NEURAL NETWORK FOR INVERSE KINEMATICS PROBLEM

In Robotics, finding solution of problems related to Inverse Kinematics (IK) is one of the problems. Increasing the complex of the joint structure of robot the traditional geometric, iterative and algebraic methods are inadequate. The inverse position kinematics solves the following problem: “Given the desired pose of the robot’s hand; what must be the angles at the joints?” Unlike to the forward problem, the solution of the inverse problem is not always unique: The same end effector’s pose can be reached by distinct joint position vectors, corresponding different configurations. The problem of IK consists of the conversion of the position and orientation of a robot manipulator end-effector from Cartesian space to joint space. In this work, it is possible to predict the angles according to coordinates x,y,z of the real-world thanks to the ability of Adaptive Neuro-Fuzzy Inference System (ANFIS) and Artificial Neural Network(ANN) to learn from training data. An exemplary trajectory for two-link planar and PUMA robot was used to perceive the correctness of system responses obtained from ANN and ANFIS output. In order to design, simulate and testing manipulators was used the Robotics System Toolbox, while scikit-learn and then TensorFlow library, that allows to define any mathematical model and provides appropriate methods for machine learning, for designing and developing artificial neural network. To build ANFIS there was need the Fuzzy Logic Toolbox . The program generating training data, learning network with backpropagation algorithm and validating obtained solution was written in Python scripting language.