A Novel Approach for Efficient Gaussian Mixture Model using Dynamics-motivated Optimal Excitation

Abstract

EffiDynaMix, a novel, efficient, gray-box, nonparametric dynamics modeling method, integrates mathematical dynamics with Gaussian Mixture Model (GMM) for simplified training data creation and improved generalization. It outperforms traditional methods like conv-GMM, GP, and LSTM in training efficiency and accuracy with new data. By leveraging dynamic equations, EffiDynaMix enhances learning efficiency and adaptability, offering advancements in robotic system precision and computational efficiency, leading to faster and more responsive robots. © 2024 IEEE.