Abstract: Aiming at the problems of high cost,low accuracy of pose estimation and poor robustness in extreme scenes of traditional robot grasping algorithm,a robot grasping method based on single view key point voting was proposed.In this method,voting was used to infer 2D key points based on a single RGB image.The 6D pose of object was calculated with EPNP algorithm by combining with the position of 3D key points,which was converted into the optimal robot grasping posture to achieve robot grasping.The experimental results showed that the proposed method could achieve better estimation results even in occluded,truncated and cluttered scenes,and the success rate of the method reached 94%,which could guide the robot to grasp accurately.

Key words: 6D pose, single view, key points, voting, robot grabbing

摘要： 针对传统机器人抓取算法成本高、位姿估计准确率低、在极端场景中鲁棒性差等问题,提出一种基于单视图关键点投票的机器人抓取方法。该方法基于单个RGB图像,采用投票推理2D关键点的方式,再结合3D关键点的位置,利用多组点对映射关系(E-Perspective-n-Point,EPnP),算法计算物体的6D位姿,并将其转换为最优的机器人抓取姿势,实现机器人抓取。实验表明,即使在遮挡、截断、杂乱场景中,也能体现较好的估计结果。所提方法抓取成功率达到了94%,能引导机器人实现准确抓取。

关键词: 6D位姿, 单视图, 关键点, 投票, 机器人抓取