Dongmyoung Lee*, Wei Chen*, Xiaoshuai Chen and Nicolas Rojas
*Equal first-author contribution
REDS Lab, Dyson School of Design Engineering, Imperial College London
Open Source File: [CAD]
IEEE ICRA 2023, Cloth Manipulation Competition, Folding Track,
Honorable Mention
Awardees: Nicolas Rojas, Wei Chen, Dongmyoung Lee, Xiaoshuai Chen
Abstract
The manipulation of garments poses research challenges due to their deformable nature and the extensive variability in shapes and sizes. Despite numerous attempts by researchers to address these via approaches involving robot perception and control, there has been a relatively limited interest in resolving it through the co-development of robot hardware. Consequently, the majority of studies employ off-the-shelf grippers in conjunction with dual robot arms to enable bimanual manipulation and high dexterity. However, this dual-arm system increases the overall cost of the robotic system as well as its control complexity in order to tackle robot collisions and other robot coordination issues.
As an alternative approach, we propose to enable bimanual cloth manipulation using a single robot arm via novel end effector design---sharing dexterity skills between manipulator and gripper rather than relying entirely on robot arm coordination. To this end, we introduce a new gripper, called G.O.G., based on a gripper-on-gripper structure where a first gripper independently regulates the span, up to 500mm, between its fingers which are in turn also grippers. These finger grippers consist of a variable friction module that enables two grasping modes: firm and sliding grasps. Household item and cloth object benchmarks are employed to evaluate the performance of the proposed design, encompassing both experiments on the gripper design itself and on cloth manipulation. Experimental results demonstrate the potential of the introduced ideas to undertake a range of bimanual cloth manipulation tasks with a single robot arm.
Challenges of Cloth Manipulation Systems
Regarding cloth grasping and manipulation, the majority of studies enables bimanual manipulation utilizing two robot arms. This natural approach introduces a variety of challenges though, encompassing collision avoidance, intricate control strategies, and larger costs.
Although unimanual manipulation is better in terms of control complexity and overall system cost, one single grasping end-effector is challenging to constrain the cloth deformation during the implementation of cloth manipulation tasks.
Gripper-On-Gripper: Overview
In this work, we introduce a novel gripper design for bimanual cloth manipulation tasks using a single robot arm. Our gripper configuration includes a Width Control Gripper (WCG) and two Variable Friction Grippers (VFG) in a Gripper-on-Gripper (G.O.G.) structure. The WCG regulates the width of the VFGs, allowing the gripper to grasp fabric corners of various sizes. The VFG operates in two modes: power grasping and sliding grasping, achieved through a variable friction (VF) module. Power grasping uses high friction for a secure hold while sliding grasping employs low friction to flatten the cloth. The mode transition is passive, triggered by the gripping force applied.
Width Control Gripper (WCG)
The WCG is actuated by a single stepper motor (NEMA 17) using tendons, which serves to reduce the dimensions of the G.O.G. and simplify the control strategy. The transmission system efficiently transmits the motor's force to each individual finger of the WCG via the utilization of tendons connecting the motor to the fingers. A linear guide is employed to ensure the smoothness of each finger's movement, thereby minimizing friction during both forward and backward motions.
Variable Friction Gripper (VFG)
In this work, we propose using VFG for secure grasping and flattening motion control by adjusting DC motor torque, as shown. The variable friction module includes a roller for low-friction assistance during cloth flattening while the WCG is in motion. As the DC motor torque increases, the VFG switches to high friction mode, allowing the gripper to securely attach to garments using a silicone part. Our variable friction mechanism is based on DC motor control, eliminating the need for extra actuation to switch friction simplifying gripper control and circuit design.
G.O.G for Bimanual Cloth Manipulation Tasks
One major advantage of our design is the implementation of bimanual cloth manipulation by employing just one robot arm. Our proposed gripper design can significantly simplify the complex motion planning procedure for cloth manipulation. In this paper, we use the bimanual cloth folding tasks as the case study to demonstrate the effectiveness of our gripper design in facilitating cloth manipulation tasks. In addition, we further investigate the potential of our gripper by the bimanual cloth hanging and ending flattening tasks.
Automatic Bimanual Cloth Folding
Folding trajectories based on the detected corner point. The Width Control Gripper (WCG) adjusts the opening width firstly. Then, pre-grasp, grasp, folding, and release actions are executed. To simplify the overall control process, we implement the perception system with a classical corner detection algorithm. With an overhead Realsense RGB-D camera, we can locate the corner points of the target cloth for the cloth grasping. We then define a set of folding trajectories based on the detected corner points. In the meantime, the grasp width is also adjusted according to the distance of two corner points. Our design is to ensure a proper distance between two small VFGs to limit the corner deformation for better folding quality. An evaluation of the bimanual cloth folding system is conducted in the following sections.
Bimanual Cloth Hanging and Cloth Edging Flattening
For the garment hanging task, a fixed configuration of the garment is applied. The following pre-defined motion, including grasp, lift, transport and release, is executed to finish the task. We then performed a cloth flattening task with the G.O.G. gripper. A similar fixed configuration of the garment is also applied in this task. In the firm grasp stage, we employed a high-friction silicone pad to ensure a robust grasping force. Subsequently, for the flattening motion, we applied a reduced torque for the grasping. The roller then took on the primary role as the main point of contact to facilitate smooth sliding motion.
Experiments Setup
To demonstrate the capabilities of our proposed gripper design, we conduct an evaluation of the G.O.G. using two fabric benchmarks: household items and clothing objects. To be more precise, our assessment comprised three individual experiments: (1) The payload capacity of the G.O.G. design, (2) cloth grasp-and-lift experiment, (3) cloth drag placement accuracy, and (4) cloth manipulation tasks to effectively demonstrate the prowess of our gripper design and its integration with a robotic system. This comprehensive evaluation serves to highlight the capabilities and performance of our innovative gripper design.
Gripper Payload Capacity
We test the gripper's strength with a Force/Torque sensor (Robotiq FT 300). The setup involves the FT sensor and gripper on a UR5 robot, moving away while fabric is fixed to a rigid platform. We measure the maximum pulling force at 30N, designed to safely handle any fabric without causing damage. The force is recorded at its peak just before the grasp fails.
Cloth Grasp-and-Lift
We used a household item benchmark to evaluate the gripper's grasping ability. This assessment helps us assess the gripper's performance with different sizes and types of materials. To assess a successful cloth grasp, we lifted the fabric from the table surface to a height of 500mm, securely held it in that position for 5 seconds.
Cloth Dragging
We evaluate dragging placement accuracy for the cloth object benchmark, allowing us to compare our gripper's performance with commercially available ones. To ensure a fair comparison, our experimental setup closely follows the methodology outlined in the household benchmark. The assessment procedure is as follows: (1) Grasp the flat edge of each cloth object benchmark, (2) lift the object securely, (3) horizontally translate it 500mm using the manipulator to drag, and (4) release the object.
Robotic System Experiment (Bimanual Folding)
In this section, we demonstrate cloth folding performance to showcase the capabilities of our bimanual robotic system with the G.O.G. gripper and a single UR5 robot arm. We use bimanual cloth folding as our case study and follow a benchmark for the experiment. The experiment comprises two phases: the first fold (1-fold) and the second fold (2-fold). For the first folding, we start with the target fabric in a flat initial state for precise performance measurement. Additionally, we rotate the cloth to various angles for robust results. Each object undergoes five trials, including both 1-fold and 2-fold, in this experiment. We obtain the evaluation results by calculating mIoU with a wrinkle penalty between the desired folding image and the actual result image.
Detailed Presentation Video and Experiment Trials
