A prosthetic hand with human-like grasping behavior outperforms what is required in the execution of activities of daily living.

Developing a prosthetic system that emulates the complexity of the human upper limb is a formidable challenge. Unfortunately, abandonment rates for such devices remain high, primarily due to the limited intuitiveness of control and poor dexterity. Specifically, inadequate wrist mobility, i.e., the absence of actively controllable flexion-extension and pronation-supination degrees of freedom, often results in subpar dexterity in upper limb prostheses. This work introduces an anthropomorphic wrist prosthesis featuring active flexion-extension and pronation-supination capabilities, integrated with the poly-articulated Hannes hand. The central focus of this study is to compare the functionality of this prosthetic system with the natural wrist movement of healthy participants, demonstrating that the biomechanical range of motion falls within that of the mechatronic system. The overarching goal is to improve the performance of trans-radial prostheses by enhancing their dexterity and overall functionality. Our preliminary findings from healthy subjects demonstrate that the incorporation of a 2 Degrees-of-Freedom active biomimetic wrist into the prosthesis can approximate human-like capabilities in upper limb prostheses. Moreover, the resulting development confirm its enhanced dexterity when operated by amputees. These results provide valuable insights into the potential applications of this technology for amputees, offering a basis for future investigations.

Noninvasive human-machine interfaces such as surface electromyography (sEMG) have long been employed for controlling robotic prostheses. However, classical controllers are limited to few degrees of freedom (DoF). More recently, machine learning methods have been proposed to learn personalized controllers from user data. While promising, they often suffer from distribution shift during long-term usage, requiring costly model re-training. Moreover, most prosthetic sEMG sensors have low spatial density, which limits accuracy and the number of controllable motions. In this work, we address both challenges by introducing a novel myoelectric prosthetic system integrating a high density-sEMG (HD-sEMG) setup and incremental learning methods to accurately control 7 motions of the Hannes prosthesis. First, we present a newly designed, compact HD-sEMG interface equipped with 64 dry electrodes positioned over the forearm. Then, we introduce an efficient incremental learning system enabling model adaptation on a stream of data. We thoroughly analyze multiple learning algorithms across 7 subjects, including one with limb absence, and 6 sessions held in different days covering an extended period of several months. The size and time span of the collected data represent a relevant contribution for studying long-term myocontrol performance. Therefore, we release the DELTA dataset together with our experimental code.

This paper presents a narrative review of incremental learning methods for myoelectric control, outlining both the historical trajectory and potential of adaptive prosthetic systems. Traditional myoelectric control has evolved from direct control techniques to advanced pattern recognition, yet persistent challenges such as signal non-stationarities and, consequently, the need for frequent recalibration remain. Incremental learning may enable a paradigm shift by continuously updating control models based on real-time, user-in-the-loop data, thereby addressing user-specific variations, environmental changes, and challenges from screen-guided-training based calibration. A central contribution of the paper is its taxonomy of incremental learning strategies, which divides the field into four categories: dedicated on-demand recalibration, unsupervised incremental learning, predictor-dependent incremental learning, and environment-dependent incremental learning. The methodology, strengths, and limitations of each category are discussed, providing a clear framework for evaluating current research and guiding future innovations. Further, this work establishes three settings for incremental learning: domain-incremental, task-incremental, and class-incremental continual learning. In addition, the paper highlights emerging trends such as transfer learning, domain adaptation, and self-supervised regression. It also emphasizes the potential of physiologically-inspired algorithms, novel end-effector designs to enhance prosthetic performance, and human-device co-adaptation. Finally, this paper discusses open challenges for incremental learning like attribution of signal changes to noise vs. behaviours, model complexity vs. data requirements, and user vs. model adaptation. Collectively, these insights pave the way for next-generation myoelectric systems that are more robust, intuitive, and adaptable to the dynamic needs and behaviours of users.

Upper-limb prosthesis control remains challenging in achieving natural and intuitive movements, especially for devices with multiple actuated degrees of freedom (DoFs), often demanding high cognitive effort. Machine learning aids in mapping phantom limb muscle patterns to prosthetic movements, but is limited by the instability of electromyographic signals over time. This study investigates two simultaneous and proportional myocontrol strategies, based on position and velocity, using incremental learning for a 3-DoFs prosthesis, allowing co-adaptation between the system and the user. Six able-bodied and five limb-difference participants performed Target Achievement Control tests over four sessions per control strategy, assessing performance, usability, workload, simultaneity, and proportionality. Results indicate that velocity control consistently outperforms position control in both populations, yielding lower errors, higher success rates and path efficiency, and lower workload. Notably, both control strategies showed significant improvement over time in the able-bodied group, while only position control improved significantly in the limb-difference group. Interestingly, no significant difference in usability was observed between the two strategies in either group. Position control promoted greater simultaneous actuation of multi-DoFs. However, the overall findings support the use of velocity-based control as a means to improve prosthetic performance and user satisfaction.

Nowadays, upper limb prosthesis acceptance remains low due to ineffective control techniques and the relative training methods. However, an engaging training method applied during early prosthetic fitting seems to have a positive impact on acceptance. To this aim, in this paper we present a Mixed Reality (MR) pre-fitting training system based on Microsoft HoloLens2, designed for users of the Hannes prosthetic hand. The system allows individuals with varying stump morphologies to control a holographic Hannes hand through a low-latency interface, replicating the physical architecture and motion of the real device in an immersive, portable environment. This pilot study presents the preliminary evaluation of the MR framework conducted with two limb difference participants (one naïve user and one experienced user in myoelectric prosthesis control) performing a novel bimanual Target Achievement Control (TAC) test. Their performance in controlling the real Hannes device was subsequently compared to a reference group of eight transradial limb difference individuals. Both MR-trained participants showed faster learning and better adaptation to the real prosthesis than the reference group. These preliminary findings suggest that the proposed MR framework, combined with an MR bimanual TAC test, may improve learning and user experience, ultimately contributing to lower prosthesis abandonment rates.