Therapeutic and assistive exoskeletons and exosuits show promise in both clinical and real-world settings. Improving their autonomy can enhance usability, effectiveness, and cost efficiency. This Review presents a generic control framework for autonomous operation of upper and lower limb devices and reviews current advancements and future directions. We highlight how data-driven machine learning aids in intention recognition, synchronization, patient assessment, and task-agnostic control. In addition, we discuss how reinforcement learning optimizes control policies through digital human twins and how generative AI supports therapy planning and patient engagement. Richer patient-specific data and more accurate digital twins are needed for clinical validation and widespread deployment.

Objective: Accurate estimation of stiffness across anatomical levels (i.e., joint, muscle, and tendon) in vivo has long been a challenge in biomechanics. Recent advances in electromyography (EMG)-driven musculoskeletal modeling have allowed the non-invasive estimation of stiffness during dynamic joint rotations. Nevertheless, validation has been limited to the joint level due to a lack of simultaneous in vivo experimental measurements of muscle and tendon stiffness. Methods: With a focus on the triceps surae, we employed a novel perturbation-based experimental technique informed by dynamometry and ultrasonography to derive reference stiffness at the joint, muscle, and tendon levels simultaneously. Here, we propose a new EMG-driven model-based approach that does not require external joint perturbation, nor ultrasonography, to estimate multi-level stiffness. We present a novel set of closed-form equations that enables the person-specific tuning of musculoskeletal parameters dictating biological stiffness, including passive force-length relationships in modeled muscles and tendons. Results: Calibrated EMG-driven musculoskeletal models estimated the reference data with average normalized root-mean-square error ≈ 20%. Moreover, only when calibrated tendons were approximately four times more compliant than typically modeled, our approach could estimate multi-level reference stiffness. Conclusion: EMG-driven musculoskeletal models can be calibrated on a larger set of reference data to provide more realistic values for the biomechanical variables across multiple anatomical levels. Moreover, the tendon models that are typically used in musculoskeletal modeling are too stiff. Significance: Calibrated musculoskeletal models informed by experimental measurements give access to an augmented range of biomechanical variables that might not be easily measured with sensors alone.

The advancement and development of human modeling have greatly benefited from principles used in robotics, for instance, multibody dynamics laid the foundations for physics engines of human movement simulation, and the robotics and control theory were used to contextualize human sensorimotor control. There are many common interests and interconnections between the fields of human modeling and robotics. In recent years, as robots have become safer and smarter, they actively participate in our lives and help us in various scenarios. Roboticists need tools and data from human modeling to build next-generation robots that better assist humans. In this survey, we focus on the connections between physical human-robot interaction and human modeling. On one hand, human neuromusculoskeletal and sensorimotor control models provide novel insights into the human response that robots can utilize to improve human performance. On the other hand, robots are becoming instrumental in quantifying the performance of the (neuro)musculoskeletal system. Thus, the combined use of human modeling and robotic methods in physical human-robot interaction can lead to both improved human understanding and functional assistance.

The simultaneous modulation of joint torque and stiffness enables humans to perform large repertoires of movements, while versatilely adapting to external mechanical demands. Multi-muscle force control is key for joint torque and stiffness modulation. However, the inability to directly measure muscle force in the intact moving human prevents understanding how muscle force causally links to joint torque and stiffness. Joint stiffness is predominantly estimated via joint perturbation-based experiments in combination with system identification techniques. However, these techniques provide joint-level stiffness estimations with no causal link to the underlying muscle forces. Moreover, the need for joint perturbations limits the generalizability and applicability to study natural movements. Here, we present an electromyography (EMG)-driven musculoskeletal modeling framework that can be calibrated to match reference joint torque and stiffness profiles simultaneously via a multi-term objective function. EMG-driven models calibrated on <2 s of reference torque and stiffness data could blindly estimate reference profiles across 100 s of data not used for calibration. Model calibrations using an objective function comprising torque and stiffness terms always provided less feasible solutions than an objective function comprising solely a torque term, thereby reducing the space of feasible muscle–tendon parameters. Results also showed the proposed framework's ability to estimate joint stiffness in unperturbed conditions, while capturing differences against stiffness profiles derived during perturbed conditions. The proposed framework may provide new ways for studying causal relationships between muscle force and joint torque and stiffness during movements in interaction with the environment, with broad implications across biomechanics, rehabilitation and robotics.

Low back joint compression forces have been linked to the development of chronic back pain. Back-support exoskeletons controllers based on low back compression force estimates could potentially reduce the incidence of chronic pain. However, progress has been hampered by the lack of robust and accurate methods for compression force estimation. Electromyography (EMG)-driven musculoskeletal models have been proposed to estimate lumbar compression forces. Nonetheless, they commonly underrepresented trunk musculoskeletal geometries or activation–contraction dynamics, preventing validation across large sets of conditions. Here, we develop and validate a subject-specific large-scale (238 muscle–tendon units) EMG-driven musculoskeletal model for the estimation of lumbosacral moments and compression forces, under eight box-lifting conditions. Ten participants performed symmetric and asymmetric box liftings under 5 and 15 kg weight conditions. EMG-driven model-based estimates of L5/S1 flexion–extension moments displayed high correlation, R² (mean range: 0.88–0.94), and root mean squared errors between 0.21 and 0.38 Nm/kg, with respect to reference inverse dynamics moments. Model-derived muscle forces were utilized to compute lumbosacral compression forces, which reached eight times participants body weight in 15 kg liftings. For conditions involving stooped postures, model-based analyses revealed a predominant decrease in peak lumbar EMG amplitude during the lowering phase of liftings, which did not translate into a decrease in muscle–tendon forces. During eccentric contraction (box-lowering), our model employed the muscle force–velocity relationship to preserve muscle force despite significant EMG reduction. Our modeling methodology can inherently account for EMG-to-force non-linearities across subjects and lifting conditions, a crucial requirement for robust real-time control of back-support exoskeletons.

In vivo joint stiffness estimation during time-varying conditions remains an open challenge. Multiple communities, e.g. system identification and biomechanics, have tackled the problem from different perspectives and using different methods, each of which entailing advantages and limitations, often complementary. System identification formulations provide data-driven estimates of stiffness at the joint level, while biomechanics often relies on musculoskeletal models to estimate stiffness at multiple levels, i.e. joint, muscle, and tendon. Collaboration across these two scientific communities seems to be a logical step toward a reliable multi-level understanding of joint stiffness. However, differences at the theoretical, computational, and experimental levels have limited inter-community interaction. In this article we present a roadmap to achieve a unified framework for the estimation of time-varying stiffness in the composite human neuromusculoskeletal system during movement. We present our perspective on future developments to obtain data-driven system identification and musculoskeletal models that are compatible at the theoretical, computational, and experimental levels. Moreover, we propose a novel combined closed-loop paradigm, in which reference estimates of joint stiffness via system identification are decomposed into underlying muscle and tendon contribution via high-density-electromyography-driven musculoskeletal modeling. We highlight the need for aligning experimental requirements to be able to compare both joint stiffness formulations. Unifying both biomechanics' and system identification's formulations is a necessary step for truly generalizing stiffness estimation across individuals, movement conditions, training and impairment levels. From an application point of view, this is central for enabling patient-specific neurorehabilitation therapies, as well as biomimetic control of assistive robotic technologies. The roadmap we propose could serve as an inspiration for future collaborations across broadly different scientific communities to truly understand joint stiffness bio- and neuromechanics.

With recent improvements in healthcare, individuals with Duchenne muscular dystrophy (DMD) have prolonged life expectancy, and it is therefore vital to preserve their independence. Hand function plays a central role in maintaining independence in daily living. This requires sufficient grip force and the ability to modulate it with no substantially added effort. Individuals with DMD have low residual grip force and its modulation is challenging and fatiguing. To assist their hand function, we developed a novel dynamic hand orthosis called SymbiHand, where the user's hand motor intention is decoded by means of surface electromyography, enabling the control of an electrohydraulic pump for actuation. Mechanical work is transported using hydraulic transmission and flexible structures to redirect interaction forces, enhancing comfort by minimizing shear forces. This paper outlines SymbiHand's design and control, and a case study with an individual with DMD. Results show that SymbiHand increased the participant's maximum grasping force from 2.4 to 8 N. During a grasping force-tracking task, muscular activation was decreased by more than 40% without compromising task performance. These results suggest that SymbiHand has the potential to decrease muscular activation and increase grasping force for individuals with DMD, adding to the hand a total mass of no more than 241 g. Changes in mass distributions and an active thumb support are necessary for improved usability, in addition to larger-scale studies for generalizing its assistive potential.

An important goal in the design of next-generation exoskeletons and limb prostheses is to replicate human limb dynamics. Joint impedance determines the dynamic relation between joint displacement and torque. Joint stiffness is the position-dependent component of joint impedance and is key in postural control and movement. However, the mechanisms to modulate joint stiffness are not fully understood yet. The goal of this study is to conduct a systematic analysis on how humans modulate ankle stiffness. Time-varying stiffness was estimated for six healthy subjects under isometric, as well as quick and slow dynamic conditions via system identification techniques; specifically, an ensemble-based algorithm using short segments of ankle torque and position recordings. Our results show that stiffness had the lowest magnitude under quick dynamic conditions. Under isometric conditions, with fixed position and varying muscle activity, stiffness exhibited a higher magnitude. Finally, under slow dynamic conditions, stiffness was found to be the highest. Our results highlight, for the first time, the variability in stiffness modulation strategies across conditions, especially across movement velocity.

Joint stiffness estimation under dynamic conditions still remains a challenge. Current stiffness estimation methods often rely on the external perturbation of the joint. In this study, a novel 'perturbation-free' stiffness estimation method via electromyography (EMG)-driven musculoskeletal modeling was validated for the first time against system identification techniques. EMG signals, motion capture, and dynamic data of the ankle joint were collected in an experimental setup to study the ankle joint stiffness in a controlled way, i.e. at a movement frequency of 0.6 Hz as well as in the presence and absence of external perturbations. The model-based joint stiffness estimates were comparable to system identification techniques. The ability to estimate joint stiffness at any instant of time, with no need to apply joint perturbations, might help to fill the gap of knowledge between the neural and the muscular systems and enable the subsequent development of tailored neurorehabilitation therapies and biomimetic prostheses and orthoses.