This research presents the results of conceptual design and modeling of an exoskeleton. It is intended for ankle joint rehabilitation in patients with musculoskeletal disorders. The exoskeleton design includes three screw actuators that smoothly control motion in the planes of dorsal and plantar flexion, inversion, and eversion. The results of the virtual tests performed on the exoskeleton device demonstrated a high degree of adaptability to varying loads and different phases of motion. Controlled torque fluctuations and linear motion provide the necessary support during different phases of rehabilitation, which has a positive impact on the patient’s recovery rate. The advantages of the design include material availability, ease of use, and flexibility in customization, making it an attractive option for use in both clinical and home settings. The study emphasizes the importance of developing affordable and accurate rehabilitation devices that can adapt to individual patient needs.

Today, there is a high demand for remote rehabilitation using mobile robotic complexes all over the world. They offer a wide range of options for convenient and effective therapy at home to patients and the elderly, especially those bedridden after musculoskeletal injuries. In this case, modern approaches to the development of exoskeletons for the rehabilitation of the lower extremities are especially relevant for the effective restoration of lost motor functions. Taking into account the advantages and features of robotic rehabilitation, this work is devoted to the development of a prototype exoskeleton for the ankle joint and experimental studies of the remote control module. The proposed new exoskeleton prototype design was integrated with a mobile wireless communication platform, allowing remote control of the position of the exoskeleton foot using a remote control device. As a result of functional testing, the root mean square error (RMSE) was 23.9° for dorsiflexion/plantarflexion movements and 12.8° for inversion and eversion movements, as well as an average signal transmission delay of about 100 ms and packet loss of 0.6%. These results reflect the technical feasibility of remote control at a distance of up to 10 m. The developed system is mobile, autonomous, and easy to use, which confirms its suitability as a laboratory platform for functional verification and testing of module consistency.

Tendon-driven robotic fingers are underactuated and strongly nonlinear, and their unidirectional, friction-attenuated tendon tension imposes actuation constraints that any practical controller must respect. A previous study established that a deterministic learning controller (DDPG) can track a grasp trajectory for such a finger and compared it against model-based schemes. The present work extends that line of investigation by asking a narrower but unresolved question: for this same single-actuator, three-link tendon-driven architecture, how does a deterministic actor-critic policy (DDPG) compare with a stochastic, maximum-entropy policy (SAC) once a tension-limit penalty is built directly into the learning objective? Using the previously validated physics-based simulation model, both agents are trained under identical conditions with a reward that jointly penalizes tracking error and tension-constraint violation, and are evaluated across random, sinusoidal, and step disturbances using steady-state error, RMS, ISE, and IAE. The results show a consistent and interpretable trade-off: the stochastic SAC policy attains lower tracking error and better disturbance rejection, especially under time-varying and abrupt loads, at the cost of noisier control and longer training, whereas the deterministic DDPG policy converges faster and produces smoother torque and tension profiles but adapts less well to coupled disturbances. The study isolates the role of the exploration mechanism and the tension penalty in shaping these behaviors and offers practical guidance on selecting between deterministic and stochastic actor-critic methods for physically constrained, underactuated tendon-driven systems.

This paper presents the development of a hardware architecture for a control system of an upper-limb bionic prosthesis based on surface electromyography (EMG). The proposed system is intended for acquisition, amplification, filtering, and digitization of bioelectrical signals generated by skeletal muscles to provide reliable input data for prosthetic control algorithms. The designed hardware architecture includes an instrumentation amplifier for acquiring low-amplitude EMG signals, an active common-mode interference suppression circuit to reduce external noise, an active band-pass filter with a passband of 60–500 Hz to isolate the informative frequency range of the EMG signal, an adjustable gain stage, and a high-resolution sigma–delta analog-to-digital converter (ADC) for signal digitization. The relevance of the proposed solution is driven by the increasing need for advanced prosthetic and rehabilitation technologies for civilian patients and military personnel with upper-limb amputations resulting from trauma and combat-related injuries. The developed hardware architecture provides a reliable platform for biosignal acquisition and processing in rehabilitation-oriented prosthetic systems, supporting restoration of motor functions and improvement of functional independence. To evaluate the performance of the proposed architecture, circuit modeling was performed in the Multisim simulation environment. Real EMG recordings obtained from the PhysioNet biomedical signal database were used as input signals. In addition, synthetic disturbances, including 50 Hz power-line interference and low-frequency respiratory artifacts, were introduced to simulate realistic operating conditions. The simulation results demonstrated effective suppression of undesirable harmonics while preserving informative spectral components of the EMG signal. A component-level analysis was conducted considering parameters such as input impedance, noise characteristics, power consumption, and physical dimensions. The developed architecture is characterized by modularity, adjustable amplification capability, energy efficiency, and scalability for multi-channel implementations. The proposed solution can be applied in development of autonomous upper-limb bionic prostheses for medical rehabilitation of civilian patients and military personnel and can serve as a hardware platform for integration of intelligent control algorithms in advanced prosthetic systems.

Traditional numerical methods for solving inverse kinematics in rehabilitation exoskeletons often suffer from high computational latency and struggle with dynamic torque optimization during high-frequency real-time control loops. Objective: Evaluation and testing a two-stage physics-informed feedforward neural network framework for solving the inverse kinematics of a 2-DOF upper-limb exoskeleton while simultaneously minimizing static gravitational joint torques. Methods. An off-line multi-start optimization routine was used to generate a workspace dataset of 99,208 optimal configurations; a feedforward architecture with 135,814 parameters incorporating a custom differentiable physical gravity layer was then trained to predict joint angles and correct residual loads. Results. Across a test set of 14,882 points, the model achieved a mean spatial positioning error of 0.385 cm (3.85 mm) with an approximate inference latency of 2.0 ms per sample, supporting closed-loop control frequencies up to 500 Hz without relying on resource-intensive online quadratic programming. Conclusions. The proposed physics-informed learning approach effectively combines high spatial accuracy, energy-efficient torque distribution, and deterministic execution speed, providing a lightweight, low-latency control solution for wearable upper-limb rehabilitation robotics.

Reliable control of upper-limb prostheses remains challenging because muscle signals can change between sessions and wearable EMG sensors may suffer from poor electrode contact. This study presents a real-time prosthetic hand control platform based on an OYMotion gForce Pro armband and a five-finger Dynamixel-actuated prosthetic hand. The system is designed to work without an external dataset, using only a short user-specific recording session and brief calibration before real-time use — specifically, five main training repetitions plus four supplementary calibration repetitions per gesture — intentionally minimising the calibration burden to enable practical single-session deployment. Five hand gestures—power grasp, open hand, OK gesture, index pointing, and thumbs-up—were classified using several machine-learning and ensemble-based methods. All models were evaluated under the same live data stream to ensure a fair comparison. Six recording sessions collected across five able-bodied participants on different recording days were included in the final statistical analysis. For every session, the main and supplementary training samples were combined to train session-specific models before live evaluation. Across six independent recording sessions from one able-bodied participant, Linear Discriminant Analysis achieved the best average live accuracy of 85.9%, followed closely by majority voting and adaptive ensemble methods. In contrast, the LSTM model showed lower performance, reaching 70.2% accuracy. The results indicate that, for low-data real-time prosthetic control, ensemble-based methods preserved near-best average accuracy while reducing reliance on any single classifier assumption, making them a more robust deployment choice than individual classifiers. Statistical analysis confirmed that session-to-session variability was the dominant factor in system performance (Kendall's W = 0.249) rather than classifier choice (W = 0.043), suggesting that practical reliability depends more on robust calibration and signal monitoring than on model selection.