School of Mechanical Engineering & Automation, Beihang University, Beijing 100191, China.
Beijing Advanced Innovation Center for Biomedical Engineering, Beihang University, Beijing 100191, China.
Math Biosci Eng. 2019 Dec 2;17(2):1495-1510. doi: 10.3934/mbe.2020077.
The present research envisages a method for the robotic grasping based on the improved Gaussian mixture model. The improved Gaussian mixture model is a method proposed by incorporating Bayesian ideas into the Gaussian model. It will use the Gaussian model to perform grasping training in a certain area which we called trained area. The improved Gaussian models utilized the trained Gaussian models as prior models. The proposed method improved the cumulative updates and the evaluation results of the improved models to make robots more adaptable to grasp in the untrained areas. The self-taught learning ability of the robot about grasping was semi-supervised. Firstly, the observable variables of objects were determined by a camera. Then, we dragged the robot to grasp object. The relationship between the variables and robot's joint angles were mapped. We obtained new samples in the close untrained area to improve the Gaussian model. With these new observable variables, the robot grasped it successfully. Finally, the effectiveness of the method was verified by experiments and comparative tests on grasping of real objects and grasping simulation of the improved Gaussian models through the virtual robot experimentation platform.
本研究提出了一种基于改进的高斯混合模型的机器人抓取方法。改进的高斯混合模型是一种通过将贝叶斯思想融入到高斯模型中提出的方法。它将使用高斯模型在我们称之为训练区域的特定区域中执行抓取训练。改进的高斯模型利用训练好的高斯模型作为先验模型。所提出的方法改进了改进模型的累积更新和评估结果，使机器人更适应在未训练区域中的抓取。机器人对抓取的自学习能力是半监督的。首先，通过相机确定物体的可观测变量。然后，我们拖动机器人去抓取物体。将变量和机器人关节角度之间的关系进行映射。在接近未训练的区域中获得新的样本，以改进高斯模型。利用这些新的可观测变量，机器人成功地抓取了物体。最后，通过在真实物体的抓取实验和通过虚拟机器人实验平台对改进的高斯模型的抓取模拟进行的对比测试，验证了该方法的有效性。
