[2M6-GS-13] AI application: Artificial General Intelligence

[2M6-GS-13-02] Double Articulation Analysis of Continuous Motions Depending on Reference Points

〇Hitoshi Kanno¹, Rikunari Sagara¹, Ryo Taguchi¹ (1. Nagoya Institute of Technology)

Keywords:Relative concept, Unsupervised Learning, Nonparametric Bayesian Double Articulation Analyzer

In this paper, we propose a method of segmenting and learning continuous motions depending on reference points by extending the model proposed by Taniguchi et al. As input, time series data of a continuous motion based on relative coordinates from each candidate are given, and reference point estimation and segmentation of the continuous motion are performed simultaneously. In experiments, the proposed method can segment continuous motions into individual motions more correctly than the conventional method. This indicates that estimation of reference objects improves the segmentation accuracy.