Unsupervised Clustering of Repetitive Task Trials via Tucker Decomposition of Joint-Angle Data Tensors

Open Access
Article
Conference Proceedings
Authors: Akisue Kuramoto, Masaya Noguchi, Kosuke GOSHIMA, Daiki KOGA, Kazuki HIRANAI, Motomu Nakashima
Abstract

Movement patterns can vary across repetitions of the same task. This study aims to develop an analysis framework for comparing repeated movement trials in a shared feature space and describing the occurrence of movement patterns across repetitions. Time-series joint angle data was obtained from motion capture data recorded during a seated repetitive lifting task performed by twelve males. Partial Tucker decomposition provided temporal and joint-DOF bases shared across trials within each participant, while retaining trial-specific core coefficients. Gaussian mixture models were applied to these coefficients to group trials with similar kinematic features. Four to seven clusters were selected for each participant. In all participants, the dominant cluster changed in stages as the experiment progressed, although trial order was not used as a feature. Clusters dominant in early trials rarely regained dominance later, while later trials often showed repeated alternations in cluster membership probabilities. These results show that the framework can describe the occurrence and progression of movement patterns during repetitive work.

Keywords: Repetitive Task, Biomechanics, Joint Kinematics, Tucker Decomposition, Motion Pattern Classification

DOI: 10.54941/ahfe1008396

Cite this paper
Downloads
0
Visits
1
Download PDF

More from this volume

← Cardiac and Sleep-Stage Responses to Rhythmic Auditory and Amplitude-Modulated Vibrotactile Pillow Stimulation During Daytime NapsA Multitask Evolutionary Framework for Procedural Content Generation in Games →
View all articles in Human Factors in Design, Engineering, and Computing →