Wei Chen, Dongmyoung Lee, Digby Chappell and Nicolas Rojas
REDS Lab, Dyson School of Design Engineering, Imperial College London
Accepted to 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2023)
Abstract
When performing cloth-related tasks, such as garment hanging, it is often important to identify and grasp certain structural regions---a shirt's collar as opposed to its sleeve, for instance. However, due to cloth deformability, these manipulation activities, which are essential in domestic, health care, and industrial contexts, remain challenging for robots. In this paper, we focus on how to segment and grasp structural regions of clothes to enable manipulation tasks, using hanging tasks as case study. To this end, a neural network-based perception system is proposed to segment a shirt's collar from areas that represent the rest of the scene in a depth image. With a 10-minute video of a human manipulating shirts to train it, our perception system is capable of generalizing to other shirts regardless of texture as well as to other types of collared garments. A novel grasping strategy is then proposed based on the segmentation to determine grasping pose. Experiments demonstrate that our proposed grasping strategy achieves 92%, 80%, and 50% grasping success rates with one folded garment, one crumpled garment and three crumpled garments, respectively. Our grasping strategy performs considerably better than tested baselines that do not take into account the structural nature of the garments. With the proposed region segmentation and grasping strategy, challenging garment hanging tasks are successfully implemented using an open-loop control policy.
The Garment Hanging Task
In this project, we investigate one of the daily tasks that most people perform: hanging garments. While hanging is easy for a human it is rather difficult for a robot. A few steps are needed to be performed: Find the collar of the crumpled garment, grasp it and transport it to the hook.
Challenges of Garment Hanging
As a deformable object, garment can deform in many ways, leading to self-occlusions. This will be particullarly challenging for robot perception and manipulation.
It is difficult to grasp the garment without a robust pose estimation, since the garments are highly deformable.
The perception of a garment is the key to the following manipulation task. Failed perception will easily cause failure of garment manipulation system.
Framework
We propose a novel cloth manipulation system by utilising the structural element of garments for both perception and grasping.
The training of this perception system is based on data acquisition of a 10-minute video of human manipulation with automatic annotation. Experiment results indicate the proposed perception system can generalize to other types of cloth.
We proposed a novel pose estimation strategy, which can implement robust grasping of folded and crumpled garments
We implement a novel garment-hanging task by the proposed perception and grasping algorithm. The results indicate that the challenging garment-hanging task can be implemented with hard-coded execution by only considering the structural regions of garments.
Perception System
Data Acquisition and Model Training
Unstructured Manipulation is randomly applied to the targeted cloth by a human operator.
One RGB-D image sequence is recorded in the meantime. We use the depth images as the input with the groundtruth extracted from RGB images (blue colour extraction).
A U-net based segmentation module is applied here for the perception. Only depth images are used for execution time.
Pose Estimation for Grasping and Manipulation
A novel grasping strategy is then proposed that outputs a suitable grasp pose based on the extracted skeleton and local geometric structure of the segmented region of clothing.
Skeleton-based Center extraction
We extract the center area by fitting the collar region into a skeleton.
Agglomerative clustering and skeletonizing are used to obtain the single skeleton. The center region is derived by finding the maximum closeness centrality point.
Grasping Pose Estimation
De-project these collar pixels from the depth image to obtain the collar point cloud
Surface variation estimation near the skeleton center of the local point cloud is applied. The point of largest surface variation is selected as the grasping point.
A PCA-based orientation estimation is conducted to obtain the final grasping pose.
Action Primitives for Grasping and Garment Hanging
A pre-grasping pose is defined to ensure the insertion grasp and avoid a potential collision.
After the robot grasps the cloth, the robot will execute pre-defined hanging trajectories, which are reaching the intermediate position (blue dot), the center of the hook (red dot), and releases.
Experiment Setup
Detailed Presentation Video and Experiment Trials
