一种用于易损物体的机器人抓取姿态预测方法

Robot Grasping Pose Prediction Method for Fragile Objects

复杂场景下机器人抓取“万物”是一个重要且具有挑战性的任务.针对使用刚性夹持器的机器人在直接使用六自由度抓取姿态预测方法抓取柔软和易碎物品过程中易被夹持器损坏问题,本文提出了一种用于易损物体的八自由度抓取姿态预测方法.通过直接处理场景点云数据,预测物体上抓取点的接近向量、面内旋转、夹持器宽度和物体种类,并根据夹持器内点云得到物体的八自由度抓取姿态,其姿态包含夹持器的旋转、平移、夹持器的宽度和作用力.然后,在公共数据集上验证算法的有效性,并搭建复杂场景进行机器人抓取实验.实验结果表明,该方法在保证抓取成功率的前提下,降低了被抓取物体的损坏率,扩展了基于视觉的机器人抓握物品种类.

It is an important and challenging task for robots to grasp"everything"in complex scenes.Aiming at the problem that a robot using a rigid gripper is easy to be damaged by the gripper when it directly uses the six-degree-of-freedom grasping pose prediction method to grasp soft and fragile objects.An eight-degree-of-freedom grasping pose prediction method for fragile objects is proposed in this paper.By directly processing the point cloud of scenic,the approach vector of grasping points on the object,in-plane rotation,gripper width and object type are predicted,and the eight-degree-of-freedom grasping pose of the object is obtained according to the gripper point cloud.The grasping pose includes rotation,translation,width and force of the gripper.And verify the effectiveness of the algorithm on the public dataset.Finally,a complex scene is constructed for robot grasping experiment.Experimental results show that this method can reduce the damage rate of captured objects and expand the types of objects grasped by visual-based robots on the premise of ensuring the success rate of grasping.