance. Results revealed significant effects of the recovery steps on all metrics, but no interaction between condition and recovery step. Wearing the exoskeleton did not affect anteroposterior MoS or step length. However, the exoskeleton induced a persistent increase in step width and led to localized changes in rotational dynamics on the perturbed side. These findings demonstrate that a 3-DoF ankle exoskeleton does not compromise sagittal plane stability during slip recovery, even without active assistance. They also highlight the importance of analyzing multi-step recovery dynamics when designing balance-assistive exoskeleton controllers, as the exoskeleton affects not only the first recovery step but also subsequent recovery steps.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance
Abstract: The knee extensor mechanism, comprised of the tibial tubercle, patellar tendon, patella, quadriceps tendon, and quadriceps femoris, is what allows the knee to extend. Proximal tibia reconstruction requires the knee extensor mechanism to be repaired. This surgery could lead to detrimental functional outcomes; however, the potential for actuated knee-ankle-foot-orthoses to reduce gait abnormalities for this population has been little studied. The aim of this study was to develop a methodology to assess the impact of actuated knee-ankle-foot-orthoses on lower limb joint kinematics and symmetry in patients who have undergone proximal tibia reconstruction. Three post proximal tibia reconstruction patients were recruited. The orthoses was customized to their specific gait characteristics before walking cycles were recorded using marker motion capture and ground reaction force. It was observed that using an actuated knee-ankle-foot orthoses with torque assistance increases surgical leg active knee range of motion (P1: 68.92°±1.82, P2: 82.58°±2.49, P3: 69.53°±3.63) in comparison to unassisted gait (P1: 60.39°±0.79, P2: 58.76°±2.85, P3: 64.55°±1.06). Additionally, gait speed decreased while using the orthoses. Future work will strengthen this study with additional participants.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance, Biologically-Inspired Robotics and Biomimetics
Abstract: Effective control of lower-limb exoskeletons depends on a comprehensive understanding of both human biomechanics and robotic actuation. We present a modular co-simulation framework that couples a physics-based exoskeleton model in Simscape with a musculoskeletal human model in OpenSim. The system supports bidirectional, stepwise interaction: joint angles from the exoskeleton model are sent to OpenSim, which returns corresponding interaction torques for use in the next simulation step. This closed-loop coupling enables the evaluation of assistive control strategies in a physiologically meaningful environment prior to hardware deployment. To demonstrate the framework's utility, we implement the adaptive Central Pattern Generator (CPG) strategy previously established in the literature as a representative case study. The results verify that the co-simulation environment successfully captures the intended closed-loop behavior, such as torque-dependent gait modulation and perturbation recovery. This confirms the platform's efficacy as a versatile tool for testing and refining human-exoskeleton interaction strategies before experimental trials

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Assistive soft tendon-driven exosuits have demonstrated effectiveness in supporting human locomotion while maintaining low weight and high comfort. However, they suffer from a key limitation: actuation is unidirectional, as tendons can pull but not push. This constraint reduces versatility, particularly for tasks requiring bidirectional torque at the hip. While bidirectional solutions have been proposed, existing designs often suffer from redundancy, high weight, off-board actuation, or limited portability. To address this gap, we present a lightweight tendon-driven hip exosuit capable of providing bidirectional assistance with a single motor per leg. The design employs a quasi-passive tendon-tensioning mechanism with a ratchet–pawl lock to eliminate cable slackness, coupled with a control algorithm that estimates gait phase to engage the clutch at precise moments. Experiments with three participants confirm that employing the clutch mechanism delivers 31 % more power compared to operation without a clutch, representing a promising advance toward narrowing the performance gap between soft exosuits and rigid exoskeletons.

Keywords: Robotic Rehabilitation and Assistance, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Wearable robotic exoskeletons are promising for enhancing gait, personal mobility, and exercise, particularly new lightweight designs that can expand their utility beyond the clinic to everyday community settings. As the accessibility of these devices increases, understanding their impact on gait mechanics and function is critical to expand their use cases across the continuum of care. This study evaluated the immediate functional, biomechanical, and physiological effects of WIM, an ultralightweight hip exoskeleton, in healthy young adults during aerobic walking. Fourteen participants completed three randomized 6-Minute Walk Tests under different device modes: standby, assist, and exercise (resist). Lower-limb kinematics were recorded using inertial motion sensors, and cardiopulmonary response was assessed via metabolic gas analysis. While modes did not significantly affect walking distance, participants reported higher exertion during the exercise mode. This mode induced significantly greater metabolic demand, including increased metabolic cost of transport and energy expenditure. Kinematic analysis revealed that the assist mode increased hip flexion and leg lift compared to the exercise mode, with compensatory changes in knee extension and flexion. These findings align with established roles of assistive exoskeletons in supporting limb motion and of resistance-based exercise in imposing neuromuscular demand. The portability and dual-mode function of WIM highlight its potential as a community-based device for older adults and individuals with gait and balance deficits.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance
Abstract: Lower-limb exoskeletons can effectively reduce the energetic cost of locomotion, yet identifying effective, user-specific assistance patterns remains challenging. Human-in-the-loop optimization can yield metabolic benefits but is time-consuming and often relies on indirect calorimetry, whereas many model-based and data-driven controllers shape joint energetics without providing compact, joint-level tunable parameterizations. This study introduces a stride-to-stride receding-horizon predictive controller that optimizes the timing of a hip torque profile for running assistance using hip kinematics and a mechanically interpretable objective. At each stride, the controller updates six timing parameters to maximize predicted positive work W^+ while penalizing negative work W^-, regularizing deviations from a nominal profile and abrupt stride-to-stride changes. Using treadmill running data from five participants wearing a hip exoskeleton in transparent mode (i.e., almost null output interaction torque), we conducted an offline evaluation of the controller, comparing its performance against a nominal profile and an offline subject-specific static optimum. Across subjects, timing adaptation increased W^+ relative to the nominal profile (+6.1% to +17.2% at the baseline gain), while reducing W^- and converging within 2–8 strides. The offline static optimum produced the highest W^+, with timing parameters similar to those reached by the adaptive controller. These results suggest that predictive timing optimization can rapidly identify hip-assistance timing and support future studies linking mechanical work objectives to physiological and performance outcomes.

Keywords: Micro/Nano Robotics in Medicine and Biology, AI and Learning in Biomedical Robotics
Abstract: This paper proposes a high-speed and robust visual feedback system that enables long-term observation of freely moving C. elegans body parts by integrating deep learning-based object detection (YOLO) with a high-speed tracking microscope. Ideally, for detailed analysis, the target should be projected to fill the Field of View as much as possible. However, applying deep learning introduces a constraint: the need to secure a context margin around the physical target, which effectively enlarges the target size required for detection. Maintaining this large effective target within the FOV leaves a minimal spatial margin for movement. In this study, we demonstrated that a high frame rate exceeding 268 fps is required merely to prevent the C. elegans from deviating from the FOV. Furthermore, by integrating high-speed inference with a low-latency tracking microscope driven by high-speed vision and an automated stage, we realized a tracking system capable of 1000-fps processing that satisfies this requirement while accounting for cumulative system latencies. In addition, false detections hinder practical application in high-speed, high-magnification environments. To address this, we ensured tracking robustness by implementing a SW Filter that rejects false positives based on spatiotemporal continuity between frames. By reconciling high speed with robustness, we report the successful realization of stable tracking under 5x and 20x magnification using the proposed system.

Keywords: AI and Learning in Biomedical Robotics
Abstract: Suturing is a fundamental surgical skill currently evaluated with methods that lack precision and reproducibility, and are time-consuming, resource-intensive and expensive. This work delivers a proof-of-concept for a novel, objective and cost-effective method to evaluate surgical suturing skills. The Surgical sUturing GrAdeR (SUGAR) tool applies YOLO object detection algorithms to segment surgical needles, the surgeon’s hands and suturing tools, assessing thirteen hard-coded suture quality metrics through Computer Vision (CV) techniques. This work produces a new CV-based tool capable of evaluating key suture quality parameters with a precision of 85.68% and a recall factor of 61.75%, exhibiting a promising potential as a medical educational tool.

Keywords: AI and Learning in Biomedical Robotics, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: This paper presents a new method for the rapid deployment and personalization of human activity recognition (HAR) models for lower-limb exoskeletons. A generalized long short-term memory (LSTM) model is first pretrained using an open-source dataset by selecting sensor modalities and feature channels that match the target HAR system. The pretrained model is then rapidly deployed and personalized through transfer learning (TL) using a small amount of data collected from new users wearing a custom hip–knee exoskeleton. The proposed TL approach is validated both internally on the open-source dataset and externally using a cross-dataset evaluation with data collected from the custom exoskeleton. Under cross-dataset TL, a generic model pretrained on assisted-condition data achieves 95% classification accuracy using only 30% of the new user’s data (approximately 6 minutes), outperforming a conventional subject-specific models by about 7%. When pretrained on both assisted and unassisted data, 95% accuracy is achieved with only 25% of the new user’s data. Pretraining exclusively on unassisted data has superior performance, achieving 99% accuracy with 35% of the new user’s data (approximately 7 minutes). Finally, the TL process is deployed on an NVIDIA Jetson Xavier embedded platform, where TL completes in approximately 30 s using 6 min of newly collected data. The resulting personalized model achieves a mean inference time of 3 ms, demonstrating its suitability for real-time HAR and exoskeleton control.

Keywords: AI and Learning in Biomedical Robotics, Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance
Abstract: Dropped head syndrome, caused by neck muscle weakness, leads to poor posture, neck pain, and a reduced ability to complete daily tasks and engage in social interactions. Current treatments using static neck braces only stabilizes the head in an upright position but cannot restore the mobility of the head-neck for many important social and daily living tasks. Powered neck exoskeletons were developed to restore head-neck mobility in individuals with dropped head syndrome. Natural coordination between the head and eye movements is a promising method to give the user control of the neck exoskeleton. The key to this control method is accurate sensing of the gaze within the head coordinate frame. Currently, commercial eye trackers are used. However, these trackers are expensive, complex, and impractical to integrate with the existing robotic system, which creates significant barriers for the translation of the neck exoskeleton technology to patient populations. In this paper, we propose alternative CNN-based gaze estimation models using a commercial tracker as the ground truth. Our results demonstrate that the proposed models achieve performance comparable to that of commercial black-box gaze estimation models and perceptually indistinguishable to healthy participants when deployed on a physical neck exoskeleton. These findings demonstrate the feasibility of developing gaze-sensing modules tailored specifically to the powered neck exoskeletons, paving the way for affordable, simple, and reliable systems to improve the quality of life of those affected by dropped head syndrome.

Keywords: AI and Learning in Biomedical Robotics, Robotic Rehabilitation and Assistance, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Surface electromyography (sEMG) is a promising control signal for assist-as-needed hand rehabilitation after stroke, but detecting intent from paretic muscles often requires lengthy, subject-specific calibration and remains brittle to variability. We propose a healthy-to-stroke adaptation pipeline that initializes an intent detector from a model pretrained on large-scale able-bodied sEMG, then fine-tunes it for each stroke participant using only a small amount of subject-specific data. Using a newly collected dataset from three individuals with chronic stroke, we compare adaptation strategies (head-only tuning, parameter-efficient LoRA adapters, and full end-to-end fine-tuning) and evaluate on held-out test sets that include realistic distribution shifts such as within-session drift, posture changes, and armband repositioning. Across conditions, healthy-pretrained adaptation consistently improves stroke intent detection relative to both zero-shot transfer and stroke-only training under the same data budget; the best adaptation methods improve average transition accuracy from 0.42 to 0.61 and raw accuracy from 0.69 to 0.78. These results suggest that transferring a reusable healthy-domain EMG representation can reduce calibration burden while improving robustness for real-time post-stroke intent detection. Our project website, video, code, and dataset are available at: https://roamlab.github.io/reactemg-stroke/.

Keywords: Robotic Rehabilitation and Assistance, AI and Learning in Biomedical Robotics, Neuro-Robotics, Neural Interfaces, and Human-Centered Design
Abstract: Leader–follower interaction is an important paradigm in human–robot interaction (HRI). Yet, assigning roles in real-time remains challenging for resource-constrained mobile and assistive robots. While large language models (LLMs) have shown promise for natural communication, their size and latency limit on-device deployment. Small language models (SLMs) offer a potential alternative, but their effectiveness for role classification in HRI has not been systematically evaluated. In this paper, we present a benchmark of SLMs for leader–follower communication, introducing a novel dataset derived from a published database and augmented with synthetic samples to capture interaction-specific dynamics. We investigate two adaptation strategies: prompt engineering and fine-tuning, studied under zero-shot and one-shot interaction modes, compared with an untrained baseline. Experiments with Qwen2.5-0.5B reveal that zero-shot fine-tuning achieves robust classification performance (86.66% accuracy) while maintaining low latency (22.2 ms per sample), significantly outperforming baseline and prompt-engineered approaches. However, results also indicate a performance degradation in one-shot modes, where increased context length challenges the model's architectural capacity. These findings demonstrate that fine-tuned SLMs provide an effective solution for direct role assignment, while highlighting critical trade-offs between dialogue complexity and classification reliability on the edge.

Keywords: Soft Robotics
Abstract: In vitro experiments are essential for studying cellular and tissue behavior, but mechanical cues strongly influence cellular responses. Integrating controlled static and dynamic stimulation is therefore critical for mechanobiology. Existing cell stretchers often lack usability, scalability, and compatibility with standard lab workflows.

We present a soft robotic 3D cell stretcher tailored for cardiac mechanobiology. Redesigned through an interdisciplinary co-design with biologists, the system emphasizes robustness, usability, and seamless integration with standard culture plates, incubators, and imaging pipelines. It delivers volumetric, cyclic stimulation compatible with physiological cardiac dynamics.

The platform improves reliability, simplifies operation, and enhances stimulation consistency, providing a practical solution for long-term in vitro mechanobiology studies.

Keywords: Soft Robotics
Abstract: Abstract— Additive manufacturing (AM) has accelerated soft robotics research, but filament-based Fused Filament Fabrication (FFF) remains limited by material stiffness, restricting the achievable softness of printed components. This work demonstrates that Fused Granular Fabrication (FGF)—a pellet-based extrusion process—can produce airtight and mechanically functional soft pneumatic systems from ultra-soft thermoplastic elastomers. Using pelletized Styrene-Ethylene-Butylene-Styrene (SEBS, Shore 47A), we optimized process parameters and toolpath design to eliminate inter-bead voids and minimize deformation during printing.

Two pneumatic components were fabricated entirely via FGF: a normally open kinked-channel control valve and a normally closed self-sealing duckbill valve. Both operated at low pressures (21–69 kPa) and were integrated into a self-regulating pneumatic gripper that achieved autonomous grasp–hold–release behavior without electronic control. These results establish FGF as a viable approach for printing airtight, logic-capable soft robots, expanding the accessible material palette for additive manufacturing and enabling sustainable, recyclable, and monolithic soft robotic systems.

Keywords: Haptics and Human-Machine Interaction, Soft Robotics, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: In virtual reality and teleoperation scenarios, tactile feedback to an operator can enhance presence and task performance. While part of the challenge of creating this technology lies in designing tactile feedback devices and control schemes, another challenge is the design of a grounding solution to convey the tactile stimuli to the body, especially for wearable systems. Current solutions, such as gloves and thimbles, are often uncomfortable and interfere with vision-based hand tracking methods. In this paper, we explore if bio-inspired suction cups can provide a solution for tactile display attachment. We use an existing sea-urchin-inspired suction cup, which we first optimize in a controlled environment and then test in a use case. In Experiment 1, we show that for individual suction cups, wall angle and thickness both play a significant role in the maximum suction force that a cup can produce. For our cup type, a 5 mm thick wall and a 60.3 degree wall angle produced the largest suction force. In Experiment 2, we show that the number of cups has an effect on the total suction force that a display can produce. There seems to be an optimum, which was 5 cups for our display geometry. Finally, we validate in a user study that the best-performing display from Experiment 2 successfully remains attached to the palm of the hand. For all 10 participants, our attachment mechanism was not only able support its own weight, but also to withstand the forces produced by a tactile actuator, both in static pressure and vibration mode. This shows the potential of suction cups as an alternative attachment mechanism for tactile displays.

Keywords: Grippers and Other End-Effectors, Mechanism Design, Agricultural Automation
Abstract: Conventional fingered grippers often struggle in confined spaces because limited lateral access prevents finger insertion and inward closing. This paper presents the DBCF-EM gripper, whose fingers combine a base-driven curved flexure, prescribing a tangential object-following trajectory, with a dual-belt eversion system that creates near-stationary contact surfaces and reduces sliding at the contact interfaces. This enables a low-disturbance caging grasp strategy in which the fingers propagate along the object surface rather than closing perpendicularly toward it. A prototype gripper was built for robotic tomato-removal experiments from a crate. Experiments showed contour following with a maximum deviation of 3~mm, negligible normal disturbance of at most 0.1~N, and a 91-97% reduction in tangential disturbance forces. In robotic trials, the gripper achieved 100% pick-up success and a 91% damage-free success rate, demonstrating its effectiveness for confined-space grasping.

Keywords: Soft Robotics, Robotic Rehabilitation and Assistance
Abstract: Home-based rehabilitation is increasingly recognised as a key strategy for providing continuous, accessible and cost-effective therapies while reducing dependence on specialised equipment and clinical supervision. This is particularly relevant for musculoskeletal and post-surgical rehabilitation. Extending functional therapies (e.g., mechanotherapy and heat therapy) to wearable platforms requires compact, lightweight devices that preserve therapeutic efficacy outside of clinical settings. This work presents the concept and preliminary investigation of a fully textile, programmable compression module that integrates localised heating into a lightweight, compact architecture. The system employs thermally activated twisted and coiled polymer actuators (TATCPAs) for controlled compression and embroidered hybrid resistive elements for efficient heating, synergistically combining mechanotherapy and heat therapy in a single wearable device for the first time. This proof-of-concept consists of a single compact drive module, intended as the basic building block to devise a fully functional wearable system. This approach provides a scalable basis for extending rehabilitation beyond clinical settings, facilitating accessible home-based therapy.

Keywords: Soft Robotics, Biologically-Inspired Robotics and Biomimetics, Robotic Rehabilitation and Assistance
Abstract: This work presents a hollow cylindrical tensegrity manipulator based on an auxetic 2D tensegrity lattice inspired by a deep-sea sponge. The key feature is the prestressed structure and controlled local buckling induced through the topology, which also allows defined bending and results in a substantially larger angular change per segment and therefore a larger and more complex workspace. The combination of auxetic properties and tunable buckling introduces new possibilities for compact yet highly deformable continuum manipulators, offering promising applications in soft robotics, biomedical engineering and human-robot interaction.

Keywords: AI and Learning in Biomedical Robotics, Robotic Surgery and Diagnosis
Abstract: Artificial intelligence (AI) is increasingly adopted in surgery for performance assessment and intraoperative decision support. A key capability in this context is the automatic recognition of tool–tissue interactions (TTIs) from surgical videos. Existing approaches for modelling surgical actions, mainly represented by triplets, do not explicitly model contact. In this work, we propose a formulation of the TTI understanding problem based on contact-aware interaction modelling, which introduces two sequential objectives: (i) spacial detection of physical contact between tool and tissue, and (ii) classification of the interaction type. By explicitly incorporating depth information, our approach moves beyond purely 2D semantic reasoning and enables inference of actual physical contact rather than visual co-occurrence. This formulation enables the use in robotic and safety-critical applications for which semantic understanding only is not sufficient. The proposed architecture integrates YOLO-based instance segmentation, depth estimation, and a Vision Transformer for interaction reasoning. Experimental results show that the proposed method achieves 87.61% end-to-end accuracy on the TTI understanding task, with F1 score of 88.03%, demonstrating the feasibility of reliable, contact-aware TTI inference.

Keywords: Robotic Surgery and Diagnosis
Abstract: Upper aerodigestive tract (UADT) treatments frequently employ transoral laser microsurgery (TLM) for procedures such as the removal of tumors or polyps. In TLM, a laser beam is used to cut target tissue, while forceps are employed to grasp, manipulate, and stabilize tissue within the UADT. Although TLM systems may rely on different technologies and interfaces, forceps manipulation is still predominantly performed manually, introducing limitations in ergonomics, precision, and controllability. This paper proposes a novel robotic system for tissue manipulation in UADT procedures, based on a novel end-effector designed for forceps control. The system is integrated within a teleoperation framework that employs a robotic manipulator with a programmed remote center of motion (RCM), enabling precise and constrained instrument motion while improving surgeon ergonomics. The proposed approach is validated through two experimental studies and a dedicated usability evaluation, demonstrating its effectiveness and suitability for UADT surgical applications.

Keywords: Robotic Surgery and Diagnosis
Abstract: Oral and maxillofacial surgeries, including mandibular fracture reduction, bilateral sagittal split osteotomy, and dental implant placement, require precise bone cutting and screw fixation. However, due to the limitation of manual operation and the absence of clear anatomical landmarks, surgeons may accidentally injure the nerves and surrounding soft tissues. The navigation and robotic technology has been introduced into oral and maxillofacial surgery and enhance the precision of bone preparation and implant placement. Bone registration is a critical step that not only impacts surgical accuracy, also affect the usability, operative time, and workflow efficiency. This study introduces three clinically applicable intraoperative mandibular registration methods: Pair-Point Registration based on screws (PP-screw), Surface Points–Based Registration (SP), and Computed Tomography (CT) Fiducial Registration (CTF), and compares their accuracy in a phantom study. The results show that CTF achieved a target registration error (TRE) of 0.58 mm, followed by SP (0.75 mm) and PP-screw (0.79 mm). Although CTF yielded the lowest TRE, SP achieved accuracy comparable to PP-screw without requiring invasive screw placement or additional intraoperative radiation exposure. Considering the trade-off between accuracy and clinical practicality, SP represents the most balanced and efficient approach for routine implementation in mandibular navigation and robot-assisted surgery.

Keywords: Robotic Surgery and Diagnosis, AI and Learning in Biomedical Robotics
Abstract: Accurate CT-to-ultrasound registration is essential for image-guided minimally invasive spine surgery, yet existing methods suffer from poor robustness to initial misalignment and incomplete data. This work presents a fully automatic and level-wise landmark-based registration method using PCA-aligned geometric clustering to extract anatomical landmarks from CT and ultrasound point clouds. The anatomical landmarks are matched between modalities, enabling both full-spine and per-vertebra alignment and serves as an initial guess for CPD. Evaluated on a lumbar phantom across 20 initial CT-to-US transformations (rotations up to 90 degrees, translations up to 100 mm) and 20 ultrasound perturbations, the proposed landmark-CPD method achieved 0.60 mm accuracy with 100% success rate (full vertebrae) and 0.88 mm with 98% success (per vertebra) outperforming ICP (15% success) and CPD alone (14.44 mm error, 46% success per vertebra, 5.5x slower). The proposed method offers clinically relevant accuracy with superior robustness, speed, and independence from initialisation, making it well-suited for intraoperative workflows.

Keywords: Micro/Nano Robotics in Medicine and Biology, Soft Robotics, Robotic Surgery and Diagnosis
Abstract: Wireless magnetic robots are being developed for minimally invasive surgical procedures. Recently, flexible, millimeter-scale magnetic filaments were introduced to navigate the human urinary tract as a potential treatment for urological diseases. Such robots are actuated by rotating magnetic fields and can be tracked and visualized by ultrasound. A key challenge in translating this robotic technology to clinical practice is achieving accurate real-time tracking of the robot’s motion. Adhesion to surrounding surfaces or stiction to tissues is a common issue observed in ex vivo experiments. In this article, we are demonstrating the ability to track the robot’s rotational motion to ensure its proper actuation and function. The detection of the robot’s rotation is achieved by intensity-tracing in ultrasound M mode.

Keywords: Micro/Nano Robotics in Medicine and Biology
Abstract: The development of wireless, drug-loaded, dissolvable, untethered robots that can navigate arteries, deliver thrombolytic agents locally, and subsequently quickly degrade represents a significant departure from current systemic treatments. Achieving a quickly dissolving robot requires addressing key challenges, i.e., as the robot decreases in size during navigation from the point of insertion to the target site, it must retain sufficient propulsive thrust for controlled locomotion, while simultaneously providing adequate contrast for detection under X-ray fluoroscopy or other clinical imaging modalities. Using a thermally moulded sucrose–alginate composite embedded with superparamagnetic iron oxide nanoparticles, we fabricate robots with screw-shaped geometry, enabling magnetic actuation and X-ray visibility. These untethered robots are actuated against physiological flow of up to 32 mL min−1 while retaining X-ray contrast-to-noise ratio of 1.5 at a radiation dose of 0.17 µGy m2. The navigation capability of the robots is validated in a three-dimensional cerebral vascular phantom, where controlled motion is achieved along the common carotid–internal carotid–middle cerebral artery pathway.

Keywords: Bionics and Prosthetics, Robotic Rehabilitation and Assistance
Abstract: Walking speed adaptability is a fundamental requirement for community ambulation but remains severely restricted in individuals with transfemoral amputation. While functional classification systems distinguish between limited (K2) and variable-cadence (K3) ambulators, the specific biomechanical mechanisms that dictate this disparity remain poorly characterized for prosthetic design purposes. This study quantified the speed-dependent kinetic and kinematic adaptations of K2 and K3 transfemoral amputees compared to able-bodied controls. Results revealed that K3 users successfully modulated gait dynamics to maintain symmetry across speeds, primarily by generating excessive hip flexion moments at late stance to compensate for the lack of prosthetic ankle propulsion. In contrast, K2 users failed to modulate joint kinetics, exhibiting speed-dependent increases in asymmetry. Furthermore, K2 users displayed a ”stiff-legged” swing phase with reduced knee flexion. These findings suggest that active devices for K3 users must prioritize speed adaptive power injection to mitigate hip compensation, whereas controllers for K2 users must prioritize stability assurance and active swing clearance to encourage limb loading and safety. Bridging the gap between K2 and K3 mobility requires robotic interventions that not only restore physiological push-off but specifically target the confidence and stability deficits inherent to lower-mobility gait.

Keywords: Bionics and Prosthetics, Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance
Abstract: Lower limb amputation affects millions worldwide, leading to impaired mobility, reduced walking speed, and reduced participation in daily activities. Powered prosthetic knees can partially restore mobility by assisting knee joint torque, improving gait symmetry, sit-to-stand transitions, and walking speed. However, the added mass of powered components may offset these benefits by negatively affecting gait mechanics and increasing metabolic cost. Reducing powered prosthesis mass is desirable, but doing so typically requires smaller motors and batteries, which can limit performance. Consequently, optimizing mass distribution, rather than minimizing total mass, may be a more effective solution. In this exploratory study, we evaluated above-knee powertrain placement in a small cohort. Compared to below-knee placement, the above-knee configuration improved walking speed (+9.2% in one participant) and cadence (+3.6%), with mixed effects on gait symmetry. Kinematic measures were similar across configurations, and additional testing on ramps and stairs confirmed robustness across tasks. These preliminary results suggest that above-knee placement is feasible and that optimizing mass distribution can preserve the benefits of powered assistance while mitigating added weight. Further studies are needed to confirm these findings.

Keywords: Bionics and Prosthetics, Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance
Abstract: Most powered knee prostheses regulate joint impedance using gait-phase estimation, mode switching, or gait speed estimation, which limit adaptability. This work introduces an indirect volitional thigh-driven variable impedance controller (IVVIC) that directly maps sagittal-plane thigh kinematics to knee impedance without any gait-phase or locomotion condition estimation. Stiffness and damping parameters are represented as bivariate Bernstein polynomial surfaces of normalized thigh roll angle and roll rate, whose coefficients are identified offline through a constrained optimization using able-bodied gait data at different walking speeds. Leave-one-subject-out validation demonstrates that the learned impedance surfaces generalize across users, reconstructing physiological knee torque with a median NRMSE of 14.7% across subjects and 20.1% across speeds. When the controller is deployed on a powered knee prosthesis and tested with a healthy participant, it produces stable, repeatable gait pattern across different walking speeds. The prosthesis replicates key biomechanical profiles, including speed-dependent increases in knee flexion angle and flexion torque, while adapting continuously to changes in walking speed without any explicit task or speed estimation. These findings suggest that IVVIC offers a sensor-minimal, intuitive, and speed-agnostic alternative to conventional phase-based prosthesis control, with strong potential for future extension to multi-task locomotion.

Keywords: Bionics and Prosthetics, Exoskeletons, Exosuits, and Wearable Assistive Devices, Haptics and Human-Machine Interaction
Abstract: Virtual environments are increasingly used to evaluate prosthetic control strategies. However, the measurement properties of virtual functional tasks remain insufficiently characterized. This study assessed the concurrent validity and test–retest reliability of a physics-based virtual implementation of the Box and Blocks Test (BBT) designed for proportional myoelectric prosthesis control. Ten able-bodied participants completed physical and virtual BBTs using a prosthetic emulator across two sessions. Performance was quantified as the number of blocks transferred per trial. Concurrent validity between virtual and physical performance was evaluated using Pearson’s correlation coefficients, and test–retest reliability of the virtual task was assessed using intraclass correlation coefficients (ICC). There was a strong correlation between virtual BBT and physical BBT performance (r=0.754). Test–retest reliability of the virtual BBT across sessions was moderate (ICC(2,1)=0.54), which was similar to the physical BBT (ICC(2,1)=0.60). These results support the use of this virtual BBT for systematic evaluation of prosthetic control performance.

Keywords: Bionics and Prosthetics, Robotic Rehabilitation and Assistance
Abstract: Muscle Fatigue (MF) is a significant challenge in upper-limb prosthesis, driven largely by device weight and prolonged wear; however, the full extent of fatigue-related consequences remains insufficiently explored. To the best of the authors' knowledge, this study is the first to investigate MF during simulated Activities of Daily Life (ADLs) in able-bodied participants using an upper-limb prosthesis. Muscle activity, joint biomechanics, movement velocity, and Rated Perceived Exertion (RPE) were assessed to evaluate the impact of prosthesis use. Ten healthy participants with intact limbs performed five ADLs: drinking from a cup, using a fork, lifting a box, reaching overhead, and hammering, both with and without the prosthesis. Results showed a significant increase in muscle activation when using the prosthesis, particularly in the trapezius (195 %) and deltoid (300 %) muscles, accompanied by increased movement intensity at the shoulder and elbow joints and reduced wrist mobility. These findings indicate fatigue-driven compensatory strategies characterized by proximal overuse and distal restriction. Movement velocity decreased in most activities, with the largest reductions observed during reaching (27 %) and lifting (22 %). RPE values were significantly higher in the prosthesis condition, especially during the fork and reach tasks. Additionally, a reduction in the number of task repetitions was observed when using the prosthesis, particularly during the fork task. Notably, high fatigue levels negatively affected the performance of the sEMG-controlled prosthesis. These results highlight the need for strategies to mitigate MF and improve prosthesis usability during ADLs.

Keywords: Bionics and Prosthetics, Robotic Rehabilitation and Assistance
Abstract: Current myoelectric hand prostheses offer limited control over wrist orientation, reducing users’ ability to reach objects and resulting in exaggerated compensatory movements. In this study, we evaluated an adaptive prosthetic wrist with switchable compliance capabilities, automatically alternating between compliant behavior when reaching to an object and stiff behavior while holding the object. A participant with a transradial amputation performed tasks of daily living using the proposed prosthetic wrist in the adaptive mode (AW) and in always stiff (SW), and always compliant (CW) modes. To evaluate compensatory movements induced by the different modes, we tracked the arm and trunk movements and compared them to a control group of able-bodied individuals. Results suggested that both CW and AW reduced compensatory trunk and arm movements when compared to SW during reaching. Moreover, the benefits of AW proved greater than or comparable to those of CW in five out of seven tasks when considering arm movements during manipulation. Finally, subjective measures suggested that the user may require more training to become accustomed to the AW mode. These findings support the potential of selectively compliant wrists to improve prosthesis use and reduce compensatory movements, and provide important guidelines for wrist design

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, AI and Learning in Biomedical Robotics, Haptics and Human-Machine Interaction
Abstract: Work-related musculoskeletal disorders (WMSDs) remain a major challenge in physically demanding occupations, motivating the development of active wearable exoskeletons to reduce physical strain during manual handling tasks. A key requirement for such systems is the ability to anticipate user motion intention and deliver timely, task-appropriate assistance in real-world settings. This paper presents an intention-driven control framework for a cable-driven upper-limb exoskeleton based on multimodal vision–inertial sensing. Visual information from an egocentric camera and kinematic data from wrist-mounted inertial measurement units are fused to recognise three fundamental manual handling actions: lifting, walking, and placing. Multimodal data were collected from 20 participants to train and evaluate the intention prediction model offline. The framework was further validated through real-time experiments with three participants on a physical exoskeleton, achieving an overall classification accuracy of 97%. Predicted actions preceded measurable muscle activation by an average of 1.62 s, enabling anticipatory assistance rather than reactive support. Electromyography-based evaluation further showed reductions of 63.5% and 42.9% in biceps and triceps muscle activation, respectively, during box-carrying tasks. These results show that multimodal intention prediction can enable timely and physically effective assistance in wearable robotic systems.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Neuro-Robotics, Neural Interfaces, and Human-Centered Design, Biologically-Inspired Robotics and Biomimetics
Abstract: Myokinetic interfaces offer a promising approach to translating user’s motor intent into control of assistive upper-limb exoskeletons via magnetic tracking of individual muscle activity. Current systems rely on localization units fixed within a single reference frame, which are suitable for prosthetic limbs as they can be embedded inside a rigid prosthetic socket. This solution is not feasible for use in soft-tissue applications like exoskeleton cuffs that do not have hard anchoring. To address this, localization systems using Independent Localization Units (ILUs), whereby each unit tracks a single pair of target magnets in its own local frame, can be employed, thus eliminating rigid inter-unit constraints and improving anatomical adaptability. However, independent localizers are sensitive to magnetic fields from nearby disturbing magnets, which can interfere with the localization of the target pair. The operational rules for minimizing the localization cross-talk, that were derived from previous studies under ideal conditions, would not be effective in clinical scenarios due to anatomical limitations. This study assessed such rules, through simulations, in a representative clinical application, i.e., controlling an assistive elbow exoskeleton with magnets implanted in the biceps/triceps. The positive results demonstrated that ILUs can be effectively positioned on the arm to provide reliable tracking of magnets, and supported the identification of optimal anatomical regions for magnet implantation.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Haptics and Human-Machine Interaction
Abstract: Human upper-limb movements are known to exhibit robust kinematic features, including bell-shaped velocity profiles and the two-thirds power law. These features are interpreted as signatures of motor primitives underlying human movement planning. While wearable exoskeletons are increasingly used to reduce physical load, their influence on these motor primitives remains unexplored. This study investigates whether upper-limb kinematic motor primitives are maintained during movements performed with a shoulder-support exoskeleton. Five healthy participants performed vertical and complex three-dimensional upper-limb movements under three conditions: no exoskeleton, low-support exoskeleton, and high-support exoskeleton. Movement elements were extracted based on velocity zero-crossings of hand kinematics. The experimental velocity profiles of these elements were compared to a bell-shaped profile, and the scaling relationship between mean velocity and movement displacement was evaluated. Results showed strong agreement with bell-shaped velocity profiles, with high correlations to the theoretical profile (0.86-0.87), and scaling exponents close to the theoretical two-thirds value (confidence intervals range of [0.57, 0.73]). Statistical analysis revealed no significant effects of exoskeleton assistance level or movement type. These preliminary findings suggest that shoulder-support exoskeleton assistance does not disrupt the fundamental kinematic structure of upper-limb motor primitives, maintaining predictable kinematic motor primitives despite load redistribution. This supports the use of model-based assistance, intent inference, and primitive-based control strategies in wearable robotics, assuming that fundamental movement structure remains intact under exoskeleton support.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Upper limb impairment and spastic hypertonia following a stroke often necessitates constant wrist stretching for symptom relief and functional recovery. Most existing rehabilitative devices, however, can be difficult for stroke survivors to don and operate with one functional hand, which creates a barrier to at-home, self-administered therapy. To address this, we present SUREHand, a novel, passive wrist exoskeleton specifically designed for hemiparetic stroke survivors to independently don and use with one hand. The device features a pneumatic support system with an inflatable air bladder to ensure a secure, comfortable fit and a body-powered ratchet mechanism that allows users to manually induce and maintain wrist extension. We recruited three stroke survivors (N=3) to evaluate the device's mechanical performance and device usability. All participants operated the device independently in under 20 minute of training, with average donning times of 72 seconds and wrist extension with the device achieving their maximum passive range of motion. High scores on the System Usability Scale and comfort survey suggest that the SUREHand is an effective, low-cost, and accessible solution for wrist rehabilitation and stretching.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance, Bionics and Prosthetics
Abstract: Assistive exoskeletons impose stringent requirements on joint actuation in terms of compactness, weight, backdrivability, and compliance, to enable safe and efficient human-robot interaction. To achieve these features, this work presents the design and experimental characterization of a lightweight, fully 3D-printed series elastic gearbox for wearable exoskeletons. The proposed transmission combines a novel flat nested compact-cam cycloidal drive with a customized torsion elastic element, both realized through additive manufacturing. Moreover, 3D-printed bearings are employed to reduce the transmission's weight and encumbrance. An experimental characterization is first performed to assess the reducer's performance in terms of backdriveability, friction, backlash, and stiffness. To achieve a target output stiffness suitable for an assistive exoskeleton, a custom compliant element is then designed and integrated in series with the reducer. The resulting gearbox presents valuable performances, a favorable form factor, low mass, and a linear stiffness, making it well-suited for assistive exoskeletons. The combination of the novel cycloidal variant and the custom torsional spring highlights the potential of additive manufacturing to realize lightweight and compact compliant transmissions with mechanical properties suitable for wearable robotics.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Soft Robotics, Robotic Rehabilitation and Assistance
Abstract: Soft robotic exogloves can provide hand rehabilitation and assistance. Fitting these gloves often relies on standardized measurements not tailored to the individual, limiting their effectiveness, especially for fine articulation necessary for dexterous manipulation. We present the design, fabrication, modeling, and testing of a personalized pneumatically-actuated soft robotic exoglove. The glove was fit to a user's hand with topological scans and fabricated with silicone mold casting. Finite element analysis (FEA) was performed to evaluate actuator bending and forces from physical human-robot interaction (pHRI) between an actuator and a simplified personalized biomechanical finger model. Pneumatic pressure control experiments were conducted to flex the user's finger with static and dynamic references. Fabrication results show that topological scans enable precise tailoring to hand anatomy. Simulations showed that anatomical personalization enables analysis of pHRI contact forces, and results indicate sufficient joint mobilization with non-ideal compression on the proximal phalanx. Pneumatic testing indicates that pressure control allows accurate and targeted mobility of the metacarpophalangeal (MCP) and proximal interphalangeal (PIP) joints with intrinsic stiffness. Testing of multiple designs showed that relaxing the strain-limiting layer improves actuator-to-finger joint alignment during actuation. This work presents personalization to the human hand in structural conformability, joint topology, modeling of pHRI contact, and time-dependent actuation-deformation profiles. This lays a groundwork for informing exoglove design optimization to enable assistance in dexterous manipulation and neuromuscular rehabilitation of fine motor skills.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance, Soft Robotics
Abstract: Temporomandibular disorders substantially affect and impair jaw functions and thus the quality of life. Yet, powered assistance for jaw rehabilitation remains widely unexplored despite advances in exoskeletons for other extremities. To the best of our knowledge, this work presents the first comprehensive computational framework for hybrid rigid-soft jaw exoskeletons, integrating a biomechanically validated 24-muscle jaw model with deformable finite element soft-body dynamics in MuJoCo. The proposed four-tendon system combines a rigid chin cup for force transmission with a compliant facial mask for user comfort. Simulation-based evaluation under six configurations (TPU/Silicone materials, muscles-only/tendons-only/combined actuation) demonstrates effective trajectory tracking with mean errors of 4.60-6.42 mm with safe interface pressures and material stresses, while revealing performance–comfort trade-offs: TPU shows lower mean strain (e.g., 11.49 % vs 14.06 %), but higher mean pressure (e.g., 12.02 kPa vs 9.89 kPa) than silicone. Mean tendon forces of 24.63-35.46 N exceed passive muscle opening forces, yet combined actuation paradoxically increases biological muscle loading by 6-10 % despite improved tracking, guiding future control strategies toward biomechanical- and effort-aware algorithms. This validated open-source computational framework establishes essential safety constraints and design prerequisites for physical prototyping, providing a systematic path toward safe, wearable robot-assisted therapy for temporomandibular disorders.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance
Abstract: Neck muscle weakness is a prominent marker for tracking the progression of amyotrophic lateral sclerosis, which is a fatal neurodegenerative disease. To quantify neck muscle weakness, scoring systems and dynamometry are used in clinical settings. However, these methods require experienced personnel and often lack consistency and quantification. There is an unmet need to continuously monitor neck muscle weakness, especially in remote, at-home settings. Previously, we have developed a powered neck exoskeleton (the Utah Neck Exoskeleton) with actuators and sensors to enable head-neck movements for patients with neck muscle weakness. In this paper, we investigate the use of this device to assess neck muscle strength in comparison to measurements from a clinic-standard hand-held dynamometer through a preliminary experiment in six healthy adults and two patients with ALS. A statistical model was developed to associate measurements from the exoskeleton (in Nm) and the dynamometer (in N) while simultaneously considering the health status of the participants (healthy vs. patients with ALS). Our results show a strong association between measurements from the exoskeletons and the dynamometer. This association did not significantly differ between the healthy group and patients with ALS. Results from this preliminary study provide the feasibility of using the powered neck exoskeleton to measure neck strength, with strong translation potential.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Soft Robotics, Robotic Rehabilitation and Assistance
Abstract: Wearable robotics have long been proposed as a solution to augment or restore hand function, which is critical for daily autonomy. Soft robotic gloves represent a promising approach, as they can operate in parallel with human muscles while ensuring comfort and compliance. Although cable-driven designs are among the most widely adopted soft actuation solutions, they are often hindered by inefficiencies brought by friction, backlash, and the requirement for continuous motor engagement to maintain static postures, resulting in increased power consumption and reduced battery life. This work presents a hybrid actuation paradigm, which integrates nonlinear multistable elastic shells with a cable-driven transmission to empower hand exosuits. By leveraging the stable equilibrium states of the shells, the cable-driven system is used primarily to trigger snap-through transitions between straight and curled configurations, thus functionally decoupling static posture maintenance from active motor control. Bench tests on a mock-up finger demonstrate that triggering transitions between stable equilibrium states enables posture holding with near-zero motor power consumption and fast finger closure time of 0.53±0.13 s across repeated snap-through events, with motion primarily governed by shell dynamics. These findings suggest that multistable elastic elements can enhance both responsiveness and efficiency of hand exosuits, potentially contributing to the design of more and more portable and energy-efficient assistive technologies.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Soft Robotics
Abstract: The effectiveness of a soft wearable back-support device in enhancing postural stability was investigated under trip-like perturbations using two experimental paradigms: perturbed standing and perturbed walking. Healthy subjects completed trials under three different back-support conditions: no device, device worn with low stiffness, and device activated with high stiffness. Whole-body stability was quantified using the minimum Margin of Stability (MOS) at the point of maximal instability. Results demonstrated increased MOS during device use, indicating enhanced postural stability. In standing, MOS increased significantly with device stiffness, whereas in walking, both device conditions improved MOS relative to no device but did not differ significantly from each other. These findings highlight the potential of soft wearable back-support devices with adjustable stiffness to improve reactive balance control against external perturbations, with important implications for fall prevention. Future research should explore personalized stiffness optimization and evaluate efficacy in populations at elevated risk of falls.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Construction workers or packers and movers need to carry heavy loads on various topographies such as horizontal surfaces, stairs, slopes, etc. Grabbing and stabilizing heavy loads requires a firm grip which puts a toll on their hands. The Auxivo CarrySuit (Auxivo AG, Switzerland) is specifically designed to redistribute the load from the arms to the shoulders and back, which releases the pressure on the hands. This paper explores the effect of the Auxivo CarrySuit on participants while carrying loads up and down a staircase. Specifically, the electromyograms of four pairs of muscles were measured to understand the effort associated with the back (erector spinae iliocostalis and erector spinae longissimus) and legs (quadriceps rectus femoris and gastrocnemius lateralis). Additionally, the Body Part Discomfort Score and Ratings of Perceived Exertion were obtained after every task to compare the effort needed and the discomfort and exertion felt between different tasks. Significant differences were found for the erector spinae iliocostalis and quadriceps rectus femoris muscles. The redistribution of the load to the lower back caused the strain on the erector spinae iliocostalis to increase, while the strain on quadriceps rectus femoris decreased. Along with this, the Body Part Discomfort Score scores associated with upper arms, lower arms, mid-back and thighs along with Ratings of Perceived Exertion indicated a significant decrease. The reduction in the perceived exertion and the discomfort level on mid-back, inspite of redistributing the load on back, indicate that the CarrySuit is beneficial to carry loads on staircase.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance, Biologically-Inspired Robotics and Biomimetics
Abstract: Exoskeletons and exosuits are a growing area of research interest. By augmenting human power, they hold great potential to improve the lives of those with mobility challenges, such as stroke survivors and paraplegic patients. Older adults are a growing population who could also benefit from wearable device mobility assistance, but for whom comfort, appearance, and cost are important factors in technology adoption. In this paper, we introduce a proof-of-concept passive mechanism for energy storage and return that aims to address these user concerns by following more advantageous and comfortable force profiles than in other passive solutions. This is achieved through a cam-and-cable-based mechanism that optimizes multiple objectives to maximize assistive force during energy return, minimize force while energy is being stored, and minimize overall device dimensions, resulting in a design that is smaller and more lightweight than actuators used in powered assistive devices. The optimization-based design methodology is discussed, and the resultant force profiles are validated with benchtop testing. Plans for future integration of the mechanism into a wearable device are discussed.

Keywords: AI for Movement and Neuromechanics, Robotic Rehabilitation and Assistance, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Unilateral muscle weakness often leads to asymmetric gait, disrupting interlimb coordination and stance timing. This study presents a reinforcement learning (RL)–based musculoskeletal simulation framework to (1) quantify how progressive unilateral muscle weakness affects gait symmetry and (2) evaluate whether ankle exoskeleton assistance can improve gait symmetry under impaired conditions. The overarching goal is to establish a simulation- and learning-based workflow that supports early controller development prior to patient experiments. Asymmetric gait was induced by reducing right-leg muscle strength to 75%, 50%, and 25% of baseline. Gait asymmetry was quantified using toe-off timing, peak contact forces, and joint-level symmetry metrics. Increasing weakness produced progressively larger temporal and kinematic asymmetry, most pronounced at the ankle. Ankle range of motion symmetry degraded from near-symmetric behavior at 100% strength (symmetry index, SI =+6.4%; correlation r=0.974) to severe asymmetry at 25% strength (SI =-47.1%, r=0.889), accompanied by a load shift toward the unimpaired limb. At 50% strength, ankle exoskeleton assistance improved kinematic symmetry relative to the unassisted impaired condition, reducing the magnitude of ankle SI from 25.8% to 18.5% and increasing ankle correlation from r=0.948 to 0.966, although peak loading remained biased toward the unimpaired side. Overall, this framework supports controlled evaluation of impairment severity and assistive strategies, and provides a basis for future validation in human experiments.

Keywords: AI for Movement and Neuromechanics, AI and Learning in Biomedical Robotics, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Predictive musculoskeletal simulations are a promising tool for rehabilitation and the design of assistive devices, eliminating the need for extensive data collection. Here, we compare four common methods of predictive gait generation, including two model-based approaches (muscle-reflex and central pattern generator controllers), optimal control and deep reinforcement learning. We also use the same sagittal plane musculoskeletal model, with small method-specific changes, to predict the kinematics, kinetics, ground reaction forces and muscle activations during walking. We validate the results against in-vivo data for healthy walking, and compute root-mean-square errors, Pearson correlation coefficients and the experimental match. Finally, we give examples of model sensitivities and typical deviations of predictive simulations. The results show that model-based methods and optimal control can predict physiological kinematics. However, the latter struggles to predict realistic ankle angles, while the central pattern generator shows significant deviations for the hip and knee. Deep reinforcement learning can be challenging to train, and showed major differences depending on the used musculoskeletal model.

Keywords: AI for Movement and Neuromechanics
Abstract: Standardized functional assessments provide measures of mobility capacity in controlled settings, but their episodic nature may not fully reflect performance in daily life. Emerging wearable-derived digital mobility outcomes (DMOs) may provide objective measures of real-world performance, yet they have largely focused on pace, rhythm, and variability metrics, which cannot directly capture deficits in dynamic stability or lower-extremity strength. To address this gap, we introduce AI-Sole, a cloud-enabled system that combines insole-embedded inertial and force sensors with smartphone-managed acquisition and AI-powered cloud analytics. This paper describes the ecological validation of novel center-of-pressure (COP)-derived DMOs—the 50th and 95th percentiles of the COP anteroposterior and mediolateral ranges (AP-COP, ML-COP)—and compares their measurement properties against two conventional stride-velocity (SV) DMOs, using AI-Sole data collected from healthy adults during month-long gait monitoring periods. AP-COP DMOs showed consistently lower variability and higher test–retest reliability across observation windows of different durations than both ML-COP and SV DMOs. Additionally, COP-derived DMOs showed significant in-clinic vs. real-world associations—stronger for AP-COP—whereas SV DMOs showed no significant associations with in-clinic values. Overall, these findings establish proof-of-concept feasibility of extended-time, insole-based spatiotemporal and kinetic real-world gait monitoring, paving the way for future clinical studies to evaluate the potential of COP-derived DMOs as sensitive and stable candidate DMOs for progressive neuromuscular conditions.

Keywords: AI for Movement and Neuromechanics, AI and Learning in Biomedical Robotics, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Wearable robots have the potential to improve balance for the millions of individuals who experience an increased risk of falling, such as above-knee amputees. However, to fulfill this potential, wearable robots must detect balance perturbations in a timely and accurate manner. Wearable robots have used machine learning models to detect anterior-posterior perturbations in above-knee amputees, but it remains unclear whether this approach can effectively detect lateral perturbations in this population. Additionally, new wearable robots may incorporate sensors beyond those currently available to train machine learning models. This bevy of new sensors invites the question: which wearable robots have the best sensors for training perturbation detectors? In this study, we investigated the ability of a variety of simulated wearable-sensor-driven models to detect perturbations in above-knee amputees. We synthesized wearable sensor signals from motion-capture data from three individuals with above-knee amputations who received perturbations to the pelvis from a cable-driven system while walking on a treadmill. We tested simulated sensor sets from four types of wearable robots: a pelvis-mounted safety system, a hip exoskeleton, a hip-knee exoskeleton, and a knee prosthesis. All of our sensor sets trained at least one model for each participant with an accuracy greater than 0.9. This finding suggests that wearable robots can use machine learning models to detect lateral perturbations in amputees. Additionally, we found that including pelvis sensor data significantly improves classification performance, suggesting that the best sensor sets for quickly and accurately detecting perturbations contain sensors on the perturbed body segment.

Keywords: AI for Movement and Neuromechanics
Abstract: The modulation of joint mechanical impedance, characterized by inertia, stiffness, and damping properties, enable human locomotor capacity and robustness driven by instantaneous muscle effort. Quantifying mechanical impedance during gait provides insight into neuromuscular control and the development of assistive technologies. However, existing estimation methods require hundreds of subject-specific trials. This inefficiency precludes mechanical impedance analysis from diagnostic or point of care environments. In this work, we propose a novel machine learning strategy, using a multitasking deep neural network, to indirectly predict the mechanical impedance of gait by estimating joint kinetics and kinematics. Specifically, we train this model on large-scale biomechanical datasets under a Markov assumption to predict nominal, unperturbed joint kinetics and kinematics without extensive subject specific data. The loss function optimizes joint torque, position, velocity, and acceleration over windowed data to predict kinetics and kinematics with high accuracy. Using k-fold cross-validation, the torque Mean Absolute Error (MAE) was 0.0013 Nm/kg and the displacement MAE of 0.92 degrees. Leave-one-out validation showed a torque MAE of 0.025 Nm/kg and a displacement MAE of 2.86 degrees, demonstrating generalization to unseen participants. Our mechanical impedance estimates from the estimated kinetics and kinematics account for up to 98% of the variance in perturbation-induced torque and follow some published trends in inertia, stiffness, and damping properties during healthy gait. The work presented shows potential in decreasing the data required for mechanical impedance estimates during gait, a major barrier to the clinical translation of these biomechanical measurements.

Keywords: AI for Movement and Neuromechanics, Robotic Rehabilitation and Assistance, Haptics and Human-Machine Interaction
Abstract: Recent advances in video-based gait analysis have enabled accurate clinical assessment of movement disorders using deep learning. Yet, the interpretability of these models and their correspondence to established biomechanical reasoning remain poorly understood, hindering their integration into clinical practice. This study complements our previous work on diffusion-augmented spatio-temporal graph convolutional networks (BA-STGCN) for gait assessment in children with cerebral palsy (CP). While our earlier research emphasized robustness and predictive accuracy, the present work shifts focus toward interpretability and clinical transparency. We investigate how data-driven spatio-temporal representations inherently capture biomechanical patterns aligned with clinically established gait descriptors. Specifically, we analyze classical metrics such as inter-knee distance (IKD), stride width (SW), symmetry index (SI), and hip adduction (HA), comparing them with latent embeddings derived from the BA-STGCN. Quantitative analyses reveal moderate-to-strong associations between latent dimensions and clinically established features, such as r=0.72 for IKD, indicating partial latent--clinical alignment at the single-dimension level rather than a full one-to-one encoding of clinical biomechanics. At the subspace level, canonical correlation analysis further supports this alignment rc = 0.78. The proposed framework achieved a GMFCS classification accuracy of 88.3% and a mean GDI estimation error of 4.3, demonstrating strong predictive performance alongside interpretable latent--clinical correspondence. These results suggest that incorporating explicit biomechanical context can improve prediction while making the learned representation space more relatable to established gait descriptor

Keywords: Soft Robotics, Exoskeletons, Exosuits, and Wearable Assistive Devices, Haptics and Human-Machine Interaction
Abstract: Soft robots enable safe human–robot interaction, but their continuous deformation complicates real-time state estimation, particularly under open-loop control. This paper presents a lightweight proprioceptive approach for reconstructing the 3D pose of a cable-driven soft actuator using a single commercial capacitive bend sensor. The sensor measures net tip bending, which is mapped to the full actuator shape using a Piecewise Constant Curvature model, avoiding complex processing and data-driven training. Experiments on a spiral-inspired actuator show accurate, low-latency pose reconstruction that significantly outperforms open-loop motor-based estimation and remains robust to cable slack and hysteresis, achieving millimeter-level accuracy suitable for wearable and real-time soft robotic applications.

Keywords: Soft Robotics, Haptics and Human-Machine Interaction, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Mechanically characterizing the human-machine interface is essential to understanding user behavior and optimizing wearable robot performance. This interface has been challenging to sensorize due to manufacturing complexity and non-linear sensor responses. Here, we measure human limb-device interaction via fluidic innervation, creating a 3Dprinted silicone pad with embedded air channels to measure forces. As forces are applied to the pad, the air channels compress, resulting in a pressure change measurable by off-the-shelf pressure transducers. We demonstrate in benchtop testing that pad pressure is highly linearly related to applied force (R2 = 0.998) and confirmed strong linear relationships to isometric knee torque in a clinical dynamometer with strategic pad placement. We built on these idealized settings to test pad performance in more unconstrained settings, including during cyclic dynamic and stepwise isometric bicep curls. Finally, we integrated the sensor into a lower-extremity robotic exoskeleton and recorded pad pressure during repeated squats with the device unpowered. Pad pressure tracked squat phase and overall task dynamics consistently. Collectively, our preliminary results suggest fluidic innervation is a readily customizable sensing modality with high signal-to-noise ratio and temporal resolution for capturing human-machine interaction. In the long-term, this modality may provide an alternative real-time sensing input to control / optimize wearable robotic systems and to capture user function during device use.

Keywords: Robotic Rehabilitation and Assistance, Soft Robotics
Abstract: We present a robotic method to systematically study trunk stability in seated posture, providing a quantitative basis to support clinical assessments in physical therapy. We used a robotic arm to deliver rapid, small-amplitude perturbations to n=11 subjects with a custom compliant end-effector, randomizing displacements and timing. Using the instrumental variable method on a second-order model, we extracted apparent trunk impedance parameters (apparent mass, damping, and stiffness) from the measured force-displacement response. We found that short-range stiffness consistently increased with perturbation magnitude and observed variations in compliance strategies between different individuals, with other parameters such as damping coefficients reliably falling within a fixed range. These findings demonstrate the method's capacity to capture complex subject-specific behaviors and support quantitative clinical assessment of trunk stability, postural control, and neuromuscular status.

Keywords: Soft Robotics, Robotic Rehabilitation and Assistance, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Weakness in ankle plantarflexion is a common gait impairment in children with neuromuscular disorders and in adults with reduced mobility, often resulting in diminished push-off power and inefficient walking. To address this limitation, we developed the Flash-Sole, a powered smart shoe midsole that integrates stacked dielectric elastomer actuators (SDEAs) as artificial muscles to directly augment ankle push-off torque. The objective of this study was to characterize the mechanical output of the Flash-Sole and evaluate its relevance for both pediatric and adult gait assistance. Benchtop testing was conducted using an instrumented pediatric foot–ankle model equipped with an ankle torque sensor and a heel force sensor. At approximately 5% compression, the system generated a total of 16.87 Nm of plantarflexion torque and 24.9 N of heel reaction force. The active artificial muscle component provided 11.7 Nm of torque and 11.2 N of force, while the passive elastic resistance of the SDEA structure contributed to 5.15 Nm and 13.7 N, respectively. Notably, the active output remained relatively stable across compression levels, whereas the passive contribution increased with strain due to material resistance. The total torque output corresponds to approximately 60–85% of the push-off demand of a 25 kg child (~20–28 Nm) and provides ~10–15% of the requirement for a 70 kg adult (~70–90 Nm). These findings demonstrate that modest sole deformations are sufficient to produce clinically meaningful assistance within a compact, footwear-integrated platform. By combining stable active actuation, intrinsic compliance, and silent operation, the Flash-Sole offers a lightweight and unobtrusive alternative to rigid motor-based orthoses.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance, Soft Robotics
Abstract: Several studies have demonstrated the potential of exoskeletons in reducing muscle load and mitigating the pathophysiological mechanisms underlying musculoskeletal disorders. Most focus on the main articulations, such as the back, shoulder, or knees, while only a few address the wrist. This paper details the design and Range of Motion (RoM) assessment of FleXoWrist, a wrist exoskeleton intended for occupational use. FleXoWrist is built on a glove with tendon-driven mechanisms actuated remotely via DC motors located in a backpack, as they offer more reliable, robust, lightweight, safe, and cost-effective solutions. FleXoWrist kinematic performance has been validated on a rigid plastic mannequin with a custom-made soft hand and compared to human subjects’ kinematic data. Experimental results show that FleXoWrist meets the range of motion requirements of the workers by assisting all main wrist movements: ±56◦ in flexion/extension, ±28◦ in radial/ulnar deviation, and 40◦ of finger flexion. Furthermore, FleXoWrist allows rotation around an axis very close to that of a subject performing the Dart Throwing Motion, which is considered the most natural wrist movement, providing a slight ulnar and radial deviation during flexion and extension movements, respectively.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices, Robotic Rehabilitation and Assistance, Soft Robotics
Abstract: Vibration therapy (VT) is a promising non-invasive intervention for neurorehabilitation, yet its clinical optimization is hindered by the absence of real-time neuromonitoring and the incompatibility of conventional VT devices in neuroimaging settings. Here, we demonstrate a magnetoencephalography (MEG)-compatible soft robotic glove designed to deliver precise, customizable vibrotactile stimulation to individual finger joints while enabling simultaneous high-resolution monitoring of the brain function. The glove employs pneumatic actuators integrated into a lightweight textile platform, controlled via a modular system that ensures frequency-accurate vibration and safety through real-time pressure monitoring. Synchronization with MEG acquisition is achieved using embedded trigger signals, allowing time-locked correlation of tactile stimulation and cortical activity. Feasibility tests in a typically developing child and a child with hemiplegic cerebral palsy demonstrated artifact-free somatosensory evoked fields (SEFs), accurate vibration delivery at 20 Hz and 30 Hz, and frequency-specific cortical entrainment in primary (S1) and secondary (S2) somatosensory areas. These findings validate the glove’s compatibility with MEG and its potential for personalized VT protocols. This platform establishes a foundation for closed-loop neuromodulation strategies and individualized neurorehabilitation setups in children.

Keywords: Soft Robotics, Bionics and Prosthetics
Abstract: The development of high-fidelity vascular simulators is pivotal for surgical training, especially in endovascular and minimally invasive procedures where tactile sensitivity and precise force control are critical. In this work, we present a robotic vessel simulator capable of reproducing both physiological and pathological flow conditions, with interchangeable vessels having different properties, including diameter, stiffness, and pressure waveform. The system integrates an active hydraulic circuit that replicates realistic pulsatile circulation using a blood-mimicking fluid, closely emulating the dynamic behavior of real vessels. Furthermore, embedded pressure sensors, individually calibrated for each artificial vessel, enable online estimation of forces applied to the vessel walls. This approach not only may enhance the surgical training but also facilitate quantitative assessment of their force application, supporting objective evaluation.

Keywords: Robotic Surgery and Diagnosis, Soft Robotics
Abstract: This paper presents a variable-curvature kinematic modeling framework for a cable-driven continuum robot with passive, non-uniform structural stiffness along its arc length. The proposed continuum section uses a graded notch pattern, with larger notch spacing near the base and smaller notch spacing toward the distal tip, producing higher proximal stiffness and greater distal flexibility. Conventional constant-curvature models approximate the backbone as a circular arc and may be insufficient for continuum robots whose stiffness varies along their length. To address this limitation, we formulate a mechanically motivated Euler-spiral approximation in which curvature increases along the arc length. The model explicitly uses differential actuation of two antagonistic cable pairs controlling pitch and yaw bending angles. Forward and inverse kinematic relations are derived for the variable-curvature backbone geometry. The proposed model is presented as a simulation-based and mechanically motivated approximation. The framework provides a tractable basis for future experimental identification and model-based control of graded-stiffness continuum robots.

Keywords: Surgical Robotics: Steerable Catheters/Needles, Optimization and Optimal Control, Motion Control
Abstract: This paper presents an error-state Model Predictive Path Integral (MPPI) framework for tendon-driven continuum robots (TDCRs). Tracking-error dynamics are formulated on a Lie group to preserve full pose geometry, yielding precise position–orientation error metrics. A nonlinear Cosserat-rod model with strain parameterization provides a closed-form TDCR dynamics representation and updates in 0.3 ± 0.3 ms. The model is calibrated via weight-release and actuation experiments on robotic ablation catheters, and its generalized coordinates are estimated through nested optimization. The MPPI controller parallelizes trajectory sampling and evaluation, uses tendon displacement actuation computed via optimization to eliminate force sensors, and is uncertainty-aware through a simple and efficient exponentially weighted moving-average (EWMA) estimator embedded in the running cost. Control trajectories are sampled around the current best sequence and evaluated with an adaptive cost and exponential weighting to bias low-cost solutions. Experiments comparing conventional model predictive control (MPC), Lie-group MPC, and offline Implicit Q-Learning (IQL) show that our MPPI method achieves the highest accuracy, significantly better computational efficiency than MPCs, and better overall accuracy than all baselines. The model further extends naturally to multi-segment TDCRs and can incorporate tendon-actuation friction.

Keywords: Soft Robotics, Biologically-Inspired Robotics and Biomimetics
Abstract: Soft, tip-extending vine robots are well suited for navigating tight, debris-filled environments, making them ideal for urban search and rescue. Sensing the full shape of a vine robot’s body is helpful both for localizing information from other sensors placed along the robot body and for determining the robot's configuration within the space being explored. Prior approaches have localized vine robot tips using a single inertial measurement unit (IMU) combined with force sensing or length estimation, while one method demonstrated full-body shape sensing using distributed IMUs on a passively steered robot in controlled maze environments. However, the accuracy of distributed IMU-based shape sensing under active steering, varying robot lengths, and different sensor spacings has not been systematically quantified. In this work, we experimentally evaluate the accuracy of vine robot shape sensing using distributed inertial measurement units along the robot body. We quantify IMU drift, measuring an average orientation drift rate of 1.33°/min across 15 sensors. For passive steering, mean tip position error was 11% of robot length. For active steering, mean tip position error increased to 16%. During growth experiments across lengths from 30–175 cm, mean tip error was 8%, with a positive trend with increasing length. We also analyze the influence of sensor spacing and observe that intermediate spacings can minimize error for single-curvature shapes. These results demonstrate the feasibility of distributed IMU-based shape sensing for vine robots while highlighting key limitations and opportunities for improved modeling and algorithmic integration for field deployment.

Keywords: Robotic Surgery and Diagnosis
Abstract: Concentric tube robots (CTRs) offer significant potential for minimally invasive surgery due to their compact form factor and ability to navigate constrained anatomical pathways. However, achieving reliable millimeter-scale tip positioning remains challenging due to unmodeled nonlinearities including friction, torsional coupling, and hysteresis. In this work, we present the first demonstration of fiber Bragg grating (FBG)-based task-space closed-loop control for a three-tube CTR. A multicore FBG sensor is integrated within the inner lumen of the CTR, enabling distributed curvature measurement and tip position estimation at 100 Hz. The measured tip position is incorporated into a proportional feedback controller that corrects the commanded trajectory. Experimental validation across three representative trajectories, a planar circular path, a planar infinity-sign path, and a 3D helical spiral, demonstrates that closed-loop control achieves mean tip tracking error of 0.98 mm, compared to 1.72 mm for open-loop execution, representing a 1.82x improvement. These results establish FBG-based closed-loop control as a viable pathway toward the mm-scale accuracy required for neurosurgical applications.

Keywords: Robotic Surgery and Diagnosis
Abstract: Tendon-driven robots are a promising technology for minimally invasive surgical procedures, particularly in anatomically constrained environments such as pediatric cranial surgery. This type of robotic system has been demonstrated to follow cranial osteotomy paths using a follow-the-leader (FTL) algorithm. However, mechanical coupling, tendon elasticity, and backlash effects resulted in open-loop tip positioning errors ranging from 2.5 mm to 21.5 mm, limiting surgical precision. This paper presents a closed-loop tip and shape control framework that integrates real-time shape feedback using embedded Fiber Bragg Grating (FBG) sensors. The multi-core FBG system enables high-frequency curvature reconstruction along the robot backbone, providing continuous shape information for tip trajectory and shape correction. A control architecture processes FBG measurements to extract tip and shape position feedback, converts them to the joints' bending angles, and generates corrective tendon commands. Experimental validation shows that the closed-loop system reduces full-shape tracking errors by up to 88%, maintains about 1 degree RMS joint error on tip trajectories, and achieves sub-millimeter FBG reconstruction accuracy (0.30 mm mean error). These improvements establish a foundation for precision surgical robotics in confined anatomical environments.

Keywords: Biologically-Inspired Robotics and Biomimetics, Bionics and Prosthetics, Haptics and Human-Machine Interaction
Abstract: Conventional underactuated grippers often sacrifice precise parallel motion for enveloping adaptability. This paper presents a novel gripper, SPEG Hand that achieves sequential pinching and enveloping grasping with a single actuator, employing a unique mechanical sequencing strategy. The gripper utilizes a Chebyshev linkage to generate initial linear motion of the distal phalanx for pinching. The key innovation lies in the transmission system: an internal preset angular idle travel separates the pinching and enveloping phases. During the first 180° of input rotation, the motion drives only the Chebyshev linkage for pure parallel closure. Upon contact, the parallel motion is obstructed. The continued rotation of the input shaft then completes the idle stroke. The cam subsequently pushes against the parallelogram linkage. This action does not deform a single parallelogram but drives the entire dual-parallelogram finger assembly, causing the fingertip mount to rotate around its limit block, thereby achieving a controlled enveloping motion—all without the need for additional sensors or torque limiters. This design embodies a purely geometric motion-conversion principle, where blocked translation is mechanically converted into rotation. Experimental results demonstrate robust and adaptive grasping across various objects. This work provides a simple, reliable, and mechanically intelligent solution for enhancing robotic dexterity in unstructured environments.

Keywords: Bionics and Prosthetics, Robotic Rehabilitation and Assistance, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Despite the increasing availability of powered upper-limb (UL) prostheses, robust and cost-efficient control of intermediate joints—specifically the elbow and wrist—remains a significant challenge. Current solutions, like high-density EMG arrays with pattern recognition or surgical interventions like Targeted Muscle Reinnervation (TMR), are complex and often insufficient for robust simultaneous multi-joint control. Consequently, the reliance on limited myoelectric sites creates a bottleneck, restricting not only the use of multiple prosthetic joints but also the functional fidelity of available advanced pros- thetic hands. In this work, we present a two-degree-of-freedom (2-DoF) extension of the Compensation Effect Amplification Control (CEAC) paradigm. This approach leverages natural body kinematics—including trunk flexion, lateral bending, and shoulder abduction and rotation—to drive the velocity of the prosthetic elbow and wrist. Crucially, this movement- based strategy liberates distal electromyography (EMG) signals, allowing them to be dedicated exclusively to hand actuation. To validate this hybrid control architecture, we conducted a study with ten able-bodied participants using a supernumerary prosthesis. We implemented a simultaneous control scheme where the elbow and wrist were driven by CEAC, while the hand was actuated via sEMG on the forearm. We benchmarked this system against a remote-controlled prosthesis representing an idealized pattern recognition classifier. Participants achieved proficiency with CEAC within one hour, and the method outperformed the remote control benchmark in terms of both completion time and error rate. Furthermore, the approach significantly reduced unergonomic compensatory behaviors and was preferred by a majority of participants.

Keywords: Bionics and Prosthetics
Abstract: Modern upper-limb prostheses offer a high number of functional degrees of freedom that still cannot be controlled in an intuitive manner. Movement-based control approaches aim to overcome these limitations by leveraging the user’s natural motor behavior as a control input. Methods such as Compensation Effect Amplification Control (CEAC) have been shown to be intuitive and effective for various tasks, but CEAC exhibits a significant loss in performance in situations where a single postural control mapping (i.e. the relationship between human and prosthetic movement) is insufficient. To increase the versatility of CEAC, we propose a hybrid control architecture, combining the postural input with a discrete voluntary input.The proposed framework was evaluated in a study with ten asymptomatic subjects wearing a supernumerary prosthesis to perform an advanced reach-to-grasp task. Trunk flexion served as the postural control input for the prosthetic elbow, while a 2-channel-myoelectric signal was used to open/close the prosthetic hand and to modulate the postural control law. Using simple cocontraction patterns, users could temporarily switch to either of two additional control mappings. Performance was compared against a fixed postural controller (CEAC) and a remote-controlled elbow as an ideal benchmark. The new control architecture allowed participants to significantly reduce compensatory movement compared to CEAC. The results indicate that adding a sparse auxiliary signal such as EMG presents a viable approach to increasing the versatility of movement-based control for upper-limb prostheses.

Keywords: Bionics and Prosthetics, AI for Movement and Neuromechanics, AI and Learning in Biomedical Robotics
Abstract: Although upper limb losses profoundly affect daily life activities of people with limb differences, the adoption of advanced prostheses remains limited. While significant progress has been made on prosthetic devices and control strategies, user–device interfaces have received comparatively less attention. Specifically, high-density surface electromyography (HD-sEMG) systems still rely on wet skin–electrode interfaces for both recording and reference electrodes, limiting their applicability in the daily use of prosthetic devices. To address this gap, we present the development of two full-dry HD-sEMG interfaces, in monopolar and bipolar configuration, comparing them with gold-standard HD-sEMG patch using conductive gel. We evaluated the performances through six signal quality metrics, and classification outcomes obtained from six data-driven algorithms. The dry systems demonstrated satisfactory performance, with the bipolar interface exhibiting cleaner spectral characteristics and superior classification accuracy, indicating the suitability of fully dry HD-sEMG systems for prosthetic myocontrol.

Keywords: Bionics and Prosthetics, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Body-powered prosthetic terminal devices can offer direct haptic feedback but are limited by predetermined voluntary-open or -close configurations, constraining grasp versatility and user comfort. This work presents a novel split-hook terminal device capable of user-initiated switching between voluntary-open and -close modes while preserving consistent cable path length, cable input position, and excursion characteristics. The proposed design uses a gear-based transmission with cam-constrained linkages for directional reversal. An adjustable strip-spring mechanism is incorporated to enable tuning of restoring forces without tools or hardware replacement. The testing prototype was manufactured in PA11 nylon with multi- jet fusion. The 101.54 g prototype achieved a 73 mm prehensor opening with 53 mm cable excursion, matching the clinical recommendation. Testing demonstrated efficient force transmission for both modes; in voluntary close mode, forces of 6.97 ± 0.09 N and 27.10 ± 0.09 N were measured at 10 N and 40 N inputs, respectively. Both modes also exhibited linear force- excursion profiles (R2 > 0.92). Having both voluntary-open and -close modes would mitigate user fatigue during sustained grasps while affording proportional force control, potentially enhancing functional independence and comfort for upper-limb amputees.

Keywords: Bionics and Prosthetics, Robotic Rehabilitation and Assistance, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Powered prostheses are capable of providing net positive work to amputees and have advanced in the past two decades. However, reducing amputee metabolic cost of walking remains an open problem. The Law of Intersegmental Coordination (ISC) has been observed across gaits and previously implicated in energy expenditure of walking, yet it has rarely been analyzed or applied within the context of lower-limb amputee gait. This law states that the elevation angles of the thigh, shank and foot over the gait cycle covary. In this work, we developed a method to analyze intersegmental coordination for lower-limb 3D kinematic data, to simplify ISC analysis. Moreover, inspired by motor control, biomechanics and robotics literature, we used our method to extend ISC to a new law of coordination of moments. We find these Elevation Space Moments (ESM), and present results showing a moment-based coordination for able bodied gait. We also analyzed ISC for amputee gait with powered and passive prostheses, and found that while elevation angles remained planar, the ESM lacked planar coordination. We present an ISC-driven powered prosthetic control framework, using healthy coordination as a constraint to predict the shank angles/moments to compensate for alterations due to a passive foot. We developed the ISC3d toolbox that is freely available online, which may be used to compute kinematic and kinetic ISC in 3D. This provides a means to further study the role of coordination in gait and may help address fundamental questions of the neural control of human movement.

Keywords: AI and Learning in Biomedical Robotics, Robotic Surgery and Diagnosis
Abstract: Age-related Macular Degeneration (AMD) is a leading cause of irreversible central vision loss in older adults. While current treatments require repeated administrations, gene therapy delivered via subretinal injection could potentially become a curative treatment. Robotic assistance enhances accuracy (i.e., precision delivery and stabilized eye and instrument) in these procedures, though most systems remain preclinical, and those tested clinically typically involve only a small number of cases. This study reports the treatment of 20 patients with submacular hemorrhage using co-manipulation Stableyeser system. As of now and as far as the authors are aware, this represents the largest cohort of robot-assisted vitreoretinal interventions to date. All injections were successfully completed with an average duration of 2min 54s, a setup time of 17min, and no robot-related adverse events. Additionally, we evaluate retinal and instrument segmentation during robot assisted surgery using SAM2 and an Adapted Mean Teacher (AMT) framework, demonstrating superior performance of AMT for iOCT segmentation. These results highlight the feasibility of robotic subretinal therapy and of automated image analysis in clinical settings.

Keywords: Robotic Rehabilitation and Assistance, Haptics and Human-Machine Interaction, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: For individuals with Spinal Cord Injury (SCI), upper-limb assessment is essential for evaluating the severity of impairment and for monitoring rehabilitation outcomes. The Graded Redefined Assessment of Strength, Sensibility, and Prehension (GRASSP) is a clinically validated multimodal assessment, but it is time-consuming and requires clinical expertise. This study investigates whether feature-space metrics derived from ultrasound-based muscle sensing or sonomyography (SMG) can serve as surrogate indicators of GRASSP subtest performance. A-mode ultrasound signals were collected using an eight-channel forearm sensor array while individuals with SCI performed eight hand gestures. Multiple features were extracted and embedded into a two-dimensional latent space using Uniform Manifold Approximation and Projection (UMAP). Feature-space outcome metrics quantifying cluster compactness and inter-class separation were computed, including the Silhouette Score, Davies–Bouldin index, Calinski–Harabasz index, minimum inter-centroid distance, and mean inter-centroid distance. Among these metrics, the silhouette score showed the strongest correlation with overall GRASSP performance and was therefore further analyzed on a gesture-specific basis. Strong associations were observed with GRASSP subtests reflecting muscle strength and prehension ability, moderate correlations with quantitative prehension performance, and weak correlations with sensation subtests, consistent with the motor-specific nature of ultrasound sensing.

Keywords: Bionics and Prosthetics, AI and Learning in Biomedical Robotics, Haptics and Human-Machine Interaction
Abstract: Myoelectric prostheses can enable intuitive control, yet achieving natural and reliable movements remains challenging and contributes to device abandonment. We present a physics-based virtual reality (VR) environment for co-adaptive training of an unsupervised incremental non-negative matrix factorization (iNMF) control scheme for hand prostheses. The VR system includes three scenarios covering initialization, closed-loop user–controller co-adaptation, and evaluation during activities of daily living (ADL). Feasibility was examined in a pilot study with five participants using task performance, reconstruction error, and perceived workload (NASA-TLX). During co-adaptation, reconstruction error generally decreased with participant-specific variability, and post-training models better reconstructed ADL data than the initialization model. VR task performance improved across repetitions, indicating learning and adaptation, while workload tended to decrease with longer training but increased for more complex ADL tasks. Overall, the results support the feasibility of VR-based training for unsupervised prosthetic myocontrol.

Keywords: Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: The velocity vector field (flow) controller is widely used control framework for lower-limb exoskeletons. In this study, we examine this control approach and propose modifications to improve its performance. We show that flow control acts as a variable proportional-derivative error regulator, where the parameter Gamma representing the desired magnitude of the hip-knee joint velocity vector (path speed). Based on this, we introduce two modifications to Gamma: (1) a constant Gamma set to the mean desired path speed, and (2) a variable Gamma that mimics natural path speed during unassisted walking. We compared the modified flow controllers with a slow-Gamma version in experiments with seven participants walking on a treadmill at 0.6m/s, 0.8 m/s, and 1.0 m/s. Compared to the slow-Gamma controller, the RMS kinematic tracking error decreased by 30.7 +- 11.3% and the range of motion of the knee increased by 48.2 +- 5.5% for the mean-Gamma controller, while the variable-Gamma controller had 32.4 +- 14.7% smaller RMS error and 50.5 +- 6.5% larger range of motion of the knee. Moreover, the slow-Gamma controller consistently produced resistive power, whereas participants reported a more comfortable and natural gait with the modified controllers. These findings demonstrate that systematic tuning of Gamma enables controlled modulation of gait kinematics, offering a practical strategy for optimizing flow controller performance.

Keywords: Biologically-Inspired Robotics and Biomimetics, Soft Robotics, Haptics and Human-Machine Interaction
Abstract: Manipulation of a dynamically complex object - a whip - remains a challenge in both human motor neuroscience and robotics. A model of the object's dynamics appears necessary for prediction and control. However, for humans, an accurate model of the whip appears unrealistic to maintain, and approaches in robotics still remain data- and computation-heavy and rely on visual feedback. This study examined humans manipulating a whip and explored the hypothesis that the whip's center of mass (CoM) presents a feasible simplification of the whip's dynamics that humans may use for predictive control. Motivated by the skill of throwing, we examined human aiming with the whip's CoM, compared to using the hand or the tip of the whip as focal points for aiming. Results showed that participants accurately aimed the whip CoM at the target at the moment when it reached its maximum velocity, like a projectile thrown at the target. The whip CoM's aiming was more accurate than that of the hand and the tip, and it predicted task error. These results suggest that the center of mass of the continuously changing object configuration may act as a low-dimensional collective variable that facilitates efficient feedforward planning in dynamic manipulation tasks.

Keywords: Bionics and Prosthetics, AI and Learning in Biomedical Robotics, Exoskeletons, Exosuits, and Wearable Assistive Devices
Abstract: Personalized wearable devices (orthoses, exoskeletons, prosthetic sockets) require accurate geometric representations of patient anatomy, yet current workflows rely on manual CAD modeling. We present an automated pipeline that learns sparse parametric representations directly from 3D scans. Our method combines an optimized anchor-based patch decomposition with PointNet-based control point regression to transform dense point clouds (10k-400k points) into compact Bezier parameters (16 control points per patch). Evaluation on clinical scans achieves mean reconstruction errors below 12mm while maintaining CAD compatibility. The resulting low-dimensional representations enable rapid design iteration, seamless fabrication workflows, and compatibility with physics simulators for robotic applications. We further compare against a closed-form least-squares baseline on the same patches and demonstrate that the learned estimator provides a bounded worst-case error (Hausdorff one to three orders of magnitude lower)