基于未知物体三维点云特征的机器人六自由度抓取

Six-degree-of-freedom robot grasping based on three-dimensional point cloud features of unknown objects

中文关键词  机器人抓取  点云特征  卷积神经网络  力平衡  非结构化环境

英文关键词  robotic grasping  point cloud feature  convolutional neural network  force balance  unstructured environment

中文摘要

      针对非结构化环境中任意位姿的未知物体, 提出了一种基于点云特征的机器人六自由度抓取位姿检测方 法, 以解决直接从点云中获取目标抓取位姿的难题. 首先, 根据点云的基本几何信息生成抓取候选, 并通过力平衡等 方法优化这些候选; 然后, 利用可直接处理点云的卷积神经网络ConvPoint评估样本, 得分最高的抓取将被执行, 其 中抓取位姿采样和评估网络都是以原始点云作为输入; 最后, 利用仿真和实际抓取实验进行测试. 结果表明, 该方法 在常用对象上实现了88.33%的抓取成功率, 并可以有效地拓展到抓取其他形状的未知物体.

英文摘要

      In this paper, a six-degree-of-freedom robot grasp detection method based on point cloud features is proposed to address the challenging problem of directly obtaining the grasp pose of unknown objects in unstructured environments from the point cloud. Firstly, several grasp candidates will be sampled with essential geometry information of the point cloud, and optimized through methods such as force balance. Secondly, each grasp candidate is evaluated through Conv- Point, a convolutional neural network which can directly process point cloud of object, then the grasp candidate with the highest score will be executed. Both grasp sampler and grasp evaluation network take 3D point clouds observed by a depth camera as input. Finally, the performance of the proposed method is measured with simulation experiments and practical test. The results show that our approach achieves 88.33% success rate on various commonly used objects, and generalizes well to other objects of unknown shape in unstructured environments.