A Computer Vision Approach for the Inverse Kinematics of 2 DOF Manipulators Using Neural Network

Solving inverse kinematics(IK) equation is a very important task in robotic manipulation. For the first time, this paper presents an improved neural network-based solution for the inverse kinematics problem in manipulators. The concept behind this work is to eliminate the need for developing explicit IK equation thereby avoiding its complexity for modelling complex configurations. The majority of existing studies for developing IK equation using neural network created training data either by using forward kinematics or inverse kinematics for known coordinates which makes the work complex. The working of the proposed method implemented using the simulation for 2 DOF planar and spatial robotic arms in Gazebo which is interfaced with robotic operating system(ROS). The learning task in the proposed method use images captured using a camera to track the arm for acquiring training data. Results obtained with this method are compared with solutions obtained using kinematic equations. Mean square error obtained between the two are insignificant which shows good consistency of new method.