Kok-Meng Lee

This article presents a physics-based, energy-guided framework for reconstructing the 3D shape of flexible beams and estimating applied loads directly from point cloud data. Unlike conventional geometric fitting methods that are sensitive to noise and lack physical interpretability, the proposed approach formulates shape reconstruction as the minimization of a pseudo-elastic energy functional combining data fidelity with beam mechanics. Point cloud measurements are coupled to the beam centerline via virtual springs, while smoothness is enforced through elastic strain energy. The framework naturally accommodates noisy, unorganized visual data and incorporates anisotropic bending via curvature-based constraints. The reconstructed shape is then used to estimate tip loads and distributed contact forces through nonlinear least-squares optimization, enabled by an efficient coordinate-transformed initial-value beam solver. Simulations demonstrate accurate shape reconstruction and robust load estimation under measurement noise and partial constraints. Experimental validation using markerless, smartphone-based 3D scanning further confirms the ability of the proposed framework to recover physically consistent external loads without embedded sensors. Results establish a unified vision–mechanics approach for noncontact deformation and load estimation in flexible beams and continuum structures.

Motivated by the significant disparity in the aspect ratio of cavity and crack defects, this paper proposes a multi-directional scanning-based eddy current (EC) testing strategy for distinguishing between these two types by analyzing the characteristics of the perturbed EC and the corresponding magnetic flux density fields. To enable efficient simulation and verification of the strategy, the distributed current source method is enhanced by incorporating solution domain rotation, allowing accurate modeling of defects at arbitrary orientations. A symmetrical relationship among scanning images is also discussed to reduce the number of required simulations. The validity of the improved numerical model is demonstrated through comparing with finite element analysis results. Simulated scanning images for various defect orientations are then used to identify a key parameter that enables reliable classification. Finally, the proposed strategy is experimentally validated using a testbed mounted on a servo-positioning platform, confirming its effectiveness even when applied to workpiece materials different from those used in the simulations.

Lightweight impedance-controllable end-effectors are increasingly important in emerging applications involving physical human-robot interaction. Motivated by this need, this paper presents a method to design a magnetic lead screw (Mag-LS) with a built-in wireless sensing system that utilizes its inherent magnetic field for simultaneous displacement and force measurements. Based on the field/force models derived in closed form, this paper analyzes the uniqueness of the inverse solutions for designing the magnetic sensing system and optimizing its parameters to minimize computational cost for implementation on a standalone microcontroller where sensor noises are filtered. Unmodeled geometrical factors are accounted for using physics-based artificial neural networks (ANNs). A prototype Mag-LS with embedded sensors and testbed have been developed, on which noise and parametric effects on sensing accuracy are experimentally investigated. The results demonstrate the effectiveness of a multi-single-output ANN for simultaneously measuring the displacements/force of a Mag-LS, offering an alternative collocated force and displacement sensing solution for impedance/force control.

This paper explores the often-overlooked importance of the palm in robotic hand design, where traditional approaches emphasize finger dexterity and multi-motor manipulation to achieve diverse degrees of freedom (DOF) for grasping. The palm’s ability to conform to objects while grasping is an additive unique feature due to its versatility to create different shapes. However, replicating this complex contouring of the human palm in a mechanical equivalent presents significant challenges. To address this, the paper introduces the concept of a compliant anatomical palmar mechanism (CAPM), using a combined experimental and numerical approach to model the kinematic characteristics of a hybrid rigid/compliant system. The study analyzes how palm movement can enhance grasping performance capabilities. The studies use human hand experiments and finite element analysis (FEA) simulations to predict grasping force distribution. This improved modeling allows for more effective robotic design by utilizing palm functions to achieve a broader force distribution in power grasps. A CAPM prototype, integrated into a robotic hand testbed, has been developed to validate the concept and evaluate its effectiveness in grasping by maintaining a stable grasp of a spherical object under an increasing torsional load.

This paper presents the design concept of a soft continuous permanent magnet (C-PM) and magnetic elasticity for the design and control of a magnetic series elastic actuator (Mag-SEA). Unlike a conventional SEA typically designed by integrating an elastic element into an actuator through a mechanical leadscrew (Mech-LS), there is no physical contact between the rotor and translator of a Mag-SEA where their “gap” plays the role of a noncontact “magnetic spring” as the rotor magnetically drives the translator like a LS through helical PMs. Using a distributed current source method verified with analytical solutions and finite element analyses, the magnetic force vector of a Mag-SEA built with C-PM is derived and evaluated by comparing it with that designed with cylindrical PMs to approximate the helical PMs. With the magnetic elasticity computed using the DCS method and a control-oriented perturbation model, the parametric effects on the dynamics and stability are investigated numerically to gain insights into the plant characteristics of a Mag-SEA.

This article presents a 3-DOF planar magnetic pantographic exoskeleton (Mag-PGE) with embedded sensors for manipulating/measuring the motion and force/torque (F/T) of an ankle joint in the sagittal plane. A biomimetic ankle-foot simulator (AFS) is designed to facilitate investigation and help visualize its potential uses for in-bed acute stroke rehabilitation where the affected leg muscles are often passive. Unlike a conventional motor in which the rotor is supported by mechanical bearings on the stator, the (stator and rotor) of the Mag-PGE are separately attached to the (shank and foot) sharing the ankle joint of the human leg such that it enables the foot to flex in the sagittal plane, imposing no mechanical constraints on the ankle joint. A Mag-PGE/AFS model with closed-form solutions to the rotor magnetic field and motor F/T is provided; both forward and inverse problems are considered. The methods to design, train, and calibrate embedded sensors utilizing the rotor magnetic field to measure the rotor/foot motion are numerically and experimentally evaluated. As illustrated with an AFS, the model along with the embedded measurements plays a pivotal role in computing the forward and inverse solutions to the 3-DOF motor force-current model.

This article presents the methods to design and model an eddy-current (EC) probe composed of one or more electromagnets (EMs) and a magnetic flux density (MFD) sensor for detecting defects near the edge of a conductive workpiece (WP). Both cavity and crack defects are considered; the former characterized by a conductivity field and the latter modeled as two overlapping boundaries, and their perturbed effects on the EC-generated MFD field are analyzed using a distributed current source (DCS) computing method. Two approaches, defect-free subdomain formulation and dimension-reduction for analyzing defect-perturbation, are presented to shorten the time to compute the DCS solutions to the defect-detection problem. The accuracy and computational efficiency of these proposed approaches have been numerically evaluated, which demonstrate significantly improved performance with greatly reduced computation when compared with finite-element analysis (FEA) and can further shorten the time to compute the matrix inversion. The effectiveness of the magnetic field-based method has been experimentally verified with two prototype EC probes designed to overcome limitations associated with impedance-based EC probes commonly designed using a lumped-parameter approach.

Lumbar force prediction has been critical to developing cost-effective wearable lumbar exoskeletons to reduce/prevent low back pain (LBP). Noninvasive prediction of the internal compressive forces on the lumbar spine remains challenging. Considering a spine-equivalent beam (SEB) model for the musculoskeletal system, this article presents a two-stage method to identify a set of subject-specific parameters when the upper torso flexes in the sagittal plane. The first stage uses the measured subject-specific spine curvatures due to bending loads to identify the SEB’s flexural rigidity (EI) and erector spinal (ES) muscle force/torque. The second stage searches for the optimal EI minimizing a residual vector based on the difference between the spine shape simulated by the SEB model and that reconstructed from measurements. The results are validated with published in vivo data, considering both with and without lifting a load. The findings reveal that the musculoskeletal structure of a human spine, unlike an engineering beam where EI can be characterized by a well-defined constant for a given material/design, responds to the load acting on it and its parameters depend on the load and its lifting configurations.

Motivated by the need to incorporate human spine mechanics into the design loop of a wearable spine-assistive device to prevent low back pain, this paper presents a novel method to model the spine as a series of pin-connected triangles and reconstruct its shape/curvature in the sagittal plane from the measured back profile along the pathlength between the neck and the pelvis by template matching without relying on the common assumption that the back profile and spine are parallel or medical images to calibrate the subject-specific parameters. The findings verify that the vertebrae of human spines are similar and can be scaled using an anatomically based spine as a template. Formulated as a constrained optimization problem to account for the deformable skin movements and subject-dependent variations (height/weight and subcutaneous fat tissue thickness), the templet-matching method that solves the template-feature positions eliminates common errors caused by tracking the skin markers as the points on the spine curve and is evaluated in two scenarios; both use the same template of an anatomically based spine. The first uses published magnetic resonance imaging data of eight subjects’ standing and flexion postures to validate the method. The second illustrates its effectiveness using images captured by a commercial human-motion capturing system experimentally. As verified with X-ray images that serve as ground truth for comparison, the findings demonstrate that the method offers an effective tool to reveal the internal kinematics of the spine from the measured profile of the human back.

Rotary to translational transmission systems play an important role in many applications from engineering to human assistance devices. Although lead and ball screws are widely available, they suffer mechanical wear/tear problems due to contact friction. Motivated by increasing demands for energy-efficient mechanisms for mobile and wearable robotic systems, this article presents an analytical method to design a magnetic leadscrew (MLS) with embedded sensing. MLS is driven by permanent magnets (PMs) converting magnetic energy to thrust forces while transmitting the rotary-to-translation motion. However, existing designs generally assume an infinitely long MLS, so its magnetic field distribution is axisymmetric and periodic. To relax these assumptions for applications that require maintaining a constant lead over a short travel, this article formulates the magnetic field and radial/thrust forces of an MLS in closed form using a distributed current source (DCS) method for developing MLS with an embedded field-based sensing system. The sensing method determines the unique solution to the inverse magnetic field model and measures the translation and rotation independently. With the DCS models, a parametric study has been conducted numerically leading to the development of a prototype PM-driven MLS with embedded sensing, upon which the magnetic field model, sensing system, and algorithm are numerically illustrated and experimentally validated.

This paper presents a design method based on distributed current source (DCS) that discretizes the permanent magnets (PMs) and electromagnets (EMs) into elemental current sources and derives the magnetic field and current-force models for design analyses of a 3-degree-of-freedom (3-DOF) planar motor with redundant inputs. The DCS models have been verified by comparing them with exact solutions and commercial finite element analysis (FEA). The results show that the DCS models are accurate (within 2.5% of exact solutions) and computationally efficient (a three-order improvement over FEA). As an illustration, the analytically derived DCS models are employed to analyze the geometrical constraints and parametric effects on the PM/EM layout and forces/torque performance of a 3-DOF planar motor. Using singular value decomposition, two designs are numerically evaluated. With the closed-form DCS models, the loci of the best/worst manipulability ellipsoids are graphically presented.

The design of a robotic hand incorporating a critical multi-DOF palm depends on studies of the anatomical structure of the human palm model. Shaping of this large grasping region is simplified to depend on arches formed by relative movement of the phalanges and metacarpal. Compliant Anatomic Palmar Mechanism (CAPM) replaces these arches to form an interconnected compliant structure that can be used in grasping to conform to contacting objects to augment stability. FEM is used to simulate the range of intrinsic movement of the palmar surface. Plotting these results determines the kinematic relationship by relating thumb rotation and metacarpal translation to the deformation of the palmar region.

This article presents a multi-DOF motion sensing system consisting of a permanent magnet (PM) and a magnetic tensor sensor (MTS) comprising a <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$3\times $ </tex-math></inline-formula> 3 array of three-axis digital magnetic flux density (MFD) sensors, and the methods to measure their relative MTS-PM position/orientation in a 3-D space. The design enables redundant differential measurements of MFD vectors and gradient tensor components to account for the singularities due to matrix inversion, providing a basis to explore different methods for multi-DOF estimation of the PM position/pose. Formulated as a two-stage linear least-square (LS) problem to take advantage of the dipole simplicity and exploit its physics revealed by its inverse model to guide the design of a fully connected artificial neural network (ANN) to account for the MTS measurement noise and un-modeled factors, a prototype multi-MTS system capable of 5-DOF motion measurements is developed and evaluated experimentally along with a study analyzing the parametric effects on the estimation accuracy; both stationary and moving sensor scenarios are considered. Enhanced with a fully connected ANN, an accuracy within a root-mean-square error (RMSE) of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$40 ~\mu \text{m}$ </tex-math></inline-formula> spatial position and 0.1° pose can be uniquely obtained without subtracting a predetermined geomagnetic field in both fixed and moving multi-MTS scenarios, representing a significant improvement over the (0.5 mm, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1^{\circ }{)}$ </tex-math></inline-formula> RMSE of the single-MTS.

This article offers an impedance sensing method taking advantage of the conductivity changes due to muscle contraction to estimate muscle-driven joint torques through a convolutional neural network (CNN), where the input images are derived from a finite set of boundary voltage measurements. Guided by a physical model combining the forearm biomechanics and the muscle electric field along with the CNN criteria considering the receptive fields (RFs), the effects of two image formats [for quasi-static (QS) and dynamic (DYN) states] on the CNN performance are experimentally studied on eight human subjects’ forearms using a prototype impedance sensing system. By comparing the CNN-estimated torques with that measured on a haptic device, the findings verify that the impedance-based method can estimate the joint torques driven by both the deep and superficial muscles within 9% errors of the three degrees-of-freedom wrist torque and 10% error of the gripping torque and that it is feasible to share data among a similar group to reduce data collection and time when training a CNN for uses on a new subject.

This article presents a three-degree-of-freedom (3-DOF) magnetic sensor, referred to here as a pantographic exoskeleton (PGE) sensor, for monitoring in real time the internal human-joint motion in the sagittal plane. With two sets of embedded magnetic sensors and a permanent magnet, the PGE wearable on a healthy leg or lower extremity exoskeleton (LEE) independently measures the 2-DOF translations and the joint angle. Two sensor estimation methods, which are the model-based and the artificial neural network (ANN), are experimentally analyzed in the presence of measurement noise. As an illustration, the PGE sensors are evaluated for sit-to-stand (STS) exercises, where the real-time measurements are verified by comparing with the joint angles determined by a commercial VICON motion capture system, and the translational deviations measured on a 4-DOF platform manipulated to follow a specified internal motion trajectory of an ankle joint during STS. With the ANNs appropriately trained to account for measurement noise, the PGE sensors can track the joint angles while measuring the internal motions of both legs with or without the LEE demonstrating that the PGE sensor has the potential to serve as an indicator of stroke rehabilitation where patients lack force perception and suffer an increased risk of falls due to the weak affected leg.

Series elastic actuators (SEAs) are increasingly used in robotic applications where tasks require safe human-robot interaction. Motivated by the growing need to develop SEAs with nonlinear stiffness to tradeoff between force tracking bandwidth and low output impedance, this paper presents an analytical approach to model the magnetic force/torque of a magnetic (Mag) SEA using the distributed current source method to provide a basis to derive the noncontact magnetic compliance; both translational and rotational stiffnesses are considered. The Mag-SEA model, which relaxes several common assumptions in the published literature that treat the magnetic lead screw as an ideal transformer, is numerically illustrated. The findings demonstrate that the simplified (ideal transformer) model fails to characterize the displacement-dependent magnetic force/torque and underestimates the mechanical advantage that can be capitalized from the spring-like magnetic energy of the embedded permanent magnets.

The primary focus of robotic hand design has been placed on integrating a series of rotational joints at each finger. The paper presents the design of a compliant anatomical palmar mechanism (CAPM) that augments the robotic hand's capabilities by adapting between precision and power grasps. A design of this robotic hand has been developed based on the anatomical inspiration of the human hand to find a direct relationship between palmar deformation and the rotation of the fingers at a shared joint. Palmar deformation has an effect on the orientation and position of the fingertip's location when contacting a target object. Previously determined simulation results of this geometrically complex, large displacement, hyper-elastic structure are compared to initial experimental results to show simultaneous varied conformity of the palm.

This paper presents a real-time 3-DOF orientation motion control for a weight-compensated spherical motor with a rotor dynamic model derived using 3D axis-angle representation to facilitate the controller design. By integrating the sensor fusion-based real-time 3-DOF orientation estimation method, a sliding mode control method is developed employing the equivalent axis and angle as the control parameters, which is followed by the investigation of system stability using the Direct Lyapunov stability method. The effectiveness and accuracy of the control system are experimentally demonstrated and evaluated on a permanent-magnet spherical motor prototype fabricated using additive manufacturing with regulation and tracking control experiments.