Design and Implementation of Intuitive Human Robot Interface System by DDPG with HER and RCA

This paper presents an intuitive human-robot interface system (iHRIS), where the deep deterministic policy gradient (DDPG) with hindsight experience replay (HER) training method is proposed to accelerate model training. The system consists of a motion capture system and a motion learning network. The motion capture system includes an RGB-D camera and an operating glove. The position of the human operator's hand is estimated by Openpose using the RGB-D images, while the hand's posture is determined by the information captured by the glove. An inertial measurement unit (IMU) and a microprocessor are equipped on the glove. The IMU data is calibrated using the Recursive Least Squares (RLS) method and computed using Madgwick's algorithm. Based on the observed position and posture of the human operator's hand, a motion can be generated. This motion is then trained using the DDPG network with the Reverse Curriculum Generation (RCG) method. The network has an Actor-Critic structure and a replay experience buffer, which makes it more feasible and helps avoid overfitting. Furthermore, HER is integrated into the network to enhance convergence speed and performance. Finally, the experiments demonstrate that the proposed iHRIS enables real-time imitation of the human operator by the robot, and the imitated motion can be learned by the DDPG network.