The importance of muscle fatigue assessment in patients with neuromuscular disorders has long been recognized. Specifically, a precise measure of muscle fatigability could provide crucial information in diagnosis, treatment planning and evaluation of therapy efficacy. Yet, there is a lack of quantitative standardized and reliable methods for fatigue assessment used in clinical practice and these may not be feasible for all pathological populations at various stages of a disease. Muscle fatigue assessment using surface electromyography signals has been widely studied under isometric conditions [
38] and in particular during maximum voluntary contractions [
18,
20,
21]. However, the capacity to voluntarily generate and maintain a maximal force in isometric conditions might be limited by a lack of motivation and it could represent a highly demanding task for people affected by severe weakness or injury. When dealing with adolescent populations, it is common to see a lack of full cooperation, which makes it challenging to measure (in a reliable manner) maximum contraction force and consequently muscle fatigue [
25]. Furthermore, it is worth highlighting that subjects with low residual motor functions are hardly ever able to perform MVCs. Hence, few tests based on dynamic exercises have been proposed and validated [
39,
40], thus representing a consistent alternative to isometric contractions. Based on these considerations, our approach aimed to evaluate muscle fatigue, regardless of individual motor capabilities. In particular, the use of a robotic task allows for the experimenter to tailor the evaluation, in term of range of motion and force required, to an individual subjects capability and strength, thus allowing for adoption by a large patient population. Additionally, the adoption of a method relying on robot mediated movements assures a more controlled and repeatable execution of the test than currently used isometric or dynamic exercises. A preliminary, but different, version of such a test has already been validated in a pilot study which revealed its feasibility and repeatability [
31]. As an extension and consolidation of that preliminary study, in this paper we report the results obtained with 40 healthy subjects tested with a novel improved version of the robot-based assessment protocol. In particular, we used a visco-elastic force instead of a pure viscous force field since, with neuromuscular patients, a force that depends on velocity could reduce the repeatability of the results. In addition, we reduced the effort required to perform the test, with a resistive visco-elastic force in one direction. This work revealed that the proposed test is easy and fast to administer, provides an objective and reliable measure of muscle fatigue and can be used in a clinical setting. It has also to be mentioned that the use of the robotic device adds the ability to measure subjects’ performances in terms of kinematic parameters, thus resulting in a more detailed assessment of the patient. In the present study, the kinematic analysis demonstrated the stability of the
OF indicators and it appears robust, given different motor control strategies. Regarding the applicability of the method, the experimental setup is minimal, requiring sEMG from two target muscles and the correct alignment between the human wrist and the robotic device. From a clinical perspective, test duration is also important and our test never lasted more than 3 min. It is even reasonable to expect a shorter test time in clinical populations compared to our healthy participants. We chose to base our indicator of
Onset of Fatigue on the
Mean frequency since its variance is tipically lower than that of
Median frequency [
41]. The shift in
Mean frequency towards the lower frequency spectrum was noticeable in both the
flexion and
extension groups, however the shift was greater in the former. This may be due to different physiological properties of the two muscles: I) from a biomechanical perspective of the human wrist joint, the amplitude of the range of motion in flexion is higher than that achievable in extension (peak flexion moment is approximately 70% higher than peak extension moment [
42]); II) the percent decrement of
Mean Frequency is proportional to the amount of catabolites produced by muscle fibers during activity [
43]. In particular, the quantity of catabolites depends on the average number of muscle fibers per square unit of the muscle crossection [
43] and consequently the higher the crossection, the higher the amount of catabolites and the greatest the rate of decrease of the
Mean Frequency ; III) muscle fiber type will influence sEMG parameters, in particular, a greater percentage of type II fibers leads to a greater rate of decrease of the
Mean Frequency [
44]. As reported from other studies, muscles belonging to the extensor group fatigue more and faster than flexors [
45], therefore,
EG population was expected to exhibit more fatigue. However, directly comparing the single muscles
extensor and
flexor carpi radialis, the physiological cross-sectional area (PCSA) of
flexor carpi radialis is about half the size of the
extensor carpi radialis [
46]. Therefore, we can assume that
flexor carpi radialis has a lower force generating capacity. We can speculate that, on a whole, when using the same force field intensity for both flexion and extension, the extensor muscles would fatigue more than the flexors. It should be noted that we measured only one wrist extensor muscle. The most important contribution of this method is the development of and evaluation of the
OF measure. The
OF in the
EG presented less variability compared to the
FG, probably because of the different muscle properties mentioned above. As for the optimal version for
Onset of Fatigue, the comparison among
OF25%,
OF50% and
OF75% revealed that
OF25% is the more consistent and less variable. We can postulate that after an initial decrease, which is very similar in both groups, the
Mean Frequency curves decreased with different slopes due to different subject training levels and muscle physiological properties. Our
OF25% is also more inline with previous studies that have suggested that a mean frequency decrease of 8% is representative of muscle fatigue onset [
47,
48]. Moreover, our subjects’ ability to perform the task correctly from the very beginning and the consequent stability of the kinematic parameters, support the adoption of
OF25% as an indicator of fatigue.
Mean speed stabilized in the first 20% of the task, suggesting that
Mean Frequency and
OF25% are not related to kinematic changes. It is worth noting that, in some cases, during the last part of the task, the trend was inverted. From a physiological perspective, this may be related to a de-recruitment of fatigued motor units, in favor of the recruitment of new motor units [
24]. This finding is also in line with previous studies showing that, during submaximal contractions, motor units recruitment can still increase when motor units start to be fatigued, while during maximal contractions such a rise is limited [
49,
50]. On the other hand, we are aware that the frequency recovery could be due to the effect of cross-talk between adjacent muscles. However, since we collected from the
flexor and the
extensor carpi radialis, further investigations recording additional muscles are needed to examine the potential effect of cross-talk during our task. Additionally, it has been reported that an increase in muscle temperature leads to an increase in
Mean Frequency [
51]. Therefore, future work might consider measuring surface temperature or muscle temperature to investigate any potential relationship between an increase in temperature and the inversion of the
Mean Frequency trend observed in our study. Regarding kinematic measurements, we found that the
Mean Speed stabilized and remained constant after an initial phase, corresponding to the first 20% of the task, in which it increased up to a plateau level. This suggests that
Mean Speed was not affected by muscle fatigue (and vice-versa), which is in line with previous finding [
52‐
54]. Conversely from what was expected, we did not find changes in kinematic strategies that correlated with increasing fatigue level. Specifically, the
TPR did not show a shift in the peak of the bell-shaped speed profile [
55] from early trials to the late trials in which fatigue appeared. A final aspect to consider was task duration. In the proposed protocol, the number of repetitions performed, was decided by the subjects and not superimposed by the experimenter. Thus, subjects were instructed to stop when they felt tired, which is crucial in a clinical scenario. The number of repetitions, therefore, could also be considered as an additional measure of performance [
56], especially for populations with neuromuscular impairments, where kinematics and sEMG might have to be cautiously interpreted. Our results in healthy participants demonstrate that
OF was independent from the amount of repetitions of the reaching movements performed. This may be a consequence of the population studied, who could tolerate a high level of resistance and may not stop the test when they feel fatigue. To conclude, the developed algorithm could be improved in the future by measuring individual wrist strength and grip force throughout the task [
45]. Our approach used two levels of force, according to sex, but customizing the force and normalizing to individual maximal force production, could improve
OF results. This aspect is particularly relevant and needs to be considered in the application of this method to pathological subjects. Lastly, this study suggested that a fatigue assessment coupling a robotic task and sEMG recordings is highly feasible and practical. The present study provided a good starting point for the application of the test in clinical practice, however pilot experiments with neuromuscular and age-matched healthy subjects are necessary to confirm the results. Finally, wrist robotic device guarantees the repeatability of the task, providing the same force and trajectory. Moreover without the addition of further measurement tools we attempted to exploit the torque and angular position data recorded by the robot as a simple measure of the mechanical work performed by the subject during the test. Although such an approach does not provide specific information about actual internal muscle work or physiological work, it allowed for an estimate of the total energy required by the task. Our test has been demonstrated to require little effort, so the impact on daily energy expenditure (avg 2500–3000 kcal) would be minimal [
57].