Robotic manipulation of deformable objects such as the garments is a challenging task because of its non-linear dynamics and a high degree of freedom. In general, it is quite difficult to model cloth state. But the Robotic handling of garments is necessary for assistive tasks like Robotic Clothing Assistance. In this paper, we are trying to automate the garment manipulation process by the robot. In our framework, the robot recognizes the garment it is handling, find appropriate grasping points on the garment and brings the garment in a particular state so that the clothing assistance task can start. In this research, we are investigating the applicability of Deep Learning framework for garment recognition and grasping point detection. We have used Real Garment Dataset generated using Kinect sensor and two types of Synthetic Garment Datasets generated using physics simulator Maya. Convolutional Neural Networks are trained using these datasets to predict the class of the garment and the spatial coordinates of the grasping points. The framework was tested using real garments and it shows promising results in the detection of the type of cloth and grasping points.
[1]
Nishanth Koganti,et al.
Bayesian Nonparametric Learning of Cloth Models for Real-Time State Estimation
,
2017,
IEEE Transactions on Robotics.
[2]
Carme Torras,et al.
Active garment recognition and target grasping point detection using deep learning
,
2018,
Pattern Recognit..
[3]
Nishanth Koganti,et al.
Robotic cloth manipulation for clothing assistance task using Dynamic Movement Primitives
,
2017,
AIR.
[4]
Takamitsu Matsubara,et al.
Reinforcement learning of clothing assistance with a dual-arm robot
,
2011,
2011 11th IEEE-RAS International Conference on Humanoid Robots.
[5]
Carme Torras,et al.
Robot-Aided Cloth Classification Using Depth Information and CNNs
,
2016,
AMDO.
[6]
Ioannis Mariolis,et al.
Pose and category recognition of highly deformable objects using deep learning
,
2015,
2015 International Conference on Advanced Robotics (ICAR).