Prediction of Inverse Kinematics Solution of a Redundant manipulator using ANFIS

Abstract

In this thesis, a method for forward and inverse kinematics analysis of a 5-DOF and a 7-DOF Redundant manipulator is proposed. Obtaining the trajectory and computing the required joint angles for a higher DOF robot manipulator is one of the important concerns in robot kinematics and control. When a robotic system possesses more degree of freedom (DOF) than those required to execute a given task is called Redundant Manipulator. The difficulties in solving the inverse kinematics (IK) equations of these redundant robot manipulator arises due to the presence of uncertain, time varying and non-linear nature of equations having transcendental functions. In this thesis, the ability of ANFIS (Adaptive Neuro-Fuzzy Inference System) is used to the generated data for solving inverse kinematics problem. The proposed hybrid neuro-fuzzy system combines the learning capabilities of neural networks with fuzzy inference system for nonlinear function approximation. A single-output Sugeno-type FIS (Fuzzy Inference System) using grid partitioning has been modeled in this work. The Denavit-Hartenberg (D-H) representation is used to model robot links and solve the transformation matrices of each joint. The forward kinematics and inverse kinematics for a 5-DOF and 7-DOF manipulator are analyzed systemically.
ANFIS have been successfully used for prediction of IKs of 5-DOF and 7-DOF Redundant manipulator in this work. After comparing the output, it is concluded that the predicting ability of ANFIS is excellent as this approach provides a general frame work for combination of NN and fuzzy logic. The Efficiency of ANFIS can be concluded by observing the surface plot, residual plot and normal probability plot. This current study in using different nonlinear models for the prediction of the IKs of a 5-DOF and 7-DOF Redundant manipulator will give a valuable source of information for other modellers.