Extraction of Grasping Motions from sEMG Signals for the Control of Robotic Hands through Autoencoding

Robotic hands are used in many applications, including prosthetic devices controlled by the nervous system, as well as end-effectors in industrial automation or medical surgery, etc. One of the most investigated techniques to control them involves the use of surface electromyographic signals. Surface electromyographic signal (sEMG) is the electrical signal generated by muscle activity and recorded by non-invasive surface electrodes placed on the skin. The basic control pattern consists in online extracting neural information from sEMG signals during muscles contraction and then using these signals to control the robotic device. My thesis will focus on the control information extracting phase. Specifically, we want to extract grasping activation intentions from sEMG signals to control robotic hands. Unsupervised regression methods for extracting control information are particularly interesting because they do not require any label, with a consequent simplification and speed up of the training process. Among these, we want to investigate autoencoders and NMF ability to extract control information describing muscle activity during different grasping motions. The project will be developed along two parallel directions: simulation and experimental case. In the simulation case, the input for NMF or autoencoder will be generated through a model of sEMG signals that I implemented for my bachelor thesis and suitably modified in order to consider single fingers movements. In the second case, we will use real sEMG signals recorded during experimental sessions. Additionally, we will analyse the role of intrinsic muscles during fingers movements. We will compare simulation results with real results, in order to understand the likelihood level of the model. As final step, we will simulate the control a robotic hand using the neural commands obtained with NMF and autoencoding. To obtain this, we will leverage on the theory of postural synergies.