A Complex Skill Learning Method Integrating Particle Swarm and Variational Autoencoder for Robot

Aiming to address the complex operation problems of robots in daily environments, this paper studies unstructured demonstration-learning method, enabling robots to complete complex tasks simply and naturally. Firstly, the demonstrator continuously demonstrates the complete process of executing complex tasks to obtain the robot's operation trajectory, and then BP-AR-HMM algorithm is used to automatically segment the complex trajectory into a sequence of single skills. Then, a variational autoencoder (VAE) and the particle swarm optimization (PSO) algorithm are combined to learn the unlabeled demonstration skill sequence to reproduce the robot's complex operation trajectory. The PSO algorithm is explored to optimize the hyperparameter <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\boldsymbol{\beta}$</tex> in VAE model for improving learning accuracy and efficiency. The proposed <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{PSO} -\boldsymbol{\beta}-\mathbf{VAE}$</tex> algorithm can enable the robot to learn complex operation skills under optimal conditions. To verify the effectiveness of our method, three demonstration scenarios were designed. The results show that the proposed method has high learning accuracy and short learning time in learning the skill sequences.