Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Center Direction Network for Grasping Point Localization on Cloths

Domen Tabernik, Jon Muhovič, Matej Urbas, Danijel Skočaj

TL;DR

CeDiRNet-3DoF, a deep-learning model for grasp point detection, with a particular focus on cloth objects, is introduced, showing robustness in real-world performance, outperforming state-of-the-art methods, including the latest transformer-based models.

Abstract

Object grasping is a fundamental challenge in robotics and computer vision, critical for advancing robotic manipulation capabilities. Deformable objects, like fabrics and cloths, pose additional challenges due to their non-rigid nature. In this work, we introduce CeDiRNet-3DoF, a deep-learning model for grasp point detection, with a particular focus on cloth objects. CeDiRNet-3DoF employs center direction regression alongside a localization network, attaining first place in the perception task of ICRA 2023's Cloth Manipulation Challenge. Recognizing the lack of standardized benchmarks in the literature that hinder effective method comparison, we present the ViCoS Towel Dataset. This extensive benchmark dataset comprises 8,000 real and 12,000 synthetic images, serving as a robust resource for training and evaluating contemporary data-driven deep-learning approaches. Extensive evaluation revealed CeDiRNet-3DoF's robustness in real-world performance, outperforming state-of-the-art methods, including the latest transformer-based models. Our work bridges a crucial gap, offering a robust solution and benchmark for cloth grasping in computer vision and robotics. Code and dataset are available at: https://github.com/vicoslab/CeDiRNet-3DoF

Center Direction Network for Grasping Point Localization on Cloths

TL;DR

CeDiRNet-3DoF, a deep-learning model for grasp point detection, with a particular focus on cloth objects, is introduced, showing robustness in real-world performance, outperforming state-of-the-art methods, including the latest transformer-based models.

Abstract

Object grasping is a fundamental challenge in robotics and computer vision, critical for advancing robotic manipulation capabilities. Deformable objects, like fabrics and cloths, pose additional challenges due to their non-rigid nature. In this work, we introduce CeDiRNet-3DoF, a deep-learning model for grasp point detection, with a particular focus on cloth objects. CeDiRNet-3DoF employs center direction regression alongside a localization network, attaining first place in the perception task of ICRA 2023's Cloth Manipulation Challenge. Recognizing the lack of standardized benchmarks in the literature that hinder effective method comparison, we present the ViCoS Towel Dataset. This extensive benchmark dataset comprises 8,000 real and 12,000 synthetic images, serving as a robust resource for training and evaluating contemporary data-driven deep-learning approaches. Extensive evaluation revealed CeDiRNet-3DoF's robustness in real-world performance, outperforming state-of-the-art methods, including the latest transformer-based models. Our work bridges a crucial gap, offering a robust solution and benchmark for cloth grasping in computer vision and robotics. Code and dataset are available at: https://github.com/vicoslab/CeDiRNet-3DoF
Paper Structure (30 sections, 4 equations, 4 figures, 4 tables)

This paper contains 30 sections, 4 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Methods
  4. CeDiRNet
  5. Dense regression of center-directions
  6. Localization from center-directions
  7. CeDiRNet-3DoF for grasp point localization
  8. Training
  9. Inference
  10. Comparison to the original CeDiRNet
  11. The ViCoS Towel Dataset
  12. Dataset construction
  13. Objects
  14. Object position
  15. Backgrounds
  16. ...and 15 more sections

Figures (4)

  • Figure 1: The proposed CeDiRNet-3DoF architecture.
  • Figure 2: Overview of images in the ViCoS Towel Dataset with different towels, corner configurations, backgrounds, clutter, and lightning.
  • Figure 3: Several examples of correct grasp point detection (green cross) and estimation of angle-of-approach (dark lines predicted, light lines ground-truth).
  • Figure 4: Examples of false and missed grasp point detections (true as a green cross, false as a red cross, and ground-truth as a green dot).