Clothing Assistance is a basic assistance activity in the daily life of the elderly and disabled. However, robotic clothing assistance is highly challenging problem that involves close interaction of the robot with non-rigid clothing materials and the assisted person whose posture can vary while the clothing task is being performed. For these reasons, the Human-Cloth relationship needs to be accurately estimated in realtime to ensure the successful completion of the clothing task. In our previous study [1], we have developed a method for the real-time estimation of the human-cloth relationship using a depth sensor. However, the accuracy of our method reduces when there is severe occlusion of the clothing article or noise from the depth sensor. To address these problems, there is a need for robust tracking of the human-cloth relation-ship ensuring smooth state transitions. An approach to solving this tracking problem can be the dynamical modelling of the human-cloth relationship and the use of this motion model as a prior for real-time tracking under noise and occlusion. In this study, we evaluate the effectiveness of dynamical modelling of nonrigid clothing articles using Gaussian Process Latent Variable Model (GP-LVM) in the domain of robotic clothing assistance.
展开▼