The application areas of robot manipulators are increasingly growing year by year. Rehabilitation exoskeletons and upper limb prostheses are special cases of robot manipulators. Such systems, being nonlinear, multidimensional, and dynamic systems, have parametric uncertainties. All this becomes a challenge in the design process of the control system. One of the advanced technologies in this field is adaptive control, which is an important and interesting area of scientific and industrial research. The nonlinear control system of the two-linked robot manipulator was considered in the work. In this case, classical control methods become inapplicable and there is a need to develop and apply other control methods. Using the gain scheduling method, a group of PID controllers are designed to ensure the stability of the system at various operating points. The non-linearity of the system is compensated by the gain scheduling method and the only problem that remained is the uncertainties of the system depending on the weight of the operator. A model reference adaptive controller (MRAC) was designed and augmented with the gain scheduling method for eliminating uncertainties. Simulation is done in the MATLAB Simulink environment. A comparative performance has been quantified with different adaptation gains for the upper limb prosthesis system. The obtained results show the viability of the proposed control method.

The use of the bionic hand requires an extensive training procedure which is a major challenge for patients. The patients need to learn to control the bionic hand before starting using it, therefore, training should be done efficiently. One of the proposed methods is controlling the virtual bionic hand via physical EMG (electromyography) sensors. In general, one of the main problems of any prosthesis is the classification of the patient's finger movements. For this reason, some well-known machine learning algorithms are discussed. Comparative analysis of machine learning algorithms is performed, the best-selected algorithm is used for the system later. The classifier for finger movement classification is trained and tested. The virtual model of the bionic hand has been developed. The kinematics of bionic fingers is analyzed. The bionic finger performs tasks in the Cartesian space, whereas actuators work in the Quaternion (joint) space. It is necessary to transform the coordinate system from Cartesian to joint space and vice versa․ The inverse and forward kinematics is obtained by using the geometry approach and the Denavit - Hartenberg (DH) methods accordingly. The control system is designed for the virtual bionic hand model. The developed method gives an opportunity to classify all the movements of fingers via two surface EMG electrodes with an ML (Machine learning) based or the NN (Neural Network) classifier, and to control the designed bionic hand model in the MATLAB / Simulink environment.

No abstract