ABSTRACT. Mapping and monitoring aquatic environments can benefit from hybrid aerial–amphibious drones able to combine flight and water-surface navigation within the same mission. This paper presents a PX4 firmware extension for such platforms, introducing manual and autonomous marine navigation modes integrated with the standard PX4 mission pipeline and QGroundControl interface. The proposed framework preserves existing flight functionalities and safety mechanisms while enabling unified planning and execution of hybrid aerial–marine missions with differentiated aerial and marine waypoints. Simulated case studies validate the implementation and demonstrate stable surface navigation under calm and wavy conditions.

ABSTRACT. Cable-Driven Parallel Robots (CDPRs) are well suited to operate on large vertical surfaces because they combine a large workspace, low moving mass and high payload-to-weight ratio. This paper presents the integration of a painting end-effector and a G-code-based control layer on the deployable Irma8 CDPR. A waterborne airspray system was selected for safe indoor validation, while the software architecture was composed of an asynchronous parsing stage, look-ahead trajectory planning for continuous path generation, and online tension distribution to manage cable constraint redundancy. Experimental validation serves as a proof of concept, confirming the potential of CDPRs as flexible systems for practical applications in construction robotics.

ABSTRACT. The increasing regulatory focus on the circular economy has placed the recycling of textile waste at the core of sustainability strategies, where accurate identification of fiber composition is a critical step for effective material recovery. This work introduces HyperS2ENSE, an Artificial Intelligence-based perception module designed for automatic textile sorting systems operating in industrial and robotic environments. The component proposed in this work relies on a Vision Transformer specifically tailored for the classification of blended textile materials from hyperspectral data. The impact of introducing activation functions in the spatial and spectral tokenization stages is evaluated, with a focus on improving the characterization of high-dimensional spectral signatures in mixed-composition fabrics. Experimental results on real data demonstrate that HyperS2ENSE effectively improves the separability of blended fabric classes, outperforming the baseline model without non-linearities, while preserving computational efficiency.

ABSTRACT. Ensuring safe and efficient navigation in shared human-robot industrial environments remains a primary challenge for standard frameworks such as ROS2 Nav2, which often lack semantic context and treat all obstacles as uniform geometric entities. This paper presents SALP: Semantic Aware Local Planner, a modular extension of the Dynamic Window Approach-based (DWB) controller designed to inject semantic awareness directly into the local trajectory evaluation process. By combining YOLO-based object detection with 3D iToF sensing, the system constructs an asynchronous semantic grid that assigns class-specific costs, prioritizing human safety, to the candidate trajectories generated by the local planner. Unlike traditional semantic costmap layers, SALP minimizes computational overhead by applying semantic reasoning exclusively to the limited set of candidate paths during the scoring phase. Experimental validation was conducted on a holonomic autonomous ground vehicle (AGV) within a real-world semi-structured industrial warehouse. The findings confirm that SALP provides a robust, parameter-driven foundation for human-safe navigation. In experimental trials, the system maintained a 10–20 Hz control frequency while processing semantic updates at 2–5 Hz. By evaluating 10–20 candidate trajectories per cycle, the planner maintained a minimum safety margin of 0.5 m from humans while maintaining a trajectory efficiency within 5% of the optimal path length.

ABSTRACT. Model-based techniques have gained popularity for their high performance, but uncertainties in the model can render them inefficient, leading to task execution failures. This paper introduces a novel approach, grounded in the concept of closed-loop state sensitivity, to ensure robust task execution for redundant manipulators in the presence of uncertainties in the robot model. Specifically designed for tracking control of predefined trajectories, our strategy computes an optimal null space input that allows the robust satisfaction of constraints against model uncertainties during motion, such as joint limit avoidance. Simulation results on a Franka Emika Panda validate the proposed approach and demonstrate its capability to satisfy task and actuator constraints under parameter uncertainty.

ABSTRACT. Robotic teleoperation is widely adopted in industrial and safety-critical environments to enable human intervention in hazardous conditions. The remote control of the platforms relies on multi-view visual interfaces combining heterogeneous visual feedback sources. Network-induced latency remains a major challenge, as it degrades system transparency, increases operator workload, and negatively affects task performance. While the impact of latency on task execution has been extensively investigated, its influence on visual attention allocation within modern multi-view teleoperation interfaces remains less understood. This work presents a gaze-based approach that aims at characterising visual strategies during repetitive robotic manipulation tasks performed through a Human-Robot Interaction Interface integrating multiple visual feedback sources, both without network latency and with artificially induced network latency. Understanding these visual adaptation strategies may support the design of more effective human-centred teleoperation interfaces for industrial and safety-critical applications.

ABSTRACT. We present a physics-informed autoregressive Deep Operator Network (DeepONet) surrogate for the two-dimensional damped wave equation with external forcing, targeting real-time deformation of a membrane in an interactive virtual reality (VR) training environment for surgeons. The membrane dynamics are modeled as a system of two coupled partial differential equations (PDEs) for the deformation u(x,y,t) and velocity v(x,y,t), subject to fixed boundary conditions and contact forces. Our DeepONet employs a trunk network that maps space-time coordinates (x,y,t) to modal basis functions for u and v, and a branch network that maps initial conditions and external forcing measurements to the corresponding modal coefficients. The networks are first trained on a reduced-order dataset obtained via singular value decomposition (SVD) of 15000 spatiotemporal samples, encompassing free evolution, constant forcing, and mixed scenarios. We then perform curriculum training that gradually transitions from purely data-driven supervision to fully physics-informed optimization, incorporating constraints on the damped wave equation residual, and rollout over increasing time horizons. The resulting surrogate supports deployment in a rollout fashion within simulation engines such as Unity, providing stable and accurate predictions of membrane deformation at interactive rates suitable for VR-based surgical training.

ABSTRACT. We present a solution to the inverse dynamics problem for robots with transmissions that display nonlinear flexibility between motors and links and include nonlinear friction on the motor and link sides of the joints. For a desired link trajectory, computation of the associated motor trajectory is performed with a numerical method, while only analytical differentiations are needed to further obtain the required motor torque. Under suitable assumptions of smoothness, this mixed algebraic-numerical approach is shown to be quite effective, with only few iterations needed at each time step. Some illustrative results are reported for a two-joint robot of the considered class, which is used as a benchmark in the literature. We also discuss extensions and efficient implementations for manipulators with a large number of flexible joints.

ABSTRACT. Mobile robots operating in low-light environments frequently rely on auxiliary active-illumination to improve scene visibility for Visual Odometry and SLAM. However, robot-mounted light sources introduce localized saturation artifacts, referred to as glow, which degrade image gradients, affect feature matching, and reduce pose estimation accuracy. This work presents a deep learning-based glow-aware perception frame- work for robotic navigation under active-illumination. The proposed ar- chitecture combines glow localization, pixel-wise confidence estimation, and ROI-based restoration within a unified multi-task pipeline. To sup- port supervised training, a synthetic glow-generation pipeline is intro- duced, producing paired degraded and clean images with controllable saturation artifacts, confidence maps, and bounding-box annotations. The estimated confidence maps are integrated into ORB-SLAM3 to sup- press unstable features in glow-affected regions, improving tracking ro- bustness. Experimental results on real indoor sequences acquired with a mobile robotic platform demonstrate improved trajectory estimation accuracy under challenging active-illumination conditions

ABSTRACT. The mechanical design of physical human–robot interaction (PHRI) systems is driven by human biomechanical limits and application-specific requirements, which together define system specifications. The human biomechanical workspace and torque capability define the upper bounds of the interaction space, while the intended application determines the range of motion (ROM) and torque ranges to be reproduced. However, these requirements are commonly specified independently along the principal degrees of freedom (DoFs), despite the inherently coupled nature of human motion and force generation. This assumption underlies conventional design approaches and may neglect coupling effects that influence human–robot interaction behavior. To address this limitation, this work introduces capability regions as a quantitative framework for the design and evaluation of multi-DoF PHRI systems. Capability regions describe the admissible combinations among multiple DoFs separately in the kinematic and torque domains, enabling direct comparison of human functional requirements and robot mechanical performance. The framework is applied to two wrist rehabilitation robots with different architectures: EDUSA® PRO and the MIT-MANUS wrist robot. Their kinematic and torque capability regions are assessed and compared with the biomechanical characteristics of a healthy human wrist. Results show that EDUSA® PRO provides a 2.03-fold larger FE-AA ROM capability area than the MIT-MANUS wrist robot. The comparison highlights how capability distributions across the multi-DoF space can reveal whether each system meets the required motion and torque combinations and how these distributions are shaped by the actuation architecture. The proposed framework therefore provides a quantitative basis for capability-driven and requirement-oriented PHRI system design.

ABSTRACT. Robot-aided rehabilitation is a promising approach for improving upper-limb motor recovery. Recent studies suggest that auditory feedback may enhance motor learning, engagement, and movement quality. However, the use of performance-dependent auditory feedback in upper-limb robotic therapy remains limited.

This pilot study investigates the effect of real-time auditory feedback during a robot-aided point-to-point reaching task in four healthy participants. Auditory stimuli were generated online according to deviations from a minimum-jerk reference trajectory, providing performance-dependent cues related to movement smoothness. Kinematic, physiological, and neurophysiological measures were assessed to evaluate the impact of the proposed feedback paradigm.

Preliminary results suggest that auditory feedback can improve upper-limb motor performance, promoting smoother and more coordinated movements, as suggested by the jerk index (Jn = 0.20 +- 0.05 and Jn = 0.18 +- 0.04 without and with feedback, respectively). In addition, it elicits measurable physiological responses indicative of increased sensorimotor engagement and modulates neural oscillations associated with performance monitoring, suggesting enhanced real-time error processing during movement. These findings support the feasibility of integrating sound-based, performance-dependent cues into robot-aided upper-limb rehabilitation as a complementary approach to conventional therapy.

ABSTRACT. The deployment of quadruped robots in complex environments requires advanced teleoperation interfaces capable of managing high-dimensional, multi-degree-of-freedom locomotion and postural control. While grounded haptic devices offer high-fidelity control and active force feedback, wearable haptic interfaces provide ergonomic, untethered mobility. However, a rigorous comparison of these paradigms for legged robotics remains largely unexplored. This paper presents a comprehensive quantitative and qualitative benchmarking of two distinct teleoperation frameworks for a quadruped robot: a grounded haptic interface (Geomagic Phantom Omni) and a multi-modal wearable haptic interface leveraging inertial and electromyographic sensors (Myo Armband). Fourteen participants navigated a Unitree Go2 robot through a simulated maze. Performance was evaluated using a kinematic metrics alongside subjective surveys (NASA-TLX and System Usability Scale). Results demonstrate that the grounded haptic interface significantly outperforms the wearable system in navigation efficiency (50% lower completion time), spatial accuracy (33% lower target error), and cognitive offloading (8 ± 11 NASA-TLX score of the grounded haptic interface against 44 ± 22 of the wearable interface). Despite the physical untethering provided by the wearable interface, its reliance on discrete muscle activations and mid-air gestures induced cautious, segmented movements (14% lower mean jerk) and rapid operator fatigue. Ultimately, this study demonstrates that for complex, high-dimensional legged teleoperation, the cognitive benefits and spatial guidance of virtual force constraints outweigh the physical constraints of grounded devices.

ABSTRACT. Conventional computer vision-based semiautonomous prosthetic systems rely on fixed object categories, limiting generalizability across manipulation scenarios. This work presents an affordance-driven control framework integrating sEMG-based user intention estimation with open-vocabulary semantic perception for grasp planning in cluttered environments. Semantic descriptions are combined through temporal voting to identify the target object and infer grasp and approach strategies. A phase-dependent controller enables concurrent hand-wrist coordination during preshaping while preserving direct user control during manipulation. Offline validation identified a 320p resolution and a confidence threshold of 0.001 as the best configuration, achieving accuracy=94.4%. Experiments with three healthy participants using a commercial prosthesis yielded a mean Success Rate of 0.92±0.10 and a completion time of 32.88±6.96s during drinking, while cluttered object manipulation was completed in 17.07±7.16s without category-dependent differences. These results demonstrate the feasibility of open-vocabulary semantic perception into semiautonomous prosthetic control for robust affordance-based object selection and efficient execution of daily-life manipulation tasks.

ABSTRACT. In human-robot collaboration (HRC), robot pacing can influence both task execution and the human operator's internal state. While this effect has been studied mainly in the context of industrial collaborative manipulators, its role in collaboration with service robots remains less explored. Unlike industrial manipulators, service robots can combine task-oriented actions with socially interpretable and prosocial behaviors, which may shape how fluent and high-quality the interaction is perceived to be.

This paper presents a multimodal user study examining how different pacing levels of a TIAGo service robot affect behavioral, psychophysiological, and subjective responses during a handover and disassembly task. Throughout the interaction, the robot orients its head toward the user's face, sustaining visual attention as a simple prosocial, user-aware behavior. Fourteen participants interacted with the robot under three pacing conditions; head-orientation dynamics, ECG, respiration, and galvanic skin response were continuously recorded. Perceived workload and affective experience were assessed using the NASA-TLX and SAM questionnaires.

Results showed that faster pacing increased visual-attentional coordination, physiological activation, and perceived task demands. In particular, head motion rate increased, while SAM arousal increased from 3.4 ± 1.8 to 6.3 ± 1.6 from slow to fast pacing. NASA-TLX temporal demand also showed a clear increase, from 1.8 ± 1.4 to 5.6 ± 2.7. Conversely, valence, frustration, and self-rated performance were not negatively affected.

ABSTRACT. Pain management is a critical challenge in pediatric healthcare, as needle-based medical procedures often provoke anxiety, distress, and negative emo-tional responses in children. In recent years, socially assistive robots have emerged as a key factor to support children during medical procedures. In this work, we present a hybrid gesture generation framework for social robots de-signed to support pediatric pain management during blood sampling. The pro-posed approach combines GAN-based co-speech gesture synthesis with prede-fined custom gestures identified through a multidisciplinary co-design process involving clinicians and psychologists from two paediatric clinical centers across eight sessions. The hybrid mechanism enables the robot to maintain rhythmic speech-gesture coordination while inserting semantically meaningful gestures aligned with specific keywords in the dialogue. The framework was implemented on the Pepper robot and integrated into an interaction scenario explaining the blood sampling procedure to children. A preliminary evaluation with 28 healthy participants assessed the perceived quality of the gestures us-ing a custom questionnaire and selected dimensions of the Godspeed question-naire. Results showed a positive evaluation of gesture quality, particularly in terms of speech-gesture alignment. Gesture evaluation was significantly asso-ciated with perceived animacy, likeability, and anthropomorphism of the ro-bot. These findings support the clinical potential of the proposed framework, which is currently being evaluated in an ongoing larger-scale study with pae-diatric patients.

ABSTRACT. This paper presents the design, modeling, and control of an underactuated two-degree-of-freedom robotic gripper capable of both grasping and plunging motions. The gripper is driven by a single actuator and exploits mechanical underactuation to distribute motion and force between the two degrees of freedom. The mechanical design was developed through CAD-based modeling, followed by the formulation of a multibody dynamic model capturing the coupled behavior of the system. Based on this model, a switching velocity/force control scheme was implemented to regulate the interaction forces during the grasping phase, and two configurations were compared (a step and a ramp grasping-force reference). Simulation results demonstrate the capability of the gripper to achieve stable and adaptive manipulation while maintaining controlled contact force. The proposed approach highlights the benefits of underactuation for compact and versatile robotic end-effectors.

ABSTRACT. In this paper, the problem of designing a fault and collision detection and isolation monitor for robot manipulators is addressed using position measurements only. Instead of resorting to velocity obtained by numerical differentiation, as in the classical setup of momentum-based residuals, we propose here an approximate residual which is fed by velocity estimates provided by a reduced-order observer. The efficiency of the method in terms of noise attenuation is illustrated by means of numerical simulations and of experiments on the KUKA LWR 4+ robot.

ABSTRACT. In this paper, we present a modular and reconfigurable soft robotic gripper that incorporates 3D-printed flexible compliant fingers based on the FinRay Effect. The fingers are attached and detached from the gripper base using a slide lock and a pull lever mechanism to provide comfortable switching between two-finger and three-finger configurations, while the fingers are actuated through a set of linear actuators. A lumped-parameter mathematical model is formulated for both gripping configurations to study the grasp properties, where each finger is modeled as a monodimensional lumped-parameter kinematic chain whose parameters are coherent with the ones obtained using Finite Element Method simulations. Finally, we present a prototype for the proposed gripper and characterize each configuration by studying the corresponding payload for both configurations. The grasping capability of each configuration is tested on a variety of objects. With its modularity and versatility, our gripper could serve a wide variety of industries that involve handling deformable objects of different shapes and sizes.

ABSTRACT. Stroke is a leading cause of long-term upper-limb disability, often requiring intensive rehabilitation to restore motor function. While robotic rehabilitation and Functional Electrical Stimulation (FES) have individually demonstrated promising results, existing hybrid systems are predominantly unilateral or provide coordinated assistance only to a lim- ited number of joints. This work presents a hybrid rehabilitation sys- tem combining bilateral robotic Mirror Therapy with contralaterally controlled EMG-driven FES (CCFES), coordinated through a finite- state machine that synchronizes robotic and neuromuscular assistance according to the movement phase. The system was validated on eight healthy participants performing five rehabilitation tasks under two lev- els of robotic assistance. Results showed successful task completion while preserving movement patterns comparable to the unassisted condition. Compared with the unassisted condition, task execution time increased by less than 1 s, mean movement speed decreased by less than 0.04 m/s, and bilateral coordination remained accurate throughout all tasks. Par- ticipants rated the system as intuitive and easy to use, supporting its potential for future post-stroke rehabilitation.

ABSTRACT. This work presents a robotic system for die reorientation developed for the DRIMS Challenge. Given a randomly placed die and a target face, the system combines RGB-based localization and top-face recognition, tactile grasp verification, discrete reasoning over the 24 valid die orientations, and recovery-aware motion planning coordinated by a finite-state machine. Additionally, when a single observation is insufficient to determine the complete orientation, the robot performs a known quarter-turn and combines the two face observations with the commanded rotation to resolve the ambiguity. Implemented on a UR5e manipulator equipped with a Robotiq Hand-E gripper, a fixed OAK camera, and fingertip tactile sensors, the system completed six full reorientation tasks during the 10-minute final challenge.

ABSTRACT. Wearable assistive devices may help preserve upper-limb function in people with neuromuscular impairment, but their effects cannot be captured by muscle activation alone. In this study we investigated muscular effort and multi-muscle coordination during passive shoulder assistance in 12 individuals with flail-arm amyotrophic lateral sclerosis, using five healthy controls as condition-matched references. Coordination was represented through electromyographic covariance matrices and quantified using Riemannian distance. Assistance produced task-dependent responses: coordination approached the healthy reference during arm lifting but diverged during drinking, while RMS measures indicated limited unloading and task-specific activation redistribution. These findings highlight the complementary value of muscle activation and coordination measures when evaluating upper-limb exoskeleton assistance

ABSTRACT. Touch is a fundamental component of human interaction, conveying affective and functional information through complex temporal and spatial patterns. Enabling robots to perceive and interpret these patterns requires tactile sensing hardware that captures contact dynamics and computational methods that extract meaningful information from these signals. In this paper, we present an ongoing work on integrating a Fiber Bragg Grating (FBG) electronic skin onto the ICubHead robot, an actuated iCub head mounted on a stationary 3D-printed upper body. The e-skin consists of 20 FBG sensors distributed over the robot's left arm and integrated with its control framework to provide real-time tactile information. To validate our system, we introduced an experimental protocol in which the robot guided participants to interact with it through different types of touch. This work provides a first step toward using neuro-inspired tactile sensing to study touch dynamics in human-robot interaction.

ABSTRACT. This paper presents the Safe Rescheduled Prioritized Planning (SRPP), a framework that augments Prioritized Planning (PP) for Multi-Agent Path Finding (MAPF) using safe start handling, subgoals reassignment, and deterministic priority promotion to increase success rate while maintaining computational efficiency. SRPP operates in continuous time using the Safe Interval Path Planning (SIPP) as a low-level planner, and does not rely on structural assumptions on start and goal locations. While safe start handling and priority promotion have been previously proposed in the literature, to the best of our knowledge, SRPP is the first framework to formally define subgoals, and integrate these components for MAPF on continuous-time graphs. Extensive simulations on large roadmaps with up to 250 agents show good solution quality, high success rate, and millisecond-scale runtimes, while highlighting how the new elements improve PP performance.

ABSTRACT. Simulators for social robot navigation commonly approximate pedestrians as circular discs, overlooking the al- ternating leg motion observed by ankle-height LiDAR and the shoe protrusions below its scan plane. We present LegNav, a JAX-based simulator that separates pedestrian sensing geometry from collision geometry. Its Non-Slip Gait (NSG) model anchors stance feet in world coordinates and advances swing feet toward targets adapted to body motion. LiDAR returns are generated from individual leg discs, while collisions are checked against shoe footprints. Using the same navigation policy and training conditions, replacing disc-based pedestrians with NSG increases success from 73.1% to 95.1% and reduces active collisions from 24.5% to 3.3% in unseen leg-aware scenarios. Zero-shot TurtleBot 4 deployment provides qualitative evidence of transfer.

ABSTRACT. Abstract—Robotic devices can serve as objective assessment tools providing quantitative data on upper-limb movement and interaction forces. This exploratory study investigated whether robot-derived kinematic and kinetic measures characterize task- and limb-related differences in children with Cerebral Palsy (CP). Fifteen children (5 with Typical Development, TD; 5 with unilateral CP, UCP; 5 with bilateral CP, BCP) performed a discrete reaching task (Clock) and two continuous tracking tasks (a Figure-8 and a Letter P trajectory) using a 2-DOF planar robot. During Figure-8 tracking with the non-dominant/more-affected limb, BCP showed lower normalized velocity than TD and UCP (0.80 ± 0.11 vs 1.13 ± 0.20 and 1.17 ± 0.16) and longer task duration (28.72s ± 2.43s vs 19.98s ± 2.57s and 19.96s ± 3.03s). No significant differences emerged during the Letter P task. The discrete reaching task highlighted significant group-by-limb interactions for outward work and power. Robot-derived measures can capture task-dependent variations in upper-limb performance, suggesting their potential as complementary quantitative outcomes for clinical assessment in CP.

ABSTRACT. Electrostatic zipping actuators are a class of variable capacitance actuators based on a multi-layer dielectric polymer-liquid structure, which use a voltage-induced zipping mechanism to produce linear actuation. While zipping actuators are a promising technology for soft robotic systems, abrupt zipping transitions can prevent smooth and continuous position control. This presentation investigates the electro-mechanical stability of a class of zipping actuators called electrostatic bellow muscles (EBMs). While theoretical model-based predictions suggest that EBM units are stable over their entire working range, experiments reveal local pull-in instabilities, possibly triggered by deviations from the expected zipping kinematics and influenced by the manufacturing process. Multi-unit stacks exhibit a smoother response over a broader actuation range, indicating that stacking can mitigate abrupt unit-level transitions and improve controllability.

ABSTRACT. Traditional assessments typically evaluate cognitive and sensorimotor functions separately, limiting the characterization of their interaction. This pilot study evaluated the feasibility of an upper-limb robotic protocol for assessing cognitive-sensorimotor integration in 11 unimpaired participants. Cognitive outcomes during robotic tasks were compared with those from corresponding non-robotic cognitive-only tests, while kinematic measures were used to characterize motor performance during cognitive tasks. All participants completed the protocol. Most cognitive outcomes did not differ significantly between modalities, although differences emerged in Information Processing Speed and Cognitive Flexibility Part A. Kinematic measures revealed task-dependent changes in motor execution. These preliminary findings support the feasibility of simultaneous cognitive and motor assessment, but further studies are needed to establish the protocol’s validity and clinical applicability.

ABSTRACT. Body-Machine Interfaces (BoMIs) enable people with sensorimotor impairments to control external technologies by mapping a redundant, high-dimensional set of residual signals onto a low-dimensional control space, which is then translated into device commands. Body signals are typically tracked and mapped onto this low-dimensional space using linear (e.g., PCA) or nonlinear (e.g., autoencoder) approaches. While usually developed and validated in simulation, their real-world deployment introduces additional constraints related to mobility, safety, manipulation and human-robot interaction. This work presents the transfer of a camera-based BoMI framework to the control of the physical humanoid robot Reachy2 and describes the adaptations required to transfer the interface to a real-world environment. The system further implements a traded-control paradigm: the user's body movements, mapped through the BoMI, drive both robot navigation and cursor-based object selection; control is then transferred to the robot for autonomous reaching, grasping, and object placement. The resulting framework extends BoMI-based control from simulated navigation to real-world assistive manipulation while preserving a unified interface. An evaluation protocol has been defined, with data collection with unimpaired participants scheduled to begin.

ABSTRACT. Reinforcement learning for dexterous manipulation depends on how well the physics-based simulation represents the actual hardware, especially for underactuated tendon-driven hands where motor dynamics and contact conditions can change the effective joint coupling. In this work, we perform hardware-informed system identification of the RH8D hand to improve its MuJoCo simulation for future sim-to-real reinforcement learning. The actuator behavior was modeled by a delayed asymmetric rate limiter identified from hardware measurements. With a constant tendon-transmission model calibrated on index-finger free-space motion, full dynamic replay reaches a combined metacarpophalangeal (MCP) and proximal interphalangeal (PIP) joint trajectory root-mean-square error (RMSE) of 5.085 degrees. It was then tested whether the same constant-transmission model can reproduce finger behavior during contact with a fixed cylindrical object. The measured hardware pose was 14.54, 23.81, and 9.33 degrees for the MCP, PIP, and distal interphalangeal (DIP) joints, whereas the simulated model predicted 13.47, 2.25, and 10.03 degrees, respectively. Further testing of different constant-transmission parameters showed a clear trade-off between free-space and contact accuracy, as improving the contact behavior worsened the free space performance. These initial results indicate that a constant-transmission model may not be adequate for representing both free space and contact behaviors, which necessitates further investigation into a configuration-dependent spatial tendon model.

ABSTRACT. Accurate prediction of functional recovery in neurorehabilitation represents an important clinical challenge. Traditional rehabilitation assessments rely on periodic, discrete clinical assessments, often failing to capture continuous, real-world physiological variation and daily strain. Moreover, personalizing neurorehabilitation pathways requires continuous and objective information on a patient's physiological status. This work proposes an artificial intelligence (AI)-based framework to predict individualized rehabilitation trajectories in neurological patients by combining continuous physiological data streams from wearable devices (e.g., WHOOP), including heart rate variability (HRV), resting heart rate, sleep architecture, respiratory rate, strain, and recovery scores, with standardized clinical outcome measures of rehabilitative performance (e.g., Trunk Control Test, Short Physical Performance Battery, Motricity Index). Daily physiological indicators combined with standard clinical scores allow machine learning models (e.g., gradient boosting and recurrent architectures) to predict recovery trajectories, stratify patients by expected responsiveness to treatment, and support physicians in early, data-driven adjustment of the rehabilitation plan. Preliminary considerations on feasibility, data integration strategy, and potential clinical impact are discussed, as well as current limitations concerning sensor accuracy and generalizability in clinical populations.

ABSTRACT. Model-based planning and control of robotic manipulators rely on the repeated evaluation of kinematic and dynamic quantities. Deriving these terms, compiling them into efficient code, and keeping them synchronized with the robot description is a recurring source of friction in both research and applied robotics. In this extended abstract, we present a plugin for Thunder Dynamics that generates a \emph{Robot Server}: a ROS~2 package that keeps the model quantities synchronized with the joint state, and exposes them as services and topics. We discuss the trade-offs of this architecture with respect to linking the generated library directly into each node, and we show that the topic interface is fast enough to be used inside a \SI{1}{\kilo\hertz} control loop by running an adaptive computed torque controller on a 7-DoF Franka Emika manipulator.

ABSTRACT. Electroadhesion suction cups (EASCs) are fully electrical grippers (no air flow needed) with very low power consumption that can grasp flat to curved objects from the top. They conform to the shape of the object by zipping from the central contact point to their edges, driven by Electroadhesion forces. Zipping requires deforming elastically the EASC membrane. The object surface curvature at contact point strongly affects zipping ability, and therefore grasp feasibility. We developed a model for grasping point selection that predicts the voltage required for full zipping on a point of given local curvature. Feasible points are the ones where the estimated zipping voltage is lower than the breakdown voltage of the EASC. The model is based on an energy balance between electrostatic work and elastic deformation, explicitly including in-plane stretching on doubly curved surfaces. Experiments on cylinders, spheres, and ellipsoids validate the predicted thresholds and curvature-dependent trends.

ABSTRACT. Collaborative robots operating in shared human-robot environments must generate motion that is simultaneously natural, predictable, and formally safe. While human-inspired motion planning has been shown to improve the acceptance and legibility of robot motion, it does not inherently provide formal guarantees on velocity bounds in the presence of nearby humans. Conversely, safety frameworks that enforce velocity constraints at runtime are typically agnostic to the qualitative properties of the planned motion. Bridging this gap is a fundamental requirement for the deployment of trustworthy robotic systems in real-world human-robot coexistence scenarios. In this paper, we present an integrated framework that couples a human-like motion planner, grounded in the analysis of human upper limb movements through functional Principal Component Analysis (fPCA), with the Safe Motion Unit (SMU), which leverages biomechanical injury data to enforce proximity-dependent velocity constraints at runtime. We validate the proposed framework through a preliminary experimental evaluation on a real collaborative robot, using a vision-based human skeleton detection system to estimate the human-robot distance in real time.

ABSTRACT. This extended abstract presents TriHSA, a soft bio-inspired robotic wrist built around three Handed Shearing Auxetic (HSA) cylinders. The architecture targets coordinated flexion/extension (F/E), radial/ulnar deviation (R/U), and pronation/supination (P/S) while retaining passive compliance and substantial load-bearing capability. A bidirectional HSA geometry and a differential/clutch-inspired drive-to-motion mapping are combined in a compact three-cylinder arrangement. Experiments on a dedicated test platform show approximately ±33.5° range of motion in F/E and R/U and ±23.6° in P/S. Under trajectory-referenced dynamic loading, the wrist sustains about 5.0 N·m in F/E and 4.9 N·m in P/S, while passive rotational stiffness reaches approximately 100 N·m/rad and 115 N·m/rad, respectively. Coupled Dart-Thrower’s Motion (DTM) remains repeatable under distal payloads up to 300 g.

ABSTRACT. Reliable control and interaction of soft robots require proprioceptive sensing that is accurate, durable and easy to integrate. This abstract presents the results obtained with multiple works on the development of a sensorized soft joint for robots for the measurement of bending angles. The developed joint consists of a silicone rubber body embedding two foil strain gauges, supported by a nylon mesh and connected in a half-bridge configuration, together with a 3D-printed thermoplastic polyurethane housing. Simulations and experiments on dedicated test benches show a linear response, a root mean square error of about 1 degree, a drift below 1 degree per 1000 cycles, and stable operation over more than 3500 cycles. The joints were tested in a real application by integrating them in a tendon-driven underactuated robotic hand.

ABSTRACT. Construction sites demand robots that can perform diverse operations with markedly different sensing, workspace, interaction, and tooling needs. Modular reconfigurable robots address this at the hardware level by allowing kinematics and end-effectors to be adapted to each operation, but hardware modularity alone does not yield a task-level abstraction that lets operators specify goals independent of the current configuration. We present an operator-centered interface for the CONCERT modular construction robot that links physical reconfiguration with autonomous task execution. Instead of teleoperating motions, the operator designates the region to be processed and reviews plans produced by an autonomous planner. Overall, the architecture decouples task specification from robot configuration, enabling sensing, planning, navigation, manipulation, and intervention to be composed around a common environment representation.

ABSTRACT. Tracking the 6D pose of a novel object in a robot's hand is within reach once a model of the object exists: trackers such as FoundationPose follow a known mesh through at frame rate, but they need that mesh first. We present the reconstruction stage that can provide it during the grasp itself. Building on our physically grounded hand-conditioned flow model, which reconstructs a grasped object as a camera-aligned signed distance field from an RGB image, the hand geometry and its contacts, we distill a 220M-parameter student six times faster than the 757M teacher and make it recursive: student-recursive receives its own previous reconstruction, warped into the current frame, as a prior, and refines the object over the first frames of a grasp at 0.56 s per frame, without camera extrinsics. Trained only on synthetic robot-hand grasps and evaluated on real human-hand captures, it raises posed IoU from 0.550 to 0.651 on DexYCB and from 0.443 to 0.538 on OakInk within a few seconds of observation. With one or two views it matches the offline multi-view baseline MV-SAM3D on surface accuracy; MV-SAM3D pulls ahead with four or more views at 12 to 14 times the compute, which sets the target for the next scale-up of our training data.

ABSTRACT. Electronic cams are widely used in automated machinery to perform a range of tasks. Compared with traditional mechanical cams, their main advantage is the ability to change the end effector’s stroke and motion profile through software, minimizing machine changeover time. For highly dynamic motions, however, careful design and tuning are needed to limit following errors and maintain adequate motion accuracy. This abstract compares two electronic-cam configurations designed to perform the same task. In the first one, the end effector is linearly coupled to a brushless motor; in the second one, a dwell mechanism is placed between the motor and the end effector. Although the dwell mechanism reduces the flexibility to change the end-effector stroke, it enables higher performance with a lower position error. The paper examines the benefits of placing a nonlinear mechanism between the motor and the end effector and seeks a practical criterion for deciding when this solution is recommended.

ABSTRACT. This abstract presents the preliminary design of a four-wheeled mobile robot for cutting operations inside ships during dismantling. The platform is intended to operate in an unstructured environment characterized by uneven and potentially wet surfaces, slopes, steps, obstacles, and confined spaces. A skid-steering and a double-wishbone suspension system are adopted to provide a compact architecture and accommodate terrain irregularities. The abstract outlines the main design requirements and the resulting platform configuration, with particular emphasis on the analytical modelling of the suspension.

ABSTRACT. Planning and control of quadrupedal robots is challenging because of their hybrid dynamics, inaccurate models, and unknown terrains. We present a pipeline that plans the base trajectory and the contact sequence simultaneously, without a predefined gait, using a planar single rigid body model in which a sigmoid function merges the contact phases into a single continuous dynamics. The planned motion is mapped to joint references, and an Iterative Learning Control (ILC) scheme tracks them, compensating for model mismatch and disturbances in a learning-while-doing fashion. Experiments on the quadruped Solo-12 show that the planner produces pronking and bounding gaits, and that ILC reduces the joint tracking error across terrains and disturbances.

ABSTRACT. This work presents the design, development, and experimental validation of a novel soft–rigid robotic finger with two granular-jamming-based joints. Granular jamming lets each joint switch from a soft, compliant state to a rigid one under negative pressure, mimicking co-contraction in human muscles. Among the tested granular media, dodecahedral grains of 3 mm were selected for their wider range of achievable stiffness. The finger consists of rigid phalanges interconnected by variable stiffness joints and actuated by tendons. Three experiments evaluated how joint jamming affects flexion trajectories and tendon force, grasp tightness and stability, and the ability to reach different pre-grasp configurations that significantly influence grasp efficiency. Results show that granular jamming is an effective strategy for local stiffness modulation in soft–rigid fingers, enhancing their versatility in handling diverse objects.

ABSTRACT. Automated visual inspection of low-contrast surface defects on metal parts requires viewpoints that satisfy tight visibility conditions, incidence angle, field of view (FOV), and depth of field (DOF), rather than mere geometric coverage. We propose an inspection planning pipeline that reduces the cost of viewpoint generation and visibility evaluation. The part is sliced into sections whose contours are reconstructed by a stochastic convex-hull carving algorithm with Fourier smoothing. Candidate viewpoints are restricted a priori to a "crown" derived from the DOF constraint: a strict crown augmented with local lobes, which guarantees full boundary coverage and is extended to 3D and to the incidence constraint. Visibility is evaluated with a three-state rule computed from each face's centroid with a margin band proportional to face size, which avoids sub-face sampling. From the candidate points, those optimal for the inspection procedure are selected by solving an optimization problem that considers both the visibility conditions and the reachability for the manipulator performing the task; a trajectory visiting them is then planned and executed by a six-axis arm with a rotary platform carrying the part. Contour reconstruction, crown, and visibility rule are verified on a real mechanical part, whereas a full industrial dataset remains to be tested; pose selection and trajectory planning are under testing.

ABSTRACT. Patients with lower limb amputation often experience gait disorders due to the loss of sensory information. While sensory feedback restoration has shown promising results, conveying terrain-related information remains unexplored. This study focuses on Transcutaneous Electrical Nerve Stimulation (TENS) as a non-invasive stimulation technique for eliciting somatotopic sensations in lower limb. The aim of the study is to propose Sinusoidal Pulse Width Modulation (SPWM) as a novel encoding strategy for terrain recognition. SPWM was evaluated in ten healthy participants in discriminating among three different terrains. SPWM significantly outperformed literature approach Linear Pulse Amplitude Modulation (LPAM) in terms of accuracy (85.1% rather than 68.2%, P=0.0039) demonstrating, for the first time, the feasibility of conveying terrain-related information through non-invasive somatotopic tactile stimulation.

ABSTRACT. This work introduces a real-time compatible segmentation algorithm for elbow movements based on wavelet decomposition. The algorithm relies purely on the reading of the elbow joint angle, and it does not require an a-priori knowledge of the movement. Preliminary results on eight unimpaired participants executing activities of daily living show comparable results with respect to a non-causal benchmark algorithm. The algorithm as proposed represents a viable solution for real-time upper-limb movement segmentation, that could be integrated in the intention decoding algorithms of wearable robots’ controllers.

ABSTRACT. This paper proposes a bidirectional robotic interpretation pipeline to support communication between hearing and Deaf sign-language users. The pipeline connects speech recognition to robotic signing in one direction and sign recognition to spoken output in the other. A limited-vocabulary prototype was developed using local speech recognition, webcam-based hand-landmark extraction and long short-term memory classification for American Sign Language recognition, and two simulated five-finger robotic hands in MuJoCo platform. Recognized spoken words triggered predefined hand movements, while recognized signs were converted to spoken output. Initial prototype testing demonstrated speech triggered simulated robotic hand movements for YES, NO, and MEET. It also showed that recognized signs were converted into speech for individual words.

ABSTRACT. Fine monitoring of physical interactions with the environment is essential in upper-limb prostheses since it enables meaningful sensory feedback and supports safe grasping, particularly when interacting with potentially harmful objects. This work presents the design and preliminary validation of a low-cost multi-cell force-sensitive sensor for prosthetic fingertips, developed to provide distributed mechanical information for pain-related encoding strategies. The sensor is implemented on a flexible PCB and comprises independent sensing cells. Benchtop validation was performed using an Instron testing machine with controlled static loads applied to each cell. The sensor showed an average release time of 19 ms and low repeatability errors, despite cell-dependent differences in sensitivity. The device was then integrated on the index fingertip of an IH2 Azzurra prosthetic hand and tested during bidigital grasping of objects with different sizes and geometries. The normalized activation profiles showed distinct spatial patterns across the five cells. Objects with small contact areas tended to activate a single sensing region more strongly, whereas larger contact areas produced lower distributed multi-cell responses. These preliminary results support the sensor as a sensing interface for future pain-encoding in upper-limb prostheses.

ABSTRACT. Ship recycling involves hazardous cutting operations that are still largely performed manually, exposing workers to unsafe environments and demanding working conditions. To address these challenges, the Horizon Europe SHEREC project is developing robotic technologies for automated ship dismantling. This abstract focuses on a mobile manipulator developed within the project, consisting of a suspended mobile platform and a redundant 7-DoF robotic arm for plasma-cutting operations inside decommissioned vessels. Manipulator motion modifies the overall CoG position of the robot-platform system, potentially causing suspension deflections and variations in the platform pose. These variations may affect the torch-to-workpiece stand-off distance, which is a critical parameter for plasma-cutting quality and process stability. A MATLAB-based optimization framework is therefore proposed to exploit the kinematic redundancy of the manipulator and minimize both incremental CoG variations and deviations from its initial position, while preserving the prescribed end-effector trajectory and satisfying the relevant process constraints.

ABSTRACT. Wearable soft robots need sensors that are compliant, lightweight, low-cost and energy efficient. The Sens4SoftRob subproject of Fit4MedRob develops such sensing technologies and validates them on the Robotic Sixth Finger (RSF), a supernumerary finger that assists users with impaired hand function. This paper summarizes the results of the last reporting period. Proprioceptive soft joints with embedded strain gauges are integrated in the RSF and, together with an accelerometer, feed a 1D convolutional network that classifies the stiffness of grasped objects: after int8 quantization it reaches 88.7 percent accuracy and runs on a microcontroller in 567 ms using 36.8 kB of RAM. A modular near-infrared optical myography armband with 64 optical channels monitors the wearer and decodes four hand gestures with 92.5 percent accuracy. A soft capacitive sensor detects the onset of slip at the finger–object interface. Finally, a gaze-based interface combining object detection, fiducial tracking and monocular depth estimation triggers the finger only when the user shows the intention to grasp.

ABSTRACT. Continuous and accurate monitoring of elbow kinematics is crucial for a broad range of applications, including neuromotor rehabilitation, sports biomechanics, and the control of assistive upper-limb exoskeletons. Wearable inertial measurement units (IMUs) can be used to that purpose, although their long-term reliability is limited by signal integration drift. While optical and vision-based systems provide highly accurate tracking, they are fundamentally non-wearable, as they require external equipment. To address these gaps, we propose a wearable sensing technology using pre-strained dielectric elastomer piezo-capacitive sensors (DES). An onboard IMU provides a temporary kinematic reference during an initial guided calibration to extract polynomial regression coefficients. After calibration, the IMU is computationally deactivated, and continuous tracking relies exclusively on the drift-free DES. Validation against a vision-based reference system demonstrated robust sensing accuracy, making this novel technology a viable solution for continuous elbow flexion monitoring.

ABSTRACT. Haptic feedback is essential in virtual reality applications, such as surgical training, teleoperation, rehabilitation, and serious games. Softness perception is crucial in these applications, as they frequently involve interactions with compliant materials. Rendering softness requires both cutaneous and kinaesthetic cues. However, devices able to provide kinaesthetic feedback are typically bulky. While wearable cutaneous-only displays offer lightweight alternatives, the perceptual cost of excluding kinaesthetic feedback remains to be quantified. To address this lack of knowledge, we present ongoing developments to implement a wearable integrated haptic system, combining a fingertip pneumatic tactile display with a Bowden-cable-driven hand exoskeleton. The system is conceived to allow for experimenting with two possible operating modes (cutaneous-only feedback, and combined cutaneous and kinaesthetic feedback), to enable comparative investigations through psychophysical testing.

ABSTRACT. Soft-rigid grippers rely on compliant joints whose stiffness determines the finger motion, yet obtaining a target stiffness from a 3D-printed joint remains difficult. The relation between joint geometry and stiffness is nonlinear, and simulations of compliant parts often deviate from the behavior of the fabricated prototypes. This paper addresses the inverse design of the WaveJoint, a wave-shaped compliant joint described by five geometric parameters. Rather than learning a full mapping from stiffness to geometry, we cast the design as a black-box global optimization problem and solve it with a Radial Basis Function (RBF) surrogate method. The method natively handles bound and integer constraints on the geometry. It also combines two sources of information: Finite Element simulations, treated as cheap but noisy evaluations, and measurements on 3D-printed prototypes, treated as expensive but exact evaluations. We validate the approach on a design task with stiffness targets derived from a gripper application, starting with no prior data. After eight printed prototypes, the measured stiffness of the joint matches the target with a residual of 0.13. A simulation-only search on the same task yields a residual of 0.46 after 170 simulations. These results show that a few physical measurements, integrated into the surrogate, effectively compensate for the bias of the simulator and reduce the number of design iterations.

ABSTRACT. Manipulators that share their workspace with people must avoid obstacles that are perceived incompletely, with noise and with delay. We present a pipeline from a single fixed RGB-D camera to the joint torques of a 7-DoF Franka Research 3 arm. A depth-space module computes, for each control point on the arm, the distance between the surface of the arm around it and the nearest obstacle pixel, and a Kalman filter estimates the obstacle velocity and its covariance from the same pixels. An acceleration-level control barrier function (CBF) safety filter uses this information to anticipate obstacle motion and to tighten the barrier with bounded margins for latency and estimation uncertainty, and combines obstacle, self-collision, joint-limit and singularity constraints in one quadratic program. In experiments on the Franka Research 3 arm in the DIAG laboratory, the estimated distance never fell below the 15 cm safety distance with static obstacles. With moving obstacles it stayed above the safety distance for 95.0 % of the time while a person moved a hand towards the arm, and for 94.3 % of the time while they moved a rod. To avoid them, the arm left its path by up to 0.95 m, and no collision reflex was triggered. Violations of the safety distance occurred mainly when the obstacle approached faster than the arm could react, given the perception latency, the update rates of the pipeline and the joint limits, and partly because inaccurate depth at the edges of the arm made parts of the robot appear as obstacles.

ABSTRACT. Mixed reality can support multi-robot supervision by presenting robot state, maps, sensing, and commands within a common spatial workspace. However, many existing interfaces remain tailored to a particular robot platform or task. This paper presents HORUS (Holistic Operational Reality for Unified Systems), an open-source, registration-driven mixed-reality operations platform for heterogeneous robot teams. HORUS combines a Meta Quest~3/3S runtime, a Python SDK, and a ROS 2 bridge to generate capability-dependent robot representations, shared-map visualizations, task interfaces, teleoperation controls, and multi-operator workspaces. Per-robot leases provide explicit command ownership when several operators access the same team. We describe the system architecture, robot capability model, mixed-reality workflow, and local runtime data paths. The platform is characterized under increasing camera, spatial-data, robot, and operator loads, and is demonstrated in a heterogeneous aerial--ground exploration deployment using H-CoRE; an open-source, ROS 2-based architecture designed to enable autonomous multi-robot operations in GNSS-denied environments.

ABSTRACT. Robot-assisted laparoscopic surgery requires accurate maintenance of the Remote Center of Motion (RCM) constraint imposed by the trocar, whose violation may generate harmful lateral forces at the abdominal wall. Existing software-based strategies either assume a static RCM, neglecting respiratory motion, or lack systematic validation under repeatable trocar mobility. This work addresses this gap through two contributions: a comparison of representative Inverse-Dynamics (ID) and Inverse-Kinematics (IK) based control strategies under fixed and dynamic RCM, and a dedicated experimental setup reproducing respiratory-like tissue displacement around the trocar. The compared strategies are a Cartesian admittance controller with null-space RCM enforcement and a force-based velocity controller that reduces trocar forces through adaptive pivoting, both evaluated on predefined linear trajectories using a eobotic arm and a laparoscopic simulator integrated with the proposed setup. Under dynamic RCM, admittance control reduces peak lateral forces (max ∥Fxy ∥: 1.16 ±0.24 vs. 1.40 ±0.23 N, p < 0.05) and peak torque loads (max |Tx|: 0.080 ±0.024 vs. 0.104 ±0.025 N·m, p < 0.05), while the velocity controller shows a lower force accumulation trend over time (slope−0.006 ±0.016 vs. 0.015 ±0.026 N/s, p < 0.05). Nevertheless, both strategies reach peak forces exceeding 1 N, indicating that neither achieves sustained trocar-force reduction under dynamic RCM and motivating approaches combining dynamic modelling with active force regulation.

ABSTRACT. Long-duration outdoor coverage with autonomous platforms remains challenging beyond classical planning: deployments face localization drift in open spaces, obstacles in cluttered sites, controller feasibility in turn-heavy maneuvers, and persistent autonomy with energy management. We propose a unified ROS 2 architecture for single-robot outdoor coverage: a dual-antenna RTK-GNSS fused in an EKF keeps both position and heading accurate across long missions; three controller-aware refinements extend a mature coverage planner; a Nav2-based behavior-tree mission executive coordinates multi-goal execution, layered recovery, cost-aware goal management, and autonomous docking for return-to-charge. In real-world trials across five outdoor areas with varying geometries and obstacle densities, the robot completed every coverage route, sweeping 93.1% to 96.1% of the planned coverage area.

ABSTRACT. In robot-assisted electromagnetic actuation systems for magnetic capsule positioning, capsule motion is typically generated through magnetic forces while the robotic manipulator continuously follows the capsule to maintain a favourable magnetic configuration. This robot-following paradigm can be dynamically inefficient because it underutilizes the reactive control capabilities of the electromagnetic subsystem and imposes unnecessary motion demands on the robot. This paper addresses this limitation by proposing a Robot–Assisted Electromagnetic Actuation Platform for three-dimensional magnetic capsule positioning. The proposed system combines a compact four-solenoid electromagnetic actuator with a 7-DoF robotic manipulator, resulting in an 11-DoF redundant actuation architecture that enables capsule motion to be generated through both electromagnetic force modulation within a local magnetic workspace and robot-driven repositioning of that workspace throughout the global environment. Capsule motion is expressed through a unified robot--magnetic Jacobian, mapping joint velocities and coil current rates to capsule velocity in a single redundant formulation. Building on this formulation, a Jacobian-Based Robot–Magnetic Actuation Control (RMAC) strategy is proposed to resolve the actuation redundancy and allocate the positioning task according to the dynamic characteristics and actuation costs of the two subsystems. As a result, fast local capsule corrections are primarily assigned to the electromagnetic actuator, while the robot performs slower workspace recentering motions. RMAC was evaluated in simulation against a robot-following baseline over planar, box-shaped and helical trajectories. Results show that, across five weighting configurations of (Wr , Wm), RMAC preserved submillimetric capsule positioning accuracy (mean error 0.13–0.15 mm, versus 0.33 mm for the robot-following baseline), while reducing mean robot velocity from 5.19 to 2.30–3.38 mm/s, acceleration from 3.33 to 1.48–2.14 mm/s2 and jerk from 38.6 to 5.71–9.75 mm/s3, without increasing in coil current demand. These results show that robot–magnetic redundancy can be exploited to reduce the dynamic burden on the robotic subsystem while preserving accurate capsule positioning.

ABSTRACT. Constructibility is a structural property to decide whether the final state of a dynamic system can be reconstructed from a set of measurements. This property is particularly challenging to verify when the system dynamics and/or the output function is nonlinear. In this paper, we study constructibility for a group of nonholonomic vehicles that can measure their relative distance when they are in sight of each other. The vehicles are also equipped with communication devices to share information on their states. We first show some necessary conditions that each vehicle has to meet on the number of collected measurements and on the shape of its trajectory in order for the whole state to be constructible. Then, we analyse weak constructibility properties of the entire multiagent system by means of the Constructibility Gramian, and discuss some geometrical insights that explain the analytical results.