Facial Expression Control Method for Humanoid Expression Robot Based on Multi-Layer Gate Recurrent Unit Network

Humanoid expression robots can perform facial expression that are highly similar to humans, which is of great significance for improving the experience of human-computer interaction. In this paper, a novel electromechanical system of humanoid expression robot based on the coupling drive of flexible skin and rigid body mechanism is developed. 22 effective control points of the flexible skin of humanoid expression robot are designed. The facial feature vector of the robot based on the facial feature point extraction method is constructed. In order to obtain the nonlinear mapping relationship between the facial feature vector and the motor control vector of the rigid-flexible coupled humanoid expression robot, and realize the mapping relationship in the whole expression formation process, an inverse kinematics mapping model of humanoid expression robot based on multi-layer gate recurrent unit (GRU) network was proposed. The effect of the multi-layer GRU network on the formation of robot facial expressions is verified by experiments, and the average temporal similarity and spatial similarity of the five basic facial expressions are 91.3% and 84.3%.