薛腾, 刘文海, 潘震宇, 王伟明. 基于视觉感知和触觉先验知识学习的机器人稳定抓取[J]. 机器人, 2021, 43(1): 1-8. DOI: 10.13973/j.cnki.robot.200040
引用本文: 薛腾, 刘文海, 潘震宇, 王伟明. 基于视觉感知和触觉先验知识学习的机器人稳定抓取[J]. 机器人, 2021, 43(1): 1-8. DOI: 10.13973/j.cnki.robot.200040
XUE Teng, LIU Wenhai, PAN Zhenyu, WANG Weiming. Stable Robotic Grasp Based on Visual Perception and Prior Tactile Knowledge Learning[J]. ROBOT, 2021, 43(1): 1-8. DOI: 10.13973/j.cnki.robot.200040
Citation: XUE Teng, LIU Wenhai, PAN Zhenyu, WANG Weiming. Stable Robotic Grasp Based on Visual Perception and Prior Tactile Knowledge Learning[J]. ROBOT, 2021, 43(1): 1-8. DOI: 10.13973/j.cnki.robot.200040

基于视觉感知和触觉先验知识学习的机器人稳定抓取

Stable Robotic Grasp Based on Visual Perception and Prior Tactile Knowledge Learning

  • 摘要: 针对机器人抓取过程中需要实时评估抓取质量以动态调整抓取构型的问题,提出了基于触觉先验知识的机器人稳定抓取方法.首先,根据抓取过程中物体抵抗外界扰动的能力,提出了一种基于触觉信息的抓取质量评估方法.在此基础上,建立了视触觉联合数据集并学习触觉先验知识.其次,提出了融合视觉图像和触觉先验知识的稳定抓取构型生成方法.最后,在搭建的机器人抓取系统中对10种目标物体进行了实验验证.结果表明,相比传统的视觉方法,该方法的抓取稳定性提高了55%;针对已知物体和未知物体,稳定抓取成功率分别为86%和79%,体现了较好的泛化效果.

     

    Abstract: Considering the problem that the robot needs to estimate the grasp quality in real time to dynamically adjust the grasp configuration in grasping process, a stable robotic grasp method based on prior tactile knowledge learning is proposed. Firstly, a tactile information based method is put forward for evaluating grasp quality according to the capacity of resisting the external perturbations in grasping process. Based on it, a visual-tactile joint dampest is built, and the prior tactile knowledge is learned. Secondly, an architecture is proposed to generate stable grasp configurations by fusing the visual image and the prior tactile knowledge. Finally, an experimental verification with 10 kinds of objects is carried out on the proposed robotic grasp system. The results show that the grasp stability is improved by 55% with the proposed method over traditional vision-based methods, and the success rates of stable grasp are 86% and 79% respectively on known and unknown objects, demonstrating good generalization performance of the proposed method.

     

/

返回文章
返回