Robotic Arm Robust Grasping and Region Smoothing for Fabric Manipulation

Manipulating and interacting with fabric is a dynamic process, but the manipulation of fabric in robotics presents significant challenges due to the complexities involved in accurately modeling flexible objects in three dimensions. This study focused on constructing comprehensive fabric datasets and labels while minimizing the disparity between simulated and real-world data through data augmentation techniques. Additionally, deep learning methods were employed to detect crucial points for fabric grasping and smoothing, which subsequently informed the identification of optimal grasping points and directions using a key-point recognition network. The results were successfully implemented on a multi-degree-of-freedom robotic arm, demonstrating the algorithm's superior performance compared to baseline approaches.