Analysis of Human Whole-Body Joint Torques During Overhead Work With a Passive Exoskeleton

Overheadwork is classifiedas one of the major risk factors for the onset of shoulder work-related musculoskeletal disorders and muscle fatigue. Upper-limb exoskeletons can be used to assist workers during the execution of industrial overhead tasks to prevent such disorders. Twelve novice participants have been equipped with inertial and force/torque sensors to simultaneously estimate the whole-body kinematics and the joint torques (i.e., internal articular stress) by means of a probabilistic estimator, while performing an overhead task with a pointing tool. An evaluation has been performed to analyze the effect at the whole-body level by considering the conditions of wearing and not-wearing PAEXO, a passive exoskeleton for upper-limb support during overhead work. Results point out that PAEXO provides a reduction of the whole-body joint effort across the experimental task blocks (from 66% to 86%). Moreover, the analysis along with five different body areas shows that 1) the exoskeleton provides support at the human shoulders by reducing the joint effort at the targeted limbs, and 2) that part of the internal wrenches is intuitively transferred from the upper body to the thighs and legs, which is shown with an increment of the torques at the legs joints. The promising outcomes show that the probabilistic estimation algorithm can be used as a validation metric to quantitatively assess PAEXO performances, paving thus the way for the next challenging milestone, such as the optimization of the human joint torques via adaptive exoskeleton control.

2) that part of the internal wrenches is intuitively transferred from the upper body to the thighs and legs, which is shown with an increment of the torques at the legs joints. The promising outcomes show that the probabilistic estimation algorithm can be used as a validation metric to quantitatively assess PAEXO performances, paving thus the way for the next challenging milestone, such as the optimization of the human joint torques via adaptive exoskeleton control.
Index Terms-Human whole-body joint torque, floating-base estimation, overhead work analysis, upper-limb exoskeleton.

I. INTRODUCTION
E XOSKELETONS have been part of the robotics scenario for several decades. The scientific interest in exoskeletonbased devices was born in the early '60s to augment human performances for military purposes [1]. Since then, exoskeleton development has evolved into a powerful tool for medical/rehabilitation purposes, where exoskeletons are used to help impaired patients or elderly people to perform daily-life activities [2]- [4]. The last decade, however, revealed a rising demand for exoskeletons tailored for industrial applications [5], [6]. For example, assembly-line workers are highly exposed to the genesis of pathologies related to physical stress due to repetitive upper-body movements (e.g., in the automotive workplace, workers have to reach above their heads thousands of times a day when working on the underside of cars). The 2019 report of the European Union on the working conditions in a global perspective [7] showed that pathologies related to repetitive hand or arm movements are the most pervasive job-related risk by affecting more than 60% of the working population in Europe. The report claims that holding a painful body posture (43%) or carrying heavy loads (32%) are other important risk factors for European workers. An industrial exoskeleton can compensate for these drawbacks by providing physical assistance to its user via assistive torques. Industrial exoskeletons have been therefore recently developed to prevent work-related musculoskeletal disorders (WMSDs) and not only to rehabilitate existing pathologies.
Overhead work has been classified as one of the major risk factors for the onset of shoulder WMSDs and muscle fatigue [8]. Occupational upper-limb exoskeletons can be used to assist workers during the execution of overhead tasks to prevent shoulder WMSDs. Several investigations have been recently This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ conducted on upper-limb exoskeletons showing a reduction of muscle fatigue while executing an overhead task [9]- [11]. PAEXO (Ottobock SE & Co. KGaA, Duderstadt, Germany), EksoVest (Ekso Bionics, Richmond, CA, USA), MATE (Comau, Turin, Italy), ShoulderX (SuitX, Emeryville, CA, USA), and SkelEx (Skel-Ex, Rotterdam, The Netherlands) are only some of the relevant commercialized examples of upper-extremity exoskeletons for arm support in overhead work. This category of exoskeletons has been conceived for 1) compensating the effect of the gravity on the arm, 2) reducing the upper-body biomechanical loads, 3) preventing long term WMSDs while 4) guaranteeing the work quality and productivity for the companies.
To reduce WMSDs, industrial exoskeletons have also the potential to save substantial costs caused by occupational injuries. It is therefore not surprising that companies are moving toward the large-scale adoption of exoskeletons for their workers for promoting modern production industrial methods. In this respect, in 2018 Ottobock announced the largest long-term pilot test of exoskeletons in Europe and successfully rolled out in Volkswagen plants several PAEXOs, their passive upper-limb exoskeleton. The pilot test has been carried out to mimic real assembly line overhead working conditions with 30 employees. The test found a remarkable success among the workers whose subjective feedbacks were very positive [12]. This reason has been the engine for the quantitative assessment analysis carried out in the in-lab study in [13], where effects of the PAEXO have been evaluated with respect to (w.r.t.) objective and subjective criteria. Results showed that the exoskeleton reduces shoulder physical strain as well as global physiological strain. However, the study did not analyze the effects of PAEXO at the whole-body internal wrenches level, which is also a pivotal aspect for a better comprehension of the long-term effects of using the exoskeleton. Therefore, a further step has been done in this article to strengthen the results achieved in [13]. It complements the previous work with a quantitative joint torque estimation assessment at the whole-body effort level. The effort analysis has been performed via a floating-base whole-body joint torque estimation algorithm based on a Maximum-A-Posteriori (MAP) estimator implementation. The estimation algorithm has been already validated in [14] w.r.t. standard inverse dynamics methods. The main MAP advantage concerns the possibility to discriminate between sensors according to their reliability (in terms of sensor covariance tuning). Furthermore, the MAP has been developed in such a way to be integrated into a real-time software architecture for the simultaneous estimation of kinematics and dynamics for robotics-based applications.
The novelty proposed in this article is twofold. On one hand, we exploit the scalability of the estimation algorithm by introducing the exoskeleton as a wearable device, i.e., a new type of sensor in the network. On the other hand, we define a new metric to evaluate the whole-body support provided by the exoskeleton. The definition of the new approach implicitly embodies a benefit when compared to standard experimental assessments of the human effort. From an experimental perspective, for example, it is impractical to distribute EMGs all over the body to gather the whole-body biomechanical effects. This article is structured as follows. Section II defines the mathematical notation and the human kinematics and dynamics modeling. It also recalls the PAEXO description. Furthermore, the algorithm formulation for computing the joint torque estimation is here discussed. In Section III, the set-up and the instrumentation for the overhead experiment are described. Section IV discusses the algorithmic approach for the joint torque estimation analysis. Section V presents analysis results. Discussion on the results and several conclusions are reported in Sections VI and VII, respectively.

A. Notation
• Let x ∈ R n denote a n-dimensional column vector when written in bold text; a scalar quantity when written in nonbold text (i.e., x). • Let I be an inertial frame with z axis pointing against the gravity (g denotes the norm of the gravitational acceleration). B denotes the base frame, i.e., a frame attached to the base link of the system.

B. Recall on System Dynamics Modeling
The human model is conceived as a rigid multibody system composed of N B rigid bodies, called links, connected by n joints with one internal degree of freedom (DoF) each. We assume that none of the links have a known a priori constant pose w.r.t. I. Thus, we say that the system is floating base. The system configuration space is a Lie group , andṡ ∈ R n the joint velocities vector. If the system is interacting with the external environment by exchanging n c wrenches, the dynamics of the floating-base system can be described by adopting the Euler-Poincaré formalism [15], (Ch. 13.5): where, M ∈ R (n+6)×(n+6) is the mass matrix, h ∈ R n+6 a vector accounting for the centrifugal and Coriolis effects and gravity terms, B := (0 n×6 , 1 n ) is a selector matrix for the joint torque vector τ ∈ R n . f ext k ∈ R 6 is a vector representing the external wrench acting on the link that has the kth contact point. The Jacobian J C k (q) ∈ R 6×(6+n) is an operator mapping the system velocity ν with the velocity v of the kth contact frame C k , such that where, J b (q) ∈ R 6×6 and J b (q) ∈ R 6×n are the Jacobians related to the base and joint configuration, respectively. Furthermore, it is assumed that a set of holonomic constraints act on the system in (1). These constraints may represent, for instance, a frame having a constant pose w.r.t. I. If this frame corresponds to C k , then I v C k = 0. Thus, we represent the holonomic constraint such as The first-order time derivative of (3) yields to Equations (1) and (4) represent the constrained dynamics of a floating-base system.

C. The Floating-Base Whole-Body Joint Torque Estimation
The floating-base estimation of the whole-body joint torques has been performed via a probabilistic approach by means of a Stack-of-Tasks Maximum-A-Posteriori (SoT MAP) estimator. To this purpose, (1) and (4) have been equivalently rearranged into a compact matrix form, i.e., where Y is a matrix accounting for the sensor measurements, D is a matrix for the system constraints. Bias terms b Y and b D are related to the above-listed matrices, respectively. The vector y contains the measurements injected into the algorithm, such as IMUs and force/torque sensors. The vector d embodies kinematics and dynamics quantities for links and joints of the model, i.e., where being α g ∈ R 6 the proper sensor acceleration, f ∈ R 6 the internal joint wrench ands the joint acceleration. The goal of the algorithm is to solve (5) in the variable d. The vector in (6) implicitly contains the joint torque τ as a projection of f on the joint motion freedom subspace S, such that τ = S f , ∀n. The algorithm maximizes the probability of d given the measurements availability. In this (Gaussian) domain, both the vectors d and y are stochastic variables. The estimator, therefore, looks for the mean of a conditional probability, i.e., d = μ d|y , and its covariance Σ d|y , such as In Fig. 1, the choice of the covariances Σ plays a pivotal role. Covariances accounts for the reliability of 1) the sensors input quantities (i.e., Σ α from IMUs and Σ f ext from force/torque sensors and exoskeleton), 2) the model constraints (i.e., Σ D ), and 3) the prior knowledge on the solution (i.e., Σ d ). Trusted covariances are typically low values. Conversely, high values stand for nontrusted covariances. Technical details for building the system in (5) and the formulation of its solution in (8) can be found in [14].

D. The Stack-of-Task (SoT) Variant
The SoT MAP variant [16] allows to solve the algorithm by decoupling the estimation of the joint torques from the internal wrench estimation. An important advantage of using the SoT variant is the possibility to estimate those noncollocated wrenches for which physical sensors for the measurements are not available, e.g., external wrenches at the hands without having sensorized gloves placed on them. From an implementation perspective, the variant solves system in (5) as a stack of two tasks, as described in the pipeline of Fig. 1.
Task 1: computation of the equation where y is a vector containing only a subset of y measurements. The subset encompasses whole-body external wrenches f ext and a new constraint on the rate of change of the centroidal momentum Bḣ expressed w.r.t. the base frame B as follows: where, B X * C[I] is the 6 × 6 adjoint transform for the external wrenches expressed w.r.t. the centroidal frame C[I] into the base frame B. Term m denotes the mass of the model,ẍ COM is the acceleration of the model COM. The Task 1 solution d is a vector containing only the estimation of the whole-body external wrenches.
Task 2: computation of (5) by considering the Task 1 output d as an new input measurement for the Task 2. The measurements vector y encompasses now the entire set of sensor measurements (i.e., the external wrench from the Task 1 estimation and input measurements to y in Task 2).

E. The Passive Exoskeleton PAEXO
PAEXO is a lightweight passive upper-limb exoskeleton developed by Ottobock SE & Co. KGaA together with Volkswagen AG [17]. PAEXO has been designed to maximize users' freedom of movement and comfort while providing a reasonable assistance. PAEXO provides a support to the user's arms by transferring a portion of the arm weight to a hip belt. It has  been designed to 1) generate support torques to the users' shoulders, 2) compensate for the arm weight while not hindering or perturbing the motion. Therefore, the provided support varies with the shoulder flexion and arm elevation angle β. It is maximum at β 90 • (i.e., upper arm horizontal) and zero when the arm is lowered along the body. The assistive structure of the exoskeleton consists of a support bar and an arm bar connected with a hinge joint (see Fig. 2). A passive actuator generates the adjustable support torque in this joint. The structure is attached to the human upper arm with bracelets and to the hip via a hip belt. The exoskeleton is adjustable to fit different body sizes. A textile stabilization structure keeps the support structure close to the body and allows it to move comparable to the shoulder blade. The result is a free movement of the trunk and upper extremities with a weight of only 1.8 kg. Furthermore, the absence of rigid elements in the back enables the full range of back movements. In addition, the level of support can be manually step-less tuned via a mechanical parameter of the passive actuator (i.e., the lever arm length) to adapt to different arm weights or compensate for the extra weight of a tool.

F. The Human Model
The human body model consists of N B = 49 links and n = 48 internal DoFs. The links have been modeled with simple geometric shapes (parallelepiped, cylinder, and sphere) whose dimensions are computed via IMU sensors estimation. This approach guarantees the model scalability with the human kinematics. The dynamic properties of each link (i.e., inertias and COM) have been computed using anthropometric data available in literature by 1) exploiting the relation between the total body mass and the mass of each link [18] and 2) assuming geometric approximations and homogeneous density for the links [19]. A universal robot description format (URDF) model has been created for representing the kinematics and dynamics of the human multibody systems [20], Fig. 3. The exoskeleton was not considered in the URDF model but modeled as a set of external wrenches acting on the contact upper-body links (i.e., upper arms, pelvis).

III. THE OVERHEAD EXPERIMENT
A lab study has been conducted to analyze the effects of PAEXO at the whole-body internal wrenches level. Participants performed a repetitive overhead pointing task, which mimics the tasks commonly observed at the assembly line in the automotive industry. The study was approved by the Slovenian National Medical Ethics Committee (No. 339/2017/7) and it has been conducted in accordance to the Declaration of Helsinki. Data collection is available on-line: https://zenodo.org/record/1472214.
Twelve healthy volunteered males were recruited for the experiment (age: 23.2 ± 1.2 yrs, height: 179.3 ± 5.9 cm, mass: 72.7 ± 5.4 kg). Participants were college students with no or limited industrial experience and novice with exoskeletons. They gave written informed consent before starting the experiment. Participants were told to keep 1) their left hand on the frame of an interactive screen horizontally positioned above their head, 2) their feet on two signs placed on the floor below the screen and 3) to perform a task with a hand-held tool (power drill, weight: 0.66 kg) with their right hand, Fig. 4. The task consisted of moving the pointing tool as fast as possible from a starting point to a target, and remaining on the target for 2 s. The screen height was adjusted for each participant so that the tool tip touched the screen when the participant was standing with the right shoulder and elbow flexed at 90 • . The task target was always localized in the center of the screen and appeared in two different sizes: 23 and 46 mm diameter. The starting point appeared in four different points equally spaced on a circle of radius 1/6th of the participant's height and centered on the target. The target turned green if hit and red if missed. If the target was missed, participants could adjust the tool position until they hit the target. Participants performed the task with the exoskeleton (WE session) and without the exoskeleton (NE session). They were equally divided into two groups: A and B. Group A started in NE session and then WE session, while Group B did the opposite to counterbalance the effects of learning and fatigue. Each session consisted of 5 blocks of 24 pointing movements with a 30 s break between blocks. Each block lasted approximately 2 min, resulting in a 12 min long session. Within a block, each combination of target size (two sizes) and starting position (four positions) appeared three times in random order. Between the two sessions, participants rested for 15 min to prevent experiment-induced fatigue.

A. Instrumentation of the Human
Participants were equipped with the Xsens inertial motion tracking wearable system (Xsens, Enschede, The Netherlands) with 17 IMUs to track the whole-body kinematics. Data were recorded with the Xsens MVN software at 60 Hz. A pair of sensorized shoes developed at IIT equipped with four six-axis force/torque sensors has been used to detect the ground reaction 6-D wrenches (three forces, three moments) [21]. Data were recorded at 1 kHz with a custom-made MATLAB script running on Simulink Real-Time.

IV. JOINT TORQUE ESTIMATION ANALYSIS
The whole-body torque analysis has been performed by using the SoT MAP probabilistic estimator described in Fig. 1 for both the NE and WE. The difference between the two sessions concern the inputs in the measurement vector y, i.e., where α g IMUs ∈ R 6 is the IMUs proper sensor acceleration, f ext SHOES ∈ R 6 the sensorized shoes external wrench and f ext EXO ∈ R 6 the external force due to presence of the exoskeleton.
PAEXO raw measurements F R consist of the following: • A support force for right upper arm, F RSRF .
• A support force for left upper arm, F RSLF .
• A force that is transferred to the hip, F RHF . Fig. 5 shows the raw PAEXO forces measurements F RSLF , F RHF , F RSRF across five blocks, respectively, at the inter-subject level (i.e., on the whole 12-subject dataset). The subjects' posture for block five requires a PAEXO maximum support at the arms and this corresponds to an increase of transferred forces at the pelvis. Per each subject, the exoskeleton manufacturer provided raw force measurements computed as a function over

A. PAEXO Versus URDF Angles Compatibility
The first analysis objective required to make the PAEXO angle measurements compatible with the human URDF model defined as in Section II-F. In the URDF model [see Fig. 6(b)], let u := [u x u y u z ] be a Euler angle vector which parametrizes a rotation matrix R of the upper arm w.r.t. the shoulder, defined by None of the angles in u coincides with β. Thus, we defined a new Euler angle vector u = [u x u y u z ] to parametrize  that, by construction, is equivalent to (13). More in detail, (14) can be written as R = ⎡ ⎢ ⎣ r 11 r 12 r 13 r 21 r 22 r 23 r 31 r 32 r 33 and the resulting ZXY Euler angle vector u [22] is such that

B. PAEXO Versus URDF Forces Matching
The further objective was to transform the raw exoskeleton forces into a compatible format for the SoT MAP algorithm, i.e., a wrench vector ∈ R 6 as required in (12). To achieve the goal, we first extracted tabulated forces values accordingly to the new computed vector u . Later, we considered a pure forces assumption (i.e., no moments) on the URDF model per each contact frame C k where the exoskeleton exerted a force, (i.e., upper arms, hips). Fig. 8. Comparison of the inter-subject whole-body normalized torque norm ||τ wb || for non-exoskeleton (NE) and exoskeleton (WE) sessions, across five blocks. The two-way repeated-measures analysis of variance (ANOVA) shows that the torque estimation is significantly affected by the presence of the exoskeleton while it is not affected by the five-block division of each session.

V. RESULTS
Whole-body joint torque estimation of NE and WE sessions have been compared to assess the effect of the exoskeleton at the inter-subject level. An instance of the estimation algorithm in Fig. 1 has been performed per each subject, each session, over time (i.e., across five blocks of a session). Algorithm covariances have been chosen by following a covariance tuning metric, i.e., the minimization of the error between measured and estimated quantities with a relative error such that The 3% threshold is the maximum allowed relative error by mainly taking into account the error of IMUs and also the propagation of the error due to the sensor fusion process. The mean of the inter-subject whole-body (wb) torques τ [N · m] for NE (orange line) and WE (green line) sessions has been performed across five blocks, with the following formulas: being τ 1 , . . . , τ 12 the subjects whole-body torques estimated by the algorithm. Fig. 7 shows a whole-body reduction of the overall joint effort module due to the exoskeleton support with the with related percentage in Table I, across five blocks. A statistical analysis has been computed for the inter-subject normalized whole-body torque norm ||τ wb || along with the sessions, across five blocks. A two-way repeated-measured analysis of variance (ANOVA) has been performed with two within-subject fixed factors (i.e., sessions and blocks). Subjects are entered as a random factor. Fig. 8 shows the evolution of the normalized torque norm across the blocks. The ANOVA reveals a significant effect of the "sessions" factor on the torque estimation (p-value < 0.01). Conversely, the blocks do not affect the estimation (p-value = 0.0919) and no statistical evidence has been detected of an interaction effect between the two factors. However, the only whole-body level analysis is not sufficient to assess the effects due to the exoskeleton. An analysis along with five different body areas has been performed to evaluate the exoskeleton effects locally to the torso (t), left and right arms (a), left and right legs (l), such as Figs. 9-11, show the difference between the mean of the intersubject torques computed for NE and WE for the arms, legs, and torso, respectively. Bar plots quantify the magnitude of the effort reduction (green bars) and the effort increase (red bars) in performing the task with the exoskeleton. Plots follow the area division as in Fig. 3. Each cluster of five bars (i.e., five blocks) represents the entity of the effort reduction/increase for a specific range of motion allowed for the joints (e.g., jL5S1 rot x represents the rotation around the x axis for the joint jL5S1).
The effort due to PAEXO is evident at the arms, especially at both the shoulders (up to ∼10 N· m), Fig. 9. Another important evidence is the effort reduction the human gathers on the torso, from L5 to T8 links, Fig. 11. Intuitively, the human gained an effort reduction at the arms and the torso wearing PAEXO at the cost of an increase of the effort at the pelvis and the hips, where the load is transferred, Fig. 10.

VI. DISCUSSION
This study analyzed the whole-body effects of the passive upper-limb exoskeleton PAEXO from a joint torque analysis perspective. The analysis has been performed by reproducing in a lab environment the conditions of an industrial overhead work scenario with a dataset of 12 novice participants. A Stack-of-Tasks-based Maximum-A-Posteriori (SoT MAP) estimator has been used for the whole-body joint torque estimation of the human modeled as a floating-base system.
A first quantitative analysis in [13] investigated the objective and subjective effects of wearing PAEXO to reduce the WMSDs among overhead workers. This article complements the previous study by analyzing the effects of PAEXO at the internal wrenches level. From a whole-body promising perspective, the analysis shows that internal whole-body torques are significantly reduced when using the exoskeleton ranging from 66% in Block 1 to a maximum of 86% in Block 3. Moreover, the analysis shows that the presence of the exoskeleton reduces also the internal articular stress at the shoulders and torso level, (se Figs. 9 and 11). This last outcome is in line with the EMG results of [13] but the analysis offers here a more exhaustive picture of the whole-body Fig. 9. Difference of the mean of the inter-subject torquesτ a [N · m] for non-exoskeleton (NE) and exoskeleton (WE) sessions, across five blocks, for the left and right arm, respectively. The torque difference is the metric to investigate the magnitude of the effort reduction (green bars) and the effort increase (red bars) at the arms joints (black labeled in Fig. 3) due to the exoskeleton presence. Fig. 10. Difference of the mean of the inter-subject torquesτ l [N · m] for non-exoskeleton (NE) and exoskeleton (WE) sessions, across five blocks, for the left and right leg, respectively. The torque difference is the metric to investigate the magnitude of the effort reduction (green bars) and the effort increase (red bars) at the legs joints (red labeled in Fig. 3) due to the exoskeleton presence. Fig. 11. Difference of the mean of the inter-subject torso torquesτ t [N · m] for non-exoskeleton (NE) and exoskeleton (WE) sessions, across 5 blocks. The torque difference is the metric to investigate the magnitude of the effort reduction (green bars) and the effort increase (red bars) at the torso joints (magenta labeled in Fig. 3) due to the exoskeleton presence. On the right-hand side, an example of a one-subject visualization comparison of the joint effort is shown. The color of the spheres represents the change in the joint effort for the same task performed in the two sessions, ranging from red (i.e., high effort) to green (i.e., low effort).
biomechanical effects, as well as the unquestioned advantage of an easier doability. However, the reduction of the effort at the upper body is compensated by a general increment of the stress at the legs. Internal wrenches are intuitively transferred to the lower body, especially to the hips (see Fig. 10). The right-hand side of Fig. 11 shows a whole-body example of one-subject joint effort visualization. The color of the spheres represents the change in the joint effort for the same task performed in the two sessions, ranging from red (i.e., high effort) to green (i.e., low effort). The time-frame visualization confirms the results just discussed.

VII. CONCLUSION
This article achievements show that the estimation algorithm can be used as a validation tool to assess if PAEXO is able to measure quantitatively the overhead workers effort. However, several factors may impact the effectiveness of the results. From a modeling perspective, the conceptual design of an exoskeleton requires alignment with the user's biological joint kinematics. The exact position of the human joints axes is impractical to be retrieved in vivo thus leading to a mismatch between the device and the real axes. Misalignments (of the order of a couple of centimeters) generate residual wrenches by the propagation of the error. This modifies the physiological kinematics of the body, may increase the physiological demand [23], and often it is perceived by the user as discomfort [24]. This can induce, in turn, an increment of biomechanical stress at a not-targeted limb because of the potential onset of physiological adaptation. Concerning the human-exoskeleton model pairing, the biggest limitation is represented by the difficulty of simulating the body interface at the hip where the exoskeleton is attached. From an experimental perspective, experiments have to be carried out with real workers who experience real working issues in an industrial environment. In this regard, the initial laboratory study we presented is a good resource for overcoming possible factories' reluctance to be involved in such trials. In general, the promising outcomes of the analysis pave the way for the very next challenging milestone, such as the optimization of the human joint torques via adaptive exoskeleton control. For the optimization computational cost, it will be extremely useful to use simplified human models rather than musculoskeletal models, thus accepting a loss of accuracy from the biological point of view. This is an acceptable trade-off for the development of control algorithms that can be complemented with musculoskeletal analyses. In general, we envision that in the long run the large-scale adoption of the exoskeleton control will be a fundamental step towards the reduction of WMSDs.

VII. ACKNOWLEDGMENT
The content of this publication is the sole responsibility of the authors. The European Commission or its services cannot be held responsible for any use that may be made of the information it contains.