基于抓取模式识别的欠驱动灵巧手抓取方法

Under-actuated dexterous hand grasping method based on grasp type detection

针对桌面上单个物体场景的抓取任务,提出一种基于抓取模式识别的欠驱动灵巧手自主抓取方法.受人类抓取策略启发,基于四种典型抓取模式建立物体的抓取模式数据集,并通过深度学习预测物体的抓取模式和抓取区域,利用图像处理获得抓取角度,从而简化欠驱动灵巧手的抓取规划.深度学习算法在测试集中的识别准确率达98.70%,对未知物体的识别准确率达82.70%,具有较好的泛化能力.当执行自主抓取时,深度学习方法的不准确性通过欠驱动手的自适应性得到了一定的补偿.通过UR3e机械臂搭载欠驱动灵巧手对24个物体进行抓取实验,在120次抓取中平均成功率为90.80%.实验结果表明所提方法能适应不同形状大小的物体,具备抓取实用性.

To grasp a single object on the desktop,an autonomous grasping method based on grasp type detection for under-actuated dexterous hand was proposed.Inspired by human grasping strategies,the object's grasp type data set was established based on four typical grasp types.The grasp type and grasping area of the object were predicted through deep learning,and the grasping angle was obtained by image processing,so the grasping planning of the under-actuated dexterous hand was simplified.The recognition accuracy of deep learning algorithm is 98.70%in the test set and 82.70%for the unknown object,showing good generalization ability.When performing autonomous grasping,the inaccuracy of the deep learning method was compensated to a certain extent by the adaptiveness of the under-actuated hand.The UR3e robotic arm was equipped with an under-actuated dexterous hand to carry out grasping experiments on 24 objects.The average success rate is 90.80%in 120 grasps.The results show that the proposed method can adapt to objects of different shapes and sizes,and is practical for grasping.