AUTHOR=Wu Liangdong , Chen Yurou , Li Zhengwei , Liu Zhiyong 

TITLE=Efficient push-grasping for multiple target objects in clutter environments

JOURNAL=Frontiers in Neurorobotics

VOLUME=Volume 17 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2023.1188468

DOI=10.3389/fnbot.2023.1188468

ISSN=1662-5218

ABSTRACT=Intelligent manipulation of robots in unstructured environment is an important application field of artificial intelligence, which means that robots need to have the ability of autonomous cognition and decision-making. A typical example of this type of environment is a cluttered scene that is stacked and close together. In clutter, the target(s) may be one or more, and how to efficiently complete the target(s) grasping task is a challenging issue. In this paper, we propose an efficient push-grasping method based on reinforcement learning for multiple target objects in clutter. The key point of this method is to take the states of all the targets into consideration, so that the pushing action can expand the grasping space of all targets as much as possible, so as to achieve the minimum total number of pushing and grasping actions, and then improve the efficiency of the whole system. Around this point, we adopt the mask fusion of multiple targets, make clear the definition of graspable-probability, and give the reward mechanism of multi-target push-grasping. Experiments are carried out in both simulation and real system, and the experimental results indicate that compared with other methods, the proposed method has a better performance not only for multiple target objects but also for single target in clutter. Note also that our policy is trained under simulation only, which is transferred into the real system without retraining or fine-tuning.