Parameter Optimization of Trajectory Imitation Learning Characterization Based on Gaussian Mixture Model

To tackle the problem of low efficiency of parameter selection for Gaussian mixture model (GMM), an integrated method was proposed to estimate the parameters of GMM in characterization of trajectory imitation learning in this paper. The k-means algorithm was improved by this mothod based on maximum minimum distance algorithm which belonged to multi-center clustering algorithm and the optimal initial cluster center was obtained. Four significant GMM parameters were obtained at the same time by using genetic algorithm based on the Bayesian information criterion (BIC). The efficiency of dividing the initial data set was im proved, and the initial cluster center and the number of the mixture model was determined, and the local extreme problem caused by initial value was effectively avoided. Multiple sets simulation results demonstrate the validity of this method.