AUTHOR=Zhang Yang , Xie Lihua , Li Yuheng , Li Yuan 

TITLE=A neural learning approach for simultaneous object detection and grasp detection in cluttered scenes

JOURNAL=Frontiers in Computational Neuroscience

VOLUME=Volume 17 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2023.1110889

DOI=10.3389/fncom.2023.1110889

ISSN=1662-5188

ABSTRACT=Simultaneously detecting objects and its grasp pose, enabling reasoning object manipulation, is essential and challenging for robots or unmanned systems working in a cluttered real-world environment. To achieve this, we propose a novel neural learning approach, namely SOGD, to predict a best grasp configuration for each detected objects in an RGB-D image. Our model adopts two separate branches to detect objects and grasp configurations respectively. Unlike previous grasp detection methods, we design an alignment module to estimate the correlations between the detected objects and grasp configurations. This module is the key component which enables our model to predict more reasonable grasp configurations for each detected object. Besides, a 3D-plane-based background removal is presented to filter out the cluttered background. We test our method on two public datasets (Cornell Grasp Dataset and Jacquard Dataset) and evaluate the performance against state-of-the-art methods. Results show that our method achieves +0.7% ~ +1.4% performance boost in terms of average accuracy compared with existing RGB-D based grasp detection methods.