Developing application techniques of kinematics and simulation model for InMoov robot

Abstract

In this work, the direct and inverse kinematic analysis of both robot arms are investigated based on the analytical and informational representation. The results of the study will be used to provide the functionality of gesturing by a robot in sign language, both Kazakh and other languages, used in educational systems, especially in children's institutions and societies for deaf people. A simulation model of the movement of the robot's arms in the workspace has been studied and built. The developed model will be further implemented and used as mathematical and information support for the created robot. The developed library contains implementations of forward kinematics and iterative algorithms for inverse kinematics. The InMoov robot is a platform widely used in research tasks, supported by the MyRobotLab package. A direct kinematic model for the left and right hands of the robot has been studied. Based on the Python programming language, the working space for robot manipulators was calculated, using the matpilotlib library, an iteration method algorithm was developed to find the probable path of robot manipulators in space. A model of a structured artificial neural network (ANN) is proposed, which is used to find a solution to the inverse kinematics of the InMoov robot with six degrees of freedom (4-dof). The applied ANN model is a multilayer perceptron neural network (MLPNN) in which the learning rule of the Adam-a gradient diskend type is applied. To solve this problem, the problem of finding the best ANN configuration was studied. It has been established that a multilayer parseptron neural network gives the minimum mean square error. The regression coefficient analysis, which shows a 95.6 % fit of all communication variables, is acceptable for obtaining the inverse kinematics of the InMoov robot.

Arduino-based implementation of kinematics for a 4 DOF robot manipulator using artificial neural network

“…Raheem et al [3] proposed a multilayer perceptron structure by using a back propagation training algorithm to find the required joint angle for the end effector position. C. Kenshimov et al [4] trained a multi-layer perceptron neural network (MLPNN) for a 4-DOFs INMOOV manipulator to investigate the desired set of angle position from a given set of end effector pose, where the data base on the iterative logic of solving the inverse kinematics, the experimental results showed an acceptable mapping of robot workspace with 95.6% fit for all joints angles. For a 4-DOFs SCARA robot, P. Jah and B. Biswal [5] studied the application of MLP, they observed that the Neural Network (NN) minimized the joints variables errors and improves the performance index.…”