A Prosthetic Hand System by Contralateral-Collaborative Control Based on Multi-task Learning

Upper-limb amputees often experience difficulty achieving efficient bimanual coordination due to the limited flexibility and adaptability of conventional prosthetic systems, which restricts their self-care capabilities and overall quality of life in daily activities. To address this issue, this study proposes a novel contralateral cooperative control system that utilizes surface electromyography signals from the intact limb to drive the prosthetic hand on the affected side, thereby enabling coordinated execution of both symmetric and auxiliary gestures. A hybrid multi-task learning model combining a convolutional neural network (CNN) and a Transformer architecture was innovatively developed, in which the CNN extracts local features while the Transformer, equipped with multi-head self-attention, captures global dependencies, enabling simultaneous multi-task recognition of four gestures and three force levels and substantially improving prosthetic operational flexibility. Experimental results demonstrate offline recognition accuracies of 96.12% for gestures and 94.57% for force intensity. In online control tests, the proposed system achieved a 93.34% success rate, with only a 1.13% performance gap between amputee and healthy participants, demonstrating excellent cross population consistency and task robustness. In terms of hardware design, the study drew inspiration from the coiling mechanism and morphological characteristics of the chameleon's tail, significantly enhancing the grasping performance of the prosthetic fingers. The redesigned fingers maintain anti-torsional grasping stability under high loads and provide reliable anti-slip grip for small objects during low-force precision operations. Experimental results confirm substantial improvements in adaptability, stability, and dexterity of the prosthetic fingers across diverse scenarios. Overall, the proposed prosthetic hand system seamlessly integrates contralateral-cooperative control, multimodal feature decoding, and bioinspired structural design, thereby enhancing user autonomy and quality of life while offering an innovative solution for natural, flexible, and coordinated intelligent prosthetic operation.