“Less is more”: Simplifying point clouds to improve grasping performance

Abstract

Object grasping is a task that humans do without major concerns. This results from self learning and by observing of other skilled humans doing such task with previous information. However, grasping novel objects in unknown positions for a robot is a complex task which encounters many problems, such as sub-optimal performance rates and the time consumption. In this paper we present a method that complements the state-of-the-art grasping algorithms with two segmentation steps, the first one which removes the largest planar surface in the point cloud of the world before the grasp detector receives them and the second one that complements this segmentation with another segmentation that calculates where the object is located and segments the point cloud by executing a crop around the object. The proposed method significantly improves the grasping success rate (100% improvement over the baseline approach) and simultaneously is able to reduce the time consumption by 23%.

Robotics Dexterous Grasping: The Methods Based on Point Cloud and Deep Learning

“…By adopting sampling method in GPD proposed by Pas et al (2017), Lopes et al (2018), Schnaubelt et al (2019), Bui et al (2020), Chen et al (2020), and Deng et al (2020) sample the grasp points in point cloud for candidates generation. Lopes et al (2018) find the largest planar surfaces which is infeasible for grasping by using RANSAC (Fischler and Bolles, 1981) and isolates the closest object to the camera from the rest of the scene to obtain object segmentation based on min-cut (Golovinskiy and Funkhouser, 2009). This work compares the experiments before and after reducing the point cloud search space, and proves that the grasping success rate has increased from 45 to 90%.…”

“…The drawback causes the low sampling accuracy and time-consuming sampling process. To avoid these defects, researchers propose object-aware sampling which aims to combine the specific information of the object to enhance the reasonability of search space in point cloud (Boularias et al, 2015 ; Zapata-Impata et al, 2017 ; Lopes et al, 2018 ). The search space refers to the points that need to be considered in the point cloud space for the grasping pose sampling algorithm.…”

“…Inspired by several state-of-arts works in computer vision (Ren et al, 2015 ; He et al, 2016 , 2017 ; Liu et al, 2016 ; Redmon et al, 2016 ), researchers begin to move the attention to grasp candidate generation. Some works attempt to use object detection and instance segmentation approaches to reduce the search space in the point cloud (Lopes et al, 2018 ; Schnaubelt et al, 2019 ; Bui et al, 2020 ) or regress the grasp part using rectangle bounding boxes directly (Vohra et al, 2019 ; Yu Q. et al, 2020 ). Specifically, models designed to process point cloud furtherly accelerate the progress of candidate generation (Lin and Cong, 2019 ; Yu S. et al, 2020 ) and evaluation (Singh et al, 2018 ; Mousavian et al, 2019 ; Yan et al, 2019 ; Fang H.-S. et al, 2020 ; Qian et al, 2020 ).…”