Improving Inverse Kinematic solutions of Robotic Systems for Accuracy and Speed Using Neural Network Ensembles

One of the significant problems in design and control arobot is robot’s kinematics, in which inverse kinematics iscritical for high DoF robots. Although, many algorithmshave been employed to solve this problem, the deficiency ofefficient methods is observable. As the Complexity of robotincreases, solving the inverse kinematics is difficult andcomputationally expensive. Traditional methods such asgeometric, iterative and algebraic are inadequate, if therobot is more complex. As an alternative approach, neuralnetworks have been widely used for inverse kinematicmodeling. This paper investigates two neural networkensembles, i.e., averaging and mixture of experts, to solvethe inverse kinematics for the robotic manipulators, then itis applied to a manipulator with 3 degrees of freedom as acase study. The obtained results from the methods arepresented and analyzed in order to prove the efficiency ofthe proposed approach in comparison with analyticalsolutions.

کلیدواژه ها:

Inverse Kinematics ، Ensembles NeuralNetwork ، Multi-layer perceptron ، back-propagationtraining algorithm