Design and Implementation of a Robot Pose Predicting Recurrent Neural Network for Visual Servoing Application

Visual servoing is the method of controlling a robot using image input from one or more image sensors to complete a predefined task. This paper examines the effectiveness of a Recurrent Neural Network (RNN) to predict the position and orientation (pose) of an industrial robot manipulator for automatic pick and place applications mainly in unstructured environment. The robot manipulator moves to the target object based on the pose commands obtained from the trained neural network. Various images obtained from the camera attached to the end-effector and corresponding pose of the end-effector are the input and the output data for training the neural network. The performance of the RNN in predicting the robot pose is compared with the feedforward neural (FFN) network and cascade forward neural (CFN) network. The proposed method is validated experimentally using ABB IRB 1200 6-DOF industrial robot manipulator.