The research team, Thomas Jürgen Klotzbier and Nadja Schott of the Department of Sport and Exercise Science, recruited 40 healthy young adults with an average age of 22.4 years. Participants were randomly assigned to one of two versions of a working-memory task known as the n-back task. In the low-load 0-back condition, 21 participants simply pressed one key when the letter “a” appeared on screen and another key for the letters “b,” “c,” or “d.” In the high-load 2-back condition, 19 participants had to judge whether each letter matched the one shown two trials earlier — a manipulation that substantially increases working-memory demands. Each session involved five blocks of 160 trials over roughly 40 minutes, with brief breaks in between.

Before and immediately after the cognitive task, participants rated their current state using the Mental and Physical State and Trait Energy and Fatigue Scales, in which each subscale combines three visual-analogue items scored from 0 to 420 millimetres. The ratings told a clear story. Perceived mental fatigue rose significantly, with a mean increase of 74.4 points, and perceived mental energy fell by 61 points, both highly significant in the repeated-measures analysis. Perceived physical fatigue also increased by a smaller but significant margin of 43.3 points, whereas the decline in physical energy — 22.2 points on average — did not reach statistical significance. Notably, the magnitude of these changes did not differ significantly between the 0-back and 2-back groups, suggesting that even a repetitive, lower-load task performed for 40 minutes was enough to shift subjective state.

Here, however, came the first of the study’s important dissociations. Although participants felt considerably more fatigued, their objective performance on the task did not deteriorate in a consistent way. Reaction times actually improved across the five task blocks in both conditions. Accuracy decreased in the 0-back condition but improved in the 2-back condition. The authors therefore describe the results as evidence of increased perceived mental fatigue rather than convincing cognitive performance fatigability — a distinction they emphasize, since subjective tiredness and objective decline in performance are related but separate phenomena. Practice, familiarization, or strategy adaptation may have offset any fatiguing effect of the task.

The second half of the study asked whether these subjective states were reflected in the body’s movement. Each participant walked on a treadmill at a fixed speed of 5 km/h under two conditions — level walking at 0 degrees and downhill walking at a 9-degree decline — once before and once after the cognitive task. Gait was captured with six APDM Opal inertial sensors attached to the wrists, ankles, lower back, and sternum, sampling at 128 Hz, with the Mobility Lab system detecting foot contact and toe-off events and computing a broad battery of spatio-temporal parameters, from stride length and cadence to trunk range of motion and foot elevation during the swing phase. On average, about 29 steps per 40-second trial entered the analysis.

The headline result is, in a sense, an absence of results. Of the 200 correlations computed between post-task fatigue and energy ratings and pre-to-post changes in gait parameters, only five reached unadjusted significance — fewer than the roughly ten that would be expected by chance alone. And after Benjamini–Hochberg false-discovery-rate or Holm correction across all 200 tests, none of the associations remained statistically significant; the smallest adjusted p-value was 0.80. The authors are careful to stress that these findings are exploratory and hypothesis-generating, not confirmatory evidence.

Still, the few nominal associations are intriguing enough to shape future research. During downhill walking, perceived mental fatigue showed a small-to-moderate positive association with elevation at midswing — the height to which the foot is raised during the middle of the swing phase — with r = 0.329, a 95% confidence interval of 0.010 to 0.587, and an unadjusted p-value of 0.044. The researchers speculate that this could reflect a more cautious foot-trajectory strategy: when walking downhill, which demands continuous braking and control of forward progression, keeping the foot higher might increase clearance from the walking surface. But they are emphatic that this hypothesis could not be tested with the present measurements. Elevation at midswing is not the same as minimum toe clearance, no direct stability or trip-response measures were taken, and the finding does not establish a safer or compensatory gait adaptation.

The other cluster of nominal findings concerned trunk coronal range of motion — lateral trunk movement in the frontal plane, which helps regulate the body’s center of mass relative to its base of support. During downhill walking, subjective physical energy (r = −0.452), mental energy (r = −0.401), and physical fatigue (r = 0.344) were each nominally associated with changes in this parameter, and mental energy also showed a nominal association during level walking (r = −0.345). In other words, participants who felt more fatigued or less energetic tended to show greater lateral trunk movement after the task. The authors caution that this pattern may reflect an altered movement strategy or simply normal movement variability rather than impaired stability. They point to prior work showing that trunk motion, step width, and center-of-mass control are functionally coupled, and to the idea that movement variability can be functional when it allows performance to be maintained under changing constraints. Crucially, the correlational design cannot show that fatigue caused the changes, and no direct measures of dynamic stability were obtained.

The study carries a detailed list of limitations that the authors themselves lay out candidly. The sample of 38 complete cases for the questionnaire analyses provided 80% statistical power only for correlations of approximately |r| = 0.44 or larger, so smaller associations could easily have been missed — and the absence of significant correlations should not be read as proof that gait was entirely unaffected. The 0-back condition cannot be considered a fatigue-free control, since 40 minutes of even repetitive responding may induce boredom, reduced arousal, or fatigue. Gait change scores were based on a single 40-second recorded trial per condition and time point, leaving them vulnerable to measurement error and regression to the mean. The fixed treadmill speed may have imposed different relative demands on different participants, and the fixed order of the walking conditions means that level-versus-downhill contrasts may partly reflect order effects. No recovery period was scheduled, the exact intervals between task and post-task walking were not recorded, and variables such as sex, body mass index, and physical activity were not controlled in the correlation analyses. Finally, because the 2-back group reported more weekly physical activity on average (295 versus 226 minutes), baseline equivalence between conditions was not established.

What emerges, then, is a picture of selectivity rather than widespread impairment: in healthy young adults, subjective fatigue and energy states showed scattered, exploratory associations with trunk and foot movement but no broad pattern of fatigue-related gait change. The authors call for larger, preregistered, adequately powered studies with prespecified primary outcomes, correction for multiple comparisons, direct measures of dynamic stability such as minimum toe clearance and mediolateral margin of stability, joint kinematics, and muscle activity, as well as randomized or counterbalanced comparisons of level, uphill, and downhill walking and comparisons across age groups. Until such work is done, the findings stand as a careful first step — evidence that a mental workout can make young adults feel drained without detectably disrupting the mechanics of how they walk, but with tantalizing hints that the most demanding terrain may subtly reshape the foot’s path.

Subject of Research: Associations between perceived mental fatigue and energy and gait characteristics during level and downhill walking in healthy young adults

Article Title: Exploring associations of perceived mental fatigue and perceived energy with gait characteristics during level and downhill walking in healthy young adults

Article References: Klotzbier, T. J., & Schott, N. (2026). Exploring associations of perceived mental fatigue and perceived energy with gait characteristics during level and downhill walking in healthy young adults. Frontiers in Sports and Active Living, 8, Article 1973398. https://doi.org/10.3389/fspor.2026.1973398

Image Credits: AI Generated

DOI: 10.3389/fspor.2026.1973398

Keywords: mental fatigue, gait analysis, downhill walking, working memory, n-back task, inertial sensors, trunk range of motion, midswing elevation, young adults, biomechanics, treadmill walking, exploratory study

The heart of the contribution is ISL-MDN, a publicly released dataset of inertial recordings capturing continuous multi-digit number signing. Forty-one participants wore a glove instrumented with six IMU sensors sampling at 100 hertz, producing streams of acceleration and angular velocity data as they signed sequences of two, three, and four digits. Crucially, the dataset includes sign boundary annotations, meaning researchers know exactly where one digit sign ends and the next begins. This annotation scheme allows the systematic evaluation of two fundamentally different recognition paradigms within a single benchmark: segment-based approaches that classify each pre-segmented sign individually, and sequence-based approaches that must infer the digit string from the unbroken sensor stream. The researchers also reserved unseen six-digit sequences specifically to test whether models trained on shorter sequences could generalize to longer, previously unencountered combinations—a rigorous test of true sequence understanding rather than memorization.

The choice of inertial sensing is deliberate and consequential. Unlike video, IMU data contains no images of the signer, so privacy is preserved by design; there is nothing to identify a face, a location, or a bystander. The data volume is also dramatically smaller. The authors calculate that, normalized by the number of represented classes, ISL-MDN requires more than fifty times less storage per class than comparable video-based datasets. That compactness matters enormously for edge AI, where memory, bandwidth, and battery budgets are tight. A glove with six tiny motion sensors is cheap, unobtrusive, and works identically in a dark room or bright sunlight—conditions that routinely defeat camera systems.

On the algorithmic side, the team benchmarked architectures spanning both recognition paradigms. For segment-based recognition, they built a ResNet convolutional baseline and a more sophisticated model they call the Transformer Core Method, or TCM. TCM exploits the transformer architecture’s celebrated attention mechanism in two directions at once: it models dependencies between the six sensor channels within each IMU frame, capturing how the motion of one part of the hand relates to another at a single instant, and it models temporal relationships across the segmented sign units, capturing how the hand’s configuration evolves from the start to the end of a digit. This dual attention lets the network weigh which sensor channels and which moments in time carry the most discriminative information for distinguishing, say, a signed four from a signed five.

Segment-based recognition, however, sidesteps the hardest part of the problem: in the real world, nobody tells the system where one sign ends and the next begins. For sequence-based recognition, the authors proposed their centerpiece, the Hierarchical Temporal Compression Transformer, or HTCT. Continuous sign language recognition from raw sensor streams is notoriously difficult because the model must implicitly align an input sequence of arbitrary length with an output label sequence of unknown length, without frame-level supervision. HTCT tackles this with a training objective based on Connectionist Temporal Classification, or CTC, a technique originally developed for speech recognition that allows the network to learn the alignment itself by summing over all possible alignments between inputs and labels.

The truly novel ingredient in HTCT is its hierarchical temporal compression. Raw IMU streams at 100 hertz are highly redundant: consecutive samples carry largely overlapping information, and the discriminative structure of a signed digit unfolds over hundreds of milliseconds, not milliseconds. HTCT combines a fixed Haar wavelet decomposition—a classical signal processing tool that decomposes a signal into coarse trends and fine details at multiple scales—with a learnable, causal temporal compression module. The wavelet stage provides a principled multi-scale representation, capturing both the rapid transients of a hand flick and the slower posture changes that define a digit, while the learnable compression stage is trained to discard redundant temporal information and retain only the features that help the CTC objective. The result is a much shorter, information-dense sequence that the transformer can attend over efficiently, improving both accuracy and speed.

The numbers tell a compelling story. The ResNet baseline achieved a Word Error Rate of 16.3 percent, where a word corresponds to a signed digit and the error rate measures how often the recognized digit sequence differs from the ground truth. TCM improved on that with 14.6 percent, confirming the value of modeling cross-channel dependencies with attention. But HTCT left both far behind, reaching a Word Error Rate of just 5.2 percent—roughly a threefold reduction in errors relative to the baseline. Even more remarkable is what that accuracy costs computationally: only 1.73 million floating-point operations per inference. For context, that is orders of magnitude below what typical video-based sign language transformers require, placing HTCT squarely in the territory of microcontrollers and low-power wearable processors. The authors note that the model achieves an effective accuracy–efficiency trade-off through low computational cost and inference latency, while candidly acknowledging that power consumption was not directly measured in the study.

The generalization test adds further weight. When confronted with six-digit sequences—longer than anything seen during training—the system’s ability to recognize previously unseen combinations demonstrates that HTCT is learning compositional structure rather than memorizing fixed-length patterns. This matters because real-world usage is inherently open-ended: a signer might need to communicate a bank account number, a serial code, or a date of arbitrary length. A recognizer that only works for the sequence lengths in its training set would be of limited practical value; one that composes digit signs flexibly can scale to the task at hand.

The release of ISL-MDN as an open resource, hosted on Zenodo with source files and documentation on GitHub, may prove as influential as the model itself. Inertial sensing remains relatively underexplored for continuous sign language recognition compared with vision, and one persistent obstacle has been the scarcity of annotated, signer-diverse datasets. By providing boundary annotations, multiple sequence lengths, held-out generalization data, and forty-one participants, ISL-MDN gives the community a unified benchmark on which segment-based and sequence-based methods can be compared fairly. It also extends coverage to Iranian Sign Language, a language underrepresented in the predominantly English- and Chinese-centric sign language recognition literature, and the data collection followed institutional ethical guidelines with informed consent and full anonymization of participant data.

The broader significance lies in what this work suggests about the future of assistive wearables. Sign language recognition has long been framed as a computer vision problem, but the wrist and fingers are where the signal actually lives, and miniature motion sensors can capture it directly, privately, and cheaply. Combined with aggressively compressed transformer architectures like HTCT, the vision of a glove that translates continuous signing into text entirely on-device—no cloud, no camera, no privacy trade-off—moves from aspiration toward engineering reality. Challenges remain, including scaling beyond digits to full vocabularies, testing across more sign languages, and validating battery life in the field. But with a 5.2 percent error rate at 1.73 million FLOPs, this study makes a persuasive case that the next breakthrough in accessible communication may come not from bigger models watching us, but from smaller ones riding along on our hands.

Subject of Research: Inertial-sensor-based continuous sign language number recognition using transformer models for edge AI

Article Title: Inertial-sensor–driven continuous multi-digit sign language number recognition using hierarchical temporal compression and transformer-based methods for edge AI

Article References: Parvizi, A., Jamshidi, K., & Shahbazi, H. (2026). Inertial-sensor–driven continuous multi-digit sign language number recognition using hierarchical temporal compression and transformer-based methods for edge AI. Multimedia Tools and Applications, 85(10), Article 792. https://doi.org/10.1007/s11042-026-21947-7

Image Credits: AI Generated

DOI: 10.1007/s11042-026-21947-7

Keywords: sign language recognition, inertial sensors, wearable technology, transformer, edge AI, IMU, CTC, wavelet decomposition, Iranian Sign Language, dataset, deep learning, accessibility