A novel human-to-robot interaction model based on transfer expert reinforcement learning with recurrent neural network

Abstract

The control tasks related to interaction tracking are mainly limited in robot manipulators-based traditional applications. In this, the desired motivations are specified based on the trajectories and the desired positions. The robots are programmed by using the teach-and-playback method in such applications that are assumed to be more convenient. Moreover, the advancements in sensing and robotic methodologies fulfill the satisfactory requirements of more demanding tasks. Several instructions are provided for interacting robots with humans in order to perform a sequence of more difficult tasks. It does not require learning the motions, but it only requires learning the positions of the motions in such applications, and this position is learned by using the robot controller. The major aim of this research work is to develop a new Transfer Expert Reinforcement Learning (TERL) method to offer efficient interaction between humans and computers. In this developed model, Reinforcement Learning (RL) is utilized to observe the movement of the robotic arm. Then, robot movement is considered with the help of a deep learning approach named Recurrent Neural Network (RNN) along with inputs of kinematic movement. Finally, the proposed model achieves a superior rate than conventional approaches in human to human-to-robot interaction model.