Upper-limb amputation severely impairs grasping and manipulation abilities, and myoelectric prosthetic hands offer an important route for restoring hand function by decoding surface EMG signals into movement intentions. In recent years, prosthetic hand control has made progress in object grasping, multi-degree-of-freedom coordination, and some daily tasks, but tool handling in real-life and work settings remains much more challenging. Unlike simple grasping, tasks such as hammering, sawing, and peeling require the prosthetic hand to maintain stability under rigid impacts, variable loads, and continuous contact while adjusting grip force and finger posture in real time. Conventional fixed-force control lacks dynamic adaptability, whereas force-following strategies may suffer from delays or instability during fast impacts. "In contrast, the human hand can rapidly correct movements through a sensorimotor loop integrating tactile feedback, proprioception, and neuromuscular signals." said the author Boao Li, a researcher at Wuhan University of Science and Technology, "Therefore, safely and efficiently transferring human dynamic tool-use skills to myoelectric prosthetic hands has become a key challenge for moving intelligent prostheses from static grasping toward practical tool use and vocational rehabilitation."
This study proposed a myoelectric prosthetic control framework based on human hand skill transfer to improve stability and adaptability during dynamic tool handling. The researchers first asked able-bodied participants to wear a data glove integrated with tactile sensors, bend sensors, and EMG sensors, and collected contact forces, joint angles, and surface EMG signals during tool-use demonstrations such as hammering and sawing. These multimodal data were then used to train a tactile, kinesthetic, and EMG bionic gripping controller (TKE-BGC). The controller uses a Transformer encoder to fuse multimodal temporal information and a multilayer perceptron to predict the next joint-angle increment in real time, allowing the prosthetic hand to rapidly correct its grasp according to current contact force, posture, and user EMG intention. During online control, the system first identifies the user's tool-grasping intention through EMG pattern recognition, and then uses TKE-BGC to perform feedback-based adjustments during dynamic interaction. To evaluate the framework, the researchers recruited both able-bodied participants and transradial amputees to complete 4 tool-operation tasks, including hammering nails, sawing wood, peeling, and desktop organization, and compared the proposed method with conventional fixed-force and force-following control strategies.
The results showed that the proposed TKE-BGC control framework achieved stronger stability, accuracy, and generalization in dynamic tool handling. In offline prediction, the method more closely followed the joint-motion trajectories demonstrated by the human hand, especially in hammering tasks with stronger rigid impacts, where it significantly reduced prediction errors compared with fixed-force and force-following control. In online tests, TKE-BGC consistently reduced tool drops and shortened task completion time not only in seen tasks such as hammering nails and sawing wood, but also in unseen tasks including peeling and desktop organization. At the same time, the average contact forces generated by the proposed method were closer to those of natural human-hand operation, indicating more appropriate grip-force regulation that could prevent slippage while avoiding excessive force. EMG analysis further showed that users required less muscular effort when using this method; for example, in the hammering task, its average EMG amplitude was lower than that of both force-following and fixed-force strategies. User questionnaires also showed that amputee participants rated the method more favorably in terms of efficiency, reduced effort, and ease of use. Ablation experiments further confirmed the critical role of tactile information in tool manipulation, while the fusion of tactile, kinesthetic, and EMG signals produced the best prediction and control performance.
The significance of this work lies not only in improving the stability of myoelectric prosthetic hands during tool handling, but also in shifting prosthetic control from static grasping toward more complex dynamic manipulation in real-life and work settings. By integrating tactile, kinesthetic, and EMG information and learning manipulation skills from human demonstrations, TKE-BGC enables the prosthetic hand to make real-time feedback corrections under impacts, slippage, and changing loads, producing more human-like contact force distributions while reducing users' muscular workload. These findings suggest that truly practical intelligent prostheses should not rely only on movement-intention recognition, but should also incorporate multimodal adaptive control mechanisms similar to the human sensorimotor loop. At the same time, several limitations remain: the current tactile sensing is relatively sparse and does not yet capture richer tactile information such as vibration, temperature, or shear force; model training mainly relies on a single demonstrator, leaving personalized manipulation preferences insufficiently represented. "And future work will need higher-density tactile sensors, more comprehensive multi-degree-of-freedom motion modeling, and adaptive mapping methods to further improve dexterity, generalization, and clinical practicality in complex tool-use scenarios." said Boao Li.
Authors of the paper include Boao Li, Shuhui Wu, Ting You, Shixian Wang, Ziming Chen, Ye Liu, Di Guo, Fuchun Sun, Guangyuan Xu, Du Jiang, Gongfa Li, and Bin Fang.
This work was supported in part by the Brain Science and Brain-like Intelligence Technology–National Science and Technology Major Project (grant no. 2025ZD0215600) and National Natural Science Foundation of China under grant nos. 62573063 and 62536001.
The paper, "Dynamic Manipulation Skill Learning for Tactile Myoelectric Prosthetic Hands in Tool Handling" was published in the journal Cyborg and Bionic Systems on May 13, 2026, at https://doi.org/10.34133/cbsystems.0572.
