Integrating computer vision into powered prosthetic devices is garnering attention as a potential avenue for enhancing functionality. These methodologies have been introduced to address challenges such as prosthetic abandonment, which can be attributed to the unreliability and lack of intuitiveness of myoelectric control. This paper provides insights into the current state, challenges, and future directions of integrating computer vision into prosthetic limb control systems. To this aim, a literature review was conducted, identifying 50 relevant studies (33 on upper-limb and 17 on lower-limb prostheses), sourced from Scopus, PubMed, and IEEE Xplore, with the search updated to May 2025. The hardware implementation aspects, including camera sensor positioning and computational units, were synthesized, as well as the computer vision approaches implemented with a specific emphasis on their implications for limb functional performance. The review identified strengths and suggested areas for improvement, emphasizing the need for studies on optimal camera placement and the feasibility of embedded computational units for real-world testing. Even if the reported results are promising, the performed analysis outlined the need to conduct usability studies and introduce vision-based approaches into lower-limb prosthesis control, which should be tested in real-world scenarios. Validation in real settings is crucial for these technologies to move beyond laboratory settings. • Literature review of vision-based control strategies for prosthetic limbs. • Identify gaps in real-world testing and multimodal integration for prosthetic control. • Guide future research on vision-based strategies for prosthetic control.

Work-related musculoskeletal disorders represent one main contributor to production workers absenteeism. In industry 5.0, exoskeletons have been proposed to mitigate risks of injury by supporting workers during repetitive tasks, with surface Electromyography (sEMG) showcasing their effects. Although existing studies have primarily evaluated exoskeletons by comparing muscle activity with and without the device, a systematic investigation of which sEMG features most effectively reflect muscle fatigue during prolonged arm elevation is still missing. This study aims to evaluate the effectiveness of different sEMG features in estimating perceived physical effort during an overhead bolting-unbolting task, performed by 10 participants both with and without a passive shoulder exoskeleton. It further assesses how the use of the exoskeleton influences the fatigue-related metrics identified. sEMG signals were collected from seven bilateral muscle groups, and 15 time-, frequency-, and spatial-domain features were extracted and correlated with two subjective effort perception models. Our results highlight strong correlations between time-domain features and perceived physical effort. Moreover, comparisons between conditions demonstrate that the exoskeleton provides a measurable fatigue-reducing effect. In contrast, spatial-domain features showed weak associations with perceived effort, suggesting limited suitability for low-intensity, long-duration tasks. These findings contribute to identifying the most informative sEMG features for fatigue estimation and provide evidence of the benefits of passive exoskeletons in industrial scenarios.

Rehabilitative therapies play a crucial role in upper limb motor recovery, as upper limbs are the most active parts in executing the activities of daily living. Because of a huge number of people with motor disorders and a shortage of therapists, the integration of data-driven AI methodologies and robots for rehabilitation could be helpful in creating personalized and challenging therapies, leading to a myriad of benefits for both patients and therapists. AI methods can be implemented in different functional modules of the robotic platform, such as user intention recognition, robot motion planning, robot interaction control, and system adaptation through different learning paradigms. This article presents a systematic literature review on the use of data-driven learning methods applied in upper limb robot-aided rehabilitation. The analysis is structured around the learning paradigms adopted, namely, supervised, unsupervised, and reinforcement learning, as well as the corresponding task types (e.g., classification, regression, and control tasks) and model types, distinguishing between machine learning and deep learning approaches. The review reveals that most studies employ supervised learning to address classification tasks, and that deep learning models are the most frequently adopted.

Social robots are increasingly being explored as interactive coaches capable of delivering personalized physical and cognitive training sessions. Improving their effectiveness entails personalized interventions through adaptive robotic systems with continuous workload quantification. This study presents a workload estimation methodology based on physiological and kinematic monitoring, designed for integration into a social robotic coach. Physical, mental, and dual-task activities were administered to 15 healthy participants, and Support Vector Regression was used to model their perceived workload levels. Physical workload was estimated with a mean absolute error (MAE) of 0.12 ± 0.01 and a correlation of 0.75 ± 0.02, demonstrating high reliability across conditions. Mental workload estimation, however, showed greater variability (MAE: 0.18 ± 0.01, correlation: 0.62 ± 0.03), particularly in cognitively demanding and high-intensity tasks. This is likely due to overlapping physiological responses to cognitive and physical demands, which introduce ambiguity in signal interpretation. The continuous workload estimation provided by the model can be leveraged to define thresholds offering a discrete interpretation of workload levels.

Rehabilitation robotics enables consistent and personalized therapy but still relies on complex, expert-driven tuning of control parameters. To address this, a reinforcement learning strategy based on Q-learning is proposed to autonomously adapt key parameters during upper-limb rehabilitation, without requiring prior task-specific knowledge. A systematic evaluation is conducted across combinations of control parameters (radial stiffness and execution time), performance-based reward functions (pointing accuracy and movement smoothness), and exploration strategies ( ɛ ɛ - greedy and Upper Confidence Bound). The Q-learning agent selects discrete actions (increase, decrease, or maintain the current value) for each control parameter, enabling real-time adaptation based on observed performance. The method is validated using a Kuka robotic arm in experiments involving 16 right-handed healthy subjects (13 males, 3 females) and 8 right-handed individuals simulating impaired motor behavior (5 males, 3 females). Motion signals are acquired through internal robot sensors, while a wearable physiological monitoring system define the Q-learning agent state. Reward improvement and exploration ratio are analyzed as key performance indicators and statistically compared across all tested conditions using the Mann–Whitney test. The results demonstrate that the proposed algorithm effectively adjusts control parameters online, with performance influenced by the reward function, exploration strategy, and selected control actions. Reward improvements of 0 . 11 ± 0 . 09 ( ɛ ɛ - greedy , reward based on pointing ability) and 0 . 13 ± 0 . 11 (reward based on smoothness, Upper Confidence Bound strategy) were observed in healthy subjects, indicating enhancements in pointing accuracy and movement smoothness. In simulated pathological cases, improvements of 0 . 08 ± 0 . 13 and 0 . 06 ± 0 . 16 were observed, respectively.

Robot-aided rehabilitation is pushing forward novel robotic architectures to provide physical therapy. This paper presents a patient-tailored control architecture for upper-limb robot-aided orthopaedic rehabilitation capable of i) adapting the robot workspace on the basis of patient Range of Motion (RoM); ii) generating a tunnel, around the desired path to be followed by the patient, which guarantees spatial autonomy; iii) introducing a back-wall inside the tunnel sliding with variable speed on the basis of patient performance to ensure temporal autonomy; iv) rehabilitating to working gestures, thanks to a DMP-based trajectory planner, with the aim of favoring an effective translation of the patient's motor recovery results to the occupational sphere; v) ensuring a patient-tailored assistance also thanks to the evaluation of performance indicators. The designed system was validated demonstrating the adaptability of the system to orthopaedic patient motor imnrovements.