Abstract

The dataset is related to human–robot interfaces (HRIs) for robotic hand control based on regression of surface electromyography (sEMG) signals without requiring instant-by-instant labeling. It includes sEMG data acquired from healthy subjects performing different grasping motions (power, tripodal, and ulnar grasps), and is used to compare state-of-the-art unsupervised regression methods—namely Non-Negative Matrix Factorization (NMF) and autoencoders (AE)—with a novel weakly supervised approach based on neural networks trained using the soft-Dynamic Time Warping (soft-DTW) divergence. The dataset was collected during experimental sessions involving real-time and simulated control of an anthropomorphic robotic hand, enabling simultaneous and proportional grasp control from forearm sEMG signals. The data were produced in the framework of the Horizon Europe IntelliMan project and are presented in the following publication: R. Meattini, A. Bernardini, G. Palli, and C. Melchiorri, “Comparison of Weakly Supervised Regression of sEMG Signals With State-of-the-Art Unsupervised Methods for Robot Hand Control: A Pilot Study,” IFAC PapersOnLine, vol. 59, no. 18, pp. 61–66, 2025. https://doi.org/10.1016/j.ifacol.2025.10.197