Assessing human muscle fatigue is critical for optimizing performance and safety in physical human-robot interaction(pHRI). This work presents a data-driven framework to estimate fatigue in dynamic, cyclic pHRI using arm-mounted surface electromyography(sEMG). Subject-specific machine-learning regression models(Random Forest, XGBoost, and Linear Regression predict the fraction of cycles to fatigue(FCF) from three frequency-domain and one time-domain EMG features, and are benchmarked against a convolutional neural network(CNN) that ingests spectrograms of filtered EMG. Framing fatigue estimation as regression (rather than classification) captures continuous progression toward fatigue, supporting earlier detection, timely intervention, and adaptive robot control. In experiments with ten participants, a collaborative robot under admittance control guided repetitive lateral (left-right) end-effector motions until muscular fatigue. Average FCF RMSE across participants was 20.8+/-4.3% for the CNN, 23.3+/-3.8% for Random Forest, 24.8+/-4.5% for XGBoost, and 26.9+/-6.1% for Linear Regression. To probe cross-task generalization, one participant additionally performed unseen vertical (up-down) and circular repetitions; models trained only on lateral data were tested directly and largely retained accuracy, indicating robustness to changes in movement direction, arm kinematics, and muscle recruitment, while Linear Regression deteriorated. Overall, the study shows that both feature-based ML and spectrogram-based DL can estimate remaining work capacity during repetitive pHRI, with the CNN delivering the lowest error and the tree-based models close behind. The reported transfer to new motion patterns suggests potential for practical fatigue monitoring without retraining for every task, improving operator protection and enabling fatigue-aware shared autonomy, for safer fatigue-adaptive pHRI control.
The field of physical human-robot interaction (pHRI) has grown significantly during the last two decades. Since the human is an integral part of pHRI, improving human comfort and ergonomics is important not only for maximizing productivity and efficiency in collaborative tasks [1], [2] but Fig. 1. Our approach to estimating human muscular fatigue in cyclic pHRI tasks employs learning-based regression models that predict the fraction of cycles to fatigue (FCF) from sEMG measurements. also for reducing safety risks to the human operator [3], including the potential adverse effects of muscle fatigue [4], [5], which can impair performance and increase injury risk over time. Fatigue is a physiological state that takes place after prolonged/accumulated high effort. If the progression of human effort is estimated during a pHRI task, the robot's contribution can be dynamically adjusted to reduce the load on the human and prevent the occurrence of muscle fatigue [6]. For example, in the manufacturing domain, estimating the progression of human effort helps to evaluate the difficulty level of the collaborative task and its sub-tasks to make changes toward reducing potential musculoskeletal injuries. Fatigue estimation is also essential in rehabilitation and sports training to prevent overexertion, tailor exercise intensity, and ensure optimal performance and safe recovery. For example, robotic rehabilitation for post-stroke patients often involves repetitive, rhythmic movements designed to help restore damaged neural pathways [7]. However, the onset of fatigue can interrupt these training sessions and restrict the overall duration of therapy. Therefore, finding a systematic and accurate approach to estimating the progression of human effort towards fatigue is considered to be a knowledge gap in the state-of-the-art.
In some of the earlier pHRI studies, human muscular effort has been indirectly inferred from the force applied by the user, which is typically acquired by a force sensor. Nonetheless, measuring the force alone often does not provide a complete assessment of human muscular effort in various scenarios. For instance, in collaborative drilling [2], [8], [9] and collaborative surface polishing [5], [10], the magnitude of the normal force applied by the human operator to the workpiece may remain constant, yet muscle fatigue could still occur due to prolonged effort. In both cases, the muscle activation can vary over time due to the dynamic nature of the task. Hence, not just the force applied by the human, but muscular effort should also be assessed by monitoring muscle activity, which is typically achieved by using electromyography (EMG) sensors [5], [8]. EMG sensors detect the electrical signals generated by muscle contractions and can provide information about the intensity and timing of muscle activation. Prolonged efforts in pHRI may lead to a progressive decline in a muscle’s ability to sustain performance as observed by the changes in EMG signals, resulting in muscle fatigue, which should be avoided to reduce musculoskeletal injuries and potential disorders.
De Luca [11] identified frequency-based features such as mean power frequency (MNF) and median power frequency (MDF) as reliable metrics for assessing muscle fatigue based on surface EMG measurements. As muscle fatigue occurs, the power spectrum shifts to lower frequencies, and hence, the MNF and MDF decrease gradually. It has also been reported that the RMS amplitude of the EMG signal increases towards the onset of fatigue.
Although these features are effective indicators of fatigue in static loading tasks (as in holding a heavy object in the mid-air) involving isometric muscle contractions, they are less pronounced in dynamic loading tasks (as in repeatedly moving a heavy object up and down), involving isotonic contractions [12], [13]. In a dynamic task, muscles cyclically shorten (concentric phase) and lengthen (eccentric phase) while generating force, promoting intermittent relaxation and improved blood circulation, which helps delay fatigue compared to static loading [13]- [15]. Fernando et al. [16] observed that the ratio of MNF to the average rectified value (ARV) provides a more accurate measure of fatigue in dynamic tasks. However, their results were not consistent across subjects. Zahedi et al. [6] relied on the integral of the normalized EMG signal to estimate overall human effort in a dynamic reaching task performed with a robot. They adjusted the robot’s damping based on the effort required, ensuring that the subjects did not experience muscle fatigue. Peternel et al. [4] incorporated EMG measurements into a first-degree ordinary differential equation (ODE) to estimate muscle fatigue in a cyclic sawing task performed with a robot. However, this approach does not account for muscle recovery, which was later investigated by Lorenzini et al. [17]. To incorporate the effect of dynamic muscle contractions, including both concentric (shorte
This content is AI-processed based on open access ArXiv data.